CoChat
Sign Up
Cited by 108Gait Recognition and AnalysisBiometric Identification and SecurityHuman Pose and Action Recognition

Gait Recognition Based on Deep Learning: A Survey

Open access
Filipi Gonçalves dos Santos, Claudio, Oliveira, Diego de Souza, A. Passos, Leandro, Gonçalves Pires, Rafael, Felipe Silva Santos, Daniel, Pascotti Valem, Lucas, P. Moreira, Thierry, Cleison S. Santana, Marcos +3 more·2023·ACM Computing Surveys
DOI 10.1145/3490235
Open this paper in CoChat — free accountOpen PDF

Abstract

In general, biometry-based control systems may not rely on individual expected behavior or cooperation to operate appropriately. Instead, such systems should be aware of malicious procedures for unauthorized access attempts. Some works available in the literature suggest addressing the problem through gait recognition approaches. Such methods aim at identifying human beings through intrinsic perceptible features, despite dressed clothes or accessories. Although the issue denotes a relatively long-time challenge, most of the techniques developed to handle the problem present several drawbacks related to feature extraction and low classification rates, among other issues. However, deep learning-based approaches recently emerged as a robust set of tools to deal with virtually any image and computer-vision-related problem, providing paramount results for gait recognition as well. Therefore, this work provides a surveyed compilation of recent works regarding biometric detection through gait recognition with a focus on deep learning approaches, emphasizing their benefits and exposing their weaknesses. Besides, it also presents categorized and characterized descriptions of the datasets, approaches, and architectures employed to tackle associated constraints.

AI snapshot

Written from the paper’s full text.

TL;DR

This survey reviews deep learning approaches for gait recognition, covering CNN, RNN, autoencoder, CapsNet, DBN, and GAN architectures along with public datasets. It reports that 2D-CNNs dominate the literature and that most datasets and models remain constrained to controlled, single-person settings.

Key findings

  • The survey reports that 2D-CNNs are the most widely used deep neural network for gait recognition, at approximately 50% of solutions; 3D-CNN and GAN are next at about 10% each, while hybrid methods constitute 26%, with CNN-RNN combinations most adopted at about 15%.
  • For gait representation, silhouettes are used in more than 85% of solutions, skeletons in only 10%, and skeleton-plus-silhouette representations in approximately 5%.
  • Table 2 reports varied results, including CNN+Siamese networks at 98.8% accuracy on OU-ISIR, CNN+LSTM at 99.75% accuracy on whuGAIT and OU-ISIR, LSTM at 99.1% recognition rate on CASIA B and OU-ISIR, Capsule networks at 74.4% accuracy on OU-ISIR, and DiGGAN at 93.2% identification rate on OU-MVLP and CASIA B.
  • The paper catalogs datasets including CMU MoBo, TUM GAID, HID-UMD, CASIA, OU-ISIR, USF, SOTON, AVAMVG, KY4D, and WhuGait, and notes that lack of training data and controlled recording conditions hamper real-world deployment.

Methods

  • Literature survey of recent gait recognition works using deep learning; search keywords were “gait recognition using deep learning technique,” with the technique replaced by architecture names such as convolutional neural networks and LSTM.
  • Selection filters discarded papers with similar techniques or architectures and considered studies published at most five years earlier.
  • The survey provides theoretical background on biometric identification and deep learning architectures (CNN, CapsNet, RNN/LSTM/GRU, autoencoders, DBN, GAN), then summarizes surveyed studies in Table 2 and discusses state-of-the-art by literature- and representation-based categories.

Limitations

  • The authors state that the presented datasets do not include more than one person per video for training or validation, and that recording environments are fully controlled, with fixed backgrounds, treadmills, and limited object or color variation; consequently, models may show lack and misbehavior in real-world operation.
  • The authors note that public or private gait datasets are scarce because gait biometry requires multiple movement recordings per subject, videos are high-dimensional and storage-heavy, and public distribution requires participant consent.
  • Evident from the text: the survey excludes works older than five years and discards papers with similar techniques or architectures, so coverage is not exhaustive; no formal quantitative meta-analysis is reported.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3490235

Cites

152 of 153 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

17 of 49 cited sources checked · 9 supported · 0 flagged · accuracy not yet measured

  • Premisecitation
    § 1 INTRODUCTION · p. 2

    Existing reviews in this context mostly comprise CNN and LSTM, leaving behind relevant and novel architectures such as Autoencoders, CapsNet, Deep Belief Networks, and GANs.

    “Moreover, even though there are reviews in this context [2], most of the surveyed works comprise Convolutional Neural Networks (CNN) [49] and Long short-term memory (LSTM) [32, 55], leaving behind some relevant and novel architectures, such as Autoencoders [108], Capsule Networks (CapsNet) [82], Deep Belief Networks [31], and Generative Adversarial Networks (GANs) [23].”
    Quote in the paperSubstantial claim3 of 4 sources checked
    • cites [2] [2] Abdullah S Alharthi, Syed U Yunas, and Krikor B Ozanyan. 2019. Deep learning for monitoring of human gait: A review. IEEE Sensors Journal 19, 21 (2019), 9575-9591. - source
    • cites [49] [49] Yann LeCun, Léon Bottou, Yoshua Bengio, Patrick Haffner, et al. 1998. Gradient-based learning applied to document recognition. Proc. IEEE 86, 11 (1998), 2278-2324. - source
    • cites [32] [32] Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735-1780. - source
    • cites [55] [55] V. C. Lobo, L. A. Passos, and J. P. Papa. 2020. Evolving Long Short-Term Memory Networks. In International Conference on Computational Science (ICCS) . IEEE. - source
  • Premisereasoning
    § 1 INTRODUCTION · p. 2

    None of the prior surveys focus on presenting the main advances regarding deep learning-based approaches for gait recognition.

    “However, none of them focus on presenting the main advances regarding deep learning-based approaches for gait recognition.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [110] [110] Changsheng Wan, Li Wang, and Vir V Phoha. 2018. A survey on gait recognition. ACM Computing Surveys (CSUR) 51, 5 (2018), 1-35. - source

      Section 1.2 (Motivation, p. 89:3) says the earlier surveys it examines missed recent work and neglected neural-network classifiers. Section 6.3.4 describes this paper’s own treatment of deep-learning gait recognition as a brief review. These passages support a gap in the older surveys, but do not establish the categorical claim that none of the prior surveys focused on the main deep-learning advances. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [79] [79] Imad Rida, Noor Almaadeed, and Somaya Almaadeed. 2018. Robust gait recognition: a comprehensive survey. IET Biometrics 8, 1 (2018), 14-28. - source
    • cites [67] [67] Athira Nambiar, Alexandre Bernardino, and Jacinto C Nascimento. 2019. Gait-based person re-identification: A survey. ACM Computing Surveys (CSUR) 52, 2 (2019), 1-34. - source
  • Premisecitation
    § 3.3 Recurrent Neural Networks · p. 14

    Other models combining accelerometer and gyroscope data for gait recognition reached recognition rates above 91%.

    “Other models combined specific information, such as data obtained from accelerometer and gyroscope, for gait recognition through deep learning-based approaches, reaching recognition rates above 91% [141].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [141] [141] Qin Zou, Yanling Wang, Yi Zhao, Qian Wang, Chao Shen, and Qingquan Li. 2018. Deep Learning Based Gait Recognition Using Smartphones in the Wild. arXiv preprint arXiv:1811.00338 (2018). source
  • Premisecitation
    § 3.7 Discussion · p. 16

    DiGGAN obtained state-of-the-art results over the largest multiview gait dataset (OU-MVLP, >10,000 people) using cross-view approaches.

    “An example of the approach can be observed in the UK's Newcastle University cross-view gait recognition system, the so-called DiGGAN, which obtained state-of-the-art results over the largest multiview gait dataset in the world [99] (comprising more than 10 , 000 people) using cross-view approaches [35].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [99] [99] Noriko Takemura, Yasushi Makihara, Daigo Muramatsu, Tomio Echigo, and Yasushi Yagi. 2018. Multi-view large population gait dataset and its performance evaluation for cross-view gait recognition. IPSJ Transactions on Computer Vision and Applications 10, 1 (2018), 4. - source
    • cites [35] [35] BingZhang Hu, Yu Guan, Yan Gao, Yang Long, Nicholas Lane, and Thomas Ploetz. 2018. Robust Cross-View Gait Recognition with Evidence: A Discriminant Gait GAN (DiGGAN) Approach. arXiv preprint arXiv:1811.10493 (2018). - source
  • Findingnone given
    § 3.7 Discussion · p. 17

    Skeletons are considered less frequently than silhouettes, corresponding to only 10% of available solutions.

    “Although it is a promising approach, skeletons have been considered less frequently in relation to silhouettes, corresponding to only 10% of the available solutions.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The surveyed literature is representative of the state of the art in deep learning-based gait recognition.
  • May combine two claims. 28
  • Classified as probably not a substantial claim. 34

  • May combine two claims.

    Premisecitation
    § 2.3 Gait Recognition · p. 10

    Gait identification can also be performed by footprint, which considers the pressure and size of the area, but the obvious disadvantage is the cost of deploying the necessary equipment.

    “Gait identification can also be performed by distinct characteristics, e.g., the footprint [11], which considers the pressure and size of the area. The obvious disadvantage of such models is the cost of deploying the equipment necessary to acquire such features.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [11] [11] Omar Costilla-Reyes, Ruben Vera-Rodriguez, Patricia Scully, and Krikor B Ozanyan. 2018. Analysis of spatio-temporal representations for robust footstep recognition with deep residual neural networks. IEEE transactions on pattern analysis and machine intelligence 41, 2 (2018), 285-296. - source

      Section 2.1 (p. 3) describes biometric footstep recognition using spatial information from accumulated pressure images, and Section 3 (p. 4) describes the pressure-sensing floor mats used to acquire footsteps. These passages support identification from pressure and spatial footprint features, but the paper does not state that equipment deployment cost is a disadvantage. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation
    § 3.3 Recurrent Neural Networks · p. 14

    Wang and Yan used LSTM for cross-view gait recognition based on frame-by-frame GEIs, and experiments on CASIA B and OU-ISIR demonstrate robustness over baselines.

    “Concerning Recurrent Neural Networks [43], Wang and Yan [115] used the Long short-term memory [32], a popular RNN architecture capable of learning long-term dependencies, for crossview human gait recognition based on frame-by-frame GEIs. Experiments conducted over CASIA B [128] and OU-ISIR [59] Large Population datasets demonstrate the model's robustness over several baselines.”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [115] [115] Xiuhui Wang and Wei Qi Yan. 2020. Human gait recognition based on frame-by-frame gait energy images and convolutional long short-term memory. International journal of neural systems 30, 01 (2020), 1950027. - source
    • cites [32] [32] Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735-1780. - source
    • cites [128] [128] Shiqi Yu, Daoliang Tan, and Tieniu Tan. 2006. A Framework for Evaluating the Effect of View Angle, Clothing and Carrying Condition on Gait Recognition. In Pattern Recognition , Vol. 4. 441-444. https://doi.org/10.1109/ICPR.2006.67 - source

  • May combine two claims.

    Premisecitation
    § 3.1 Convolutional Neural Networks · p. 12

    Shiraga et al. used an architecture similar to LeNet to create a GEI-based recognizer, achieving 91.5% accuracy on the OU-ISIR Large Population dataset.

    “In the context of gait recognition, Shiraga et al. [86] used an architecture similar to LENet [49] to create a gleaming-based recognizer through Gait Energy Image (GEI). The model achieved an accuracy of 91 . 5% in the OU-ISIR [59] Large Population dataset.”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [86] [86] Kohei Shiraga, Yasushi Makihara, Daigo Muramatsu, Tomio Echigo, and Yasushi Yagi. 2016. Geinet: View-invariant gait recognition using a convolutional neural network. In 2016 international conference on biometrics (ICB) . IEEE, 1-8. - source
    • cites [49] [49] Yann LeCun, Léon Bottou, Yoshua Bengio, Patrick Haffner, et al. 1998. Gradient-based learning applied to document recognition. Proc. IEEE 86, 11 (1998), 2278-2324. - source
    • cites [59] [59] Y. Makihara, H. Mannami, A. Tsuji, M.A. Hossain, K. Sugiura, A. Mori, and Y. Yagi. 2012. The OU-ISIR Gait Database Comprising the Treadmill Dataset. IPSJ Trans. on Computer Vision and Applications 4 (Apr. 2012), 53-62. - source

  • May combine two claims.

    Premisecitation
    § 3.2 Capsule Networks · p. 14

    Zhao et al. introduced ASTCapsNet, trained on multi-sensor datasets, showing that multi-modality data is more conducive to gait recognition.

    “Finally, Zhao et al. [134] introduced an automated learning system called Associated SpatioTemporal Capsule Network (ASTCapsNet). The model was trained over multi-sensor datasets to show that multi-modality data is more conducive to gait recognition.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [134] [134] Aite Zhao, Junyu Dong, Jianbo Li, Lin Qi, and Huiyu Zhou. 2021. Associated Spatio-Temporal Capsule Network for Gait Recognition. arXiv preprint arXiv:2101.02458 (2021). - source

      The Abstract says the authors introduced ASTCapsNet, trained on multi-sensor datasets, and states that multimodality is conducive to gait research. But the Discussion says the model recognizes gait from different sources without considering relationships between modalities; the reported experiments do not establish that multimodal data outperforms single-modality data for gait recognition. The copy provides section labels but no page numbers. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation
    § 3.3 Recurrent Neural Networks · p. 14

    Tran et al. proposed an IMU-based gait recognition approach using LSTMs, achieving state-of-the-art performance for verification and identification on whuGAIT and OU-ISIR.

    “Recently, Tran et al. [103] proposed an Inertial Measurement Units (IMUs)-based gait recognition approach. The authors employed LSTMs to exploit the temporal information on video sequences, thus extracting hidden patterns inside such sequences. Experiments conducted over whuGAIT [141] and OU-ISIR [59] provided state-of-the-art performance for both verification and identification tasks.”
    Quote in the paperNot a substantial claim2 of 3 sources checked
    • cites [103] [103] Lam Tran, Thang Hoang, Thuc Nguyen, Hyunil Kim, and Deokjai Choi. 2021. Multi-Model Long Short-Term Memory Network for Gait Recognition Using Window-Based Data Segment. IEEE Access 9 (2021), 23826-23839. - source

      The Introduction and Conclusion describe an IMU-based LSTM approach evaluated on whuGAIT and OU-ISIR; they report state-of-the-art identification and verification performance for the hybrid LSTM–CNN model. The claim's reference to video sequences is unsupported: the LSTMs process IMU gait-data sequences, not video. The source also does not claim state-of-the-art performance for LSTM alone. (LLM, conf=0.92)

      Experimental — accuracy not yet measured
    • cites [141] [141] Qin Zou, Yanling Wang, Yi Zhao, Qian Wang, Chao Shen, and Qingquan Li. 2018. Deep Learning Based Gait Recognition Using Smartphones in the Wild. arXiv preprint arXiv:1811.00338 (2018). source

      The whuGAIT Datasets and User-Identification Performance sections report inertial-sensor gait data and LSTM experiments on whuGAIT and OU-ISIR; LSTM performs best among the compared methods for OU-ISIR identification. But this is Zou et al.’s paper, not Tran et al.’s, and its LSTM processes inertial time series rather than video sequences. The User-Authentication Performance section reports that CNN+LSTM, not LSTM alone, performs best on OU-ISIR, so the claim of state-of-the-art performance for both tasks is overstated. (LLM, conf=0.90)

      Experimental — accuracy not yet measured
    • cites [59] [59] Y. Makihara, H. Mannami, A. Tsuji, M.A. Hossain, K. Sugiura, A. Mori, and Y. Yagi. 2012. The OU-ISIR Gait Database Comprising the Treadmill Dataset. IPSJ Trans. on Computer Vision and Applications 4 (Apr. 2012), 53-62. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.3 Gait Recognition · p. 11

    Template-based methods aim to obtain the movement of the trunk or legs, focusing on dynamics of movement through space or spatiotemporal based methods.

    “Template-based methods aim to obtain the movement of the trunk or legs, i.e., they usually focus on the dynamics of movement through space or spatiotemporal based methods [124].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [124] [124] Sang Seok Yeo and Ga Young Park. 2020. Accuracy verification of spatio-temporal and kinematic parameters for gait using inertial measurement unit system. Sensors 20, 5 (2020), 1343. - source

      The Methods section (2.1.1 and 2.1.3; page unavailable) describes sensors on the pelvis and legs measuring spatiotemporal and kinematic gait parameters. It does not identify this approach as template-based or establish what template-based methods generally aim to do. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.3 Gait Recognition · p. 11

    Non-template-based methods consider the shape and its attributes as the more relevant characteristics, performing recognition with measurements that reflect its shape.

    “On the other hand, nontemplate based methods consider the shape and its attributes as the more relevant characteristics, i.e., the recognition of the individual is performed with measurements that reflect its shape [16].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [16] [16] Muqing Deng et al. 2018. Gait recognition under different clothing conditions via deterministic learning. in IEEE/CAA Journal of Automatica Sinica (2018). - source

      Section IV (p. 3) says the method measures silhouette widths and area, supporting the reference to shape-related measurements. But the Introduction (p. 1) distinguishes appearance-based from model-based methods, not template-based from nontemplate-based methods, and the experiments (p. 6) say recognition relies on gait dynamics rather than shape information. The source therefore does not support the claim that nontemplate-based methods treat shape as the more relevant characteristic for recognition. (LLM, conf=0.83)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.7 Discussion · p. 16

    Gait Energy Image reflects the sequence of a simple energy image cycle using weighted average, and sequences in a travel cycle are processed to align the binary silhouette.

    “Concerning the most used methods to represent gait image data, Gait Energy Image reflects the sequence of a simple energy image cycle using the weighted average method. Further, the sequences in a travel cycle are processed to align the binary silhouette [42].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [42] [42] Chunyuan Zi Ying Niu Huixin Tian Jing Luo, Jianliang Zhang and Chunbo Xiu. 2015. Gait Recognition Using GEI and AFDEI. International Journal of Optics (2015). - source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation
    § 4.10 WhuGAIT Datasets · p. 25

    The whuGAIT Datasets were released in 2018 by the Wuhan University and made available along with the source code and pre-trained models to replicate the paper results.

    “The whuGAIT Datasets [140] were released in 2018 by the Wuhan University and made available along with the source code and pre-trained models to replicate the paper results 9 .”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [140] [140] Qin Zou, Yanling Wang, Yi Zhao, Qian Wang, Chao Shen, and Qingquan Li. 2018. Deep Learning Based Gait Recognition Using Smartphones in the Wild. CoRR abs/1811.00338 (2018). arXiv:1811.00338 http://arxiv.org/abs/1811.00338 - source

      The whuGAIT Datasets section describes the datasets, and the Conclusion says the authors shared the datasets, code, and trained model. Our copy does not establish that Wuhan University released them specifically in 2018 or that the shared materials replicate the paper’s results. (LLM, conf=0.85)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingnone given
    § 3.7 Discussion · p. 17

    DAE, RNN, CapsNet, DBN, and graph convolutional networks correspond to 5%, 3%, 2%, 1%, and 1% of DNNs, respectively.

    “In addition, deep autoencoder (DAE), RNN, CapsNet, DBN, and graph convolutional networks are less considered among DNNs, corresponding to 5%, 3%, 2%, 1%, and 1%, respectively.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The surveyed literature is representative of the state of the art in deep learning-based gait recognition.
    • Rests on an assumption: The classification of techniques into categories (CNN, RNN, etc.) is exhaustive and mutually exclusive enough for the survey's analysis.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given
    § 3.7 Discussion · p. 17

    Silhouettes are the most widely adopted representation for gait recognition, corresponding to more than 85% of solutions.

    “Regarding state-of-the-art solutions based on representation, the silhouettes are the most widely adopted for gait recognition, corresponding to more than 85% of the solutions.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The surveyed literature is representative of the state of the art in deep learning-based gait recognition.

  • May combine two claims.

    Findingnone given
    § 3.7 Discussion · p. 17

    2D-CNNs are the most widely used type of DNN for gait recognition, with approximately 50% of solutions based on 2D-CNN architectures for classification.

    “Regarding the literature-based, 2D-CNNs are the most widely used type of deep neural network (DNN) for gait recognition using deep learning, with approximately 50% of solutions just based on 2D-CNN architectures for classification.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The surveyed literature is representative of the state of the art in deep learning-based gait recognition.
    • Rests on an assumption: The classification of techniques into categories (CNN, RNN, etc.) is exhaustive and mutually exclusive enough for the survey's analysis.

  • May combine two claims.

    Findingown data
    § 3.7 Discussion · p. 17

    Hybrid methods constitute 26% of solutions, with CNN-RNN combinations most widely adopted at about 15%, DAE with GAN and RNN at 10%, and RNN-CapsNet at 2%.

    “On the other hand, the hybrid methods that constitute 26% of the solutions, in which the CNN-RNN combinations are the most widely adopted approach with about 15% of presence, while the combination of DAE with GAN and RNN corresponds to 10% of the methods, followed by the RNN-CapsNet methods that constitute 2% of the solutions.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The surveyed literature is representative of the state of the art in deep learning-based gait recognition.
    • Rests on an assumption: The classification of techniques into categories (CNN, RNN, etc.) is exhaustive and mutually exclusive enough for the survey's analysis.

  • May combine two claims.

    Findingown data
    § 3.7 Discussion · p. 17

    3D-CNN and GAN are the next popular categories, each corresponding to 10% of published content.

    “Solutions using 3D-CNN and GAN are the next popular categories, each corresponding to 10% of the published content.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The surveyed literature is representative of the state of the art in deep learning-based gait recognition.
    • Rests on an assumption: The classification of techniques into categories (CNN, RNN, etc.) is exhaustive and mutually exclusive enough for the survey's analysis.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation
    § 4.6 University of South Florida Dataset · p. 23

    The USF dataset comprises 1,870 sequences from 122 subjects using two different shoe types.

    “The University of South Florida (USF) dataset [84] comprises 1 , 870 sequences from 122 subjects using two different shoe types.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [84] [84] Sudeep Sarkar, P Jonathon Phillips, Zongyi Liu, Isidro Robledo Vega, Patrick Grother, and Kevin W Bowyer. 2005. The humanid gait challenge problem: Data sets, performance, and analysis. IEEE transactions on pattern analysis and machine intelligence 27, 2 (2005), 162-177. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given
    § 4.5 OU-ISIR Biometric Database · p. 22

    The OU-ISIR project has been creating, since 2007, the largest dataset for gait recognition in the world.

    “The Institute of Scientific and Industrial Research (ISIR) of Osaka University (OU) has been creating, since 2007, the largest dataset for gait recognition in the world.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingnone given
    § 5 CONCLUSIONS AND FUTURE DIRECTIONS · p. 28

    The works surveyed discuss the main concerns and efforts made by the scientific community towards the development of techniques to guarantee access to monitored resources via gait identification and authorization.

    “The works surveyed in this paper discuss the main concerns and efforts made by the scientific community towards the development of techniques to guarantee access to monitored resources via gait identification and authorization.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation
    § 4.4 CASIA · p. 20

    The CASIA Gait Database is a collection of four datasets designed for gait recognition purposes.

    “The Institute of Automation, Chinese Academy of Sciences (CASIA) provides the CASIA Gait Database [137], a collection of four datasets designed for gait recognition purposes”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [137] [137] Shuai Zheng, Junge Zhang, Kaiqi Huang, Ran He, and Tieniu Tan. 2011. Robust view transformation model for gait recognition. In 2011 18th IEEE International Conference on Image Processing . IEEE, 2073-2076. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation
    § 4.1 CMUMoBoDataset · p. 19

    The CMU MoBo dataset is available for download without reservations or need to sign forms of agreement.

    “One of the main advantages of the dataset is that it is available for download without reservations or need to sign forms of agreement, requiring just connecting to the Calgary University File Transfer Protocol (FTP) server 3 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] [24] Ralph Gross and Jianbo Shi. 2001. The CMU Motion of Body (MoBo) Database . Technical Report CMU-RI-TR-01-18. Carnegie Mellon University. -

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation
    § 4.2 TUMGAID Dataset · p. 19

    The TUM GAID database is compounded by RGB images, audio, and depth, recorded from 305 people in three different variations at first.

    “The TUM Gait from Audio, Image, and Depth (GAID) [33] database, as the name suggests, is compounded by RGB images, audio, and depth, recorded from 305 people in three different variations at first.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [33] [33] Martin Hofmann, Jürgen Geiger, Sebastian Bachmann, Björn Schuller, and Gerhard Rigoll. 2014. The tum gait from audio, image and depth (gaid) database: Multimodal recognition of subjects and traits. Journal of Visual Communication and Image Representation 25, 1 (2014), 195-206. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation
    § 4.2 TUMGAID Dataset · p. 19

    The TUM GAID dataset has a well-defined evaluation protocol.

    “A strong advantage of this dataset is a well-defined evaluation protocol.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [33] [33] Martin Hofmann, Jürgen Geiger, Sebastian Bachmann, Björn Schuller, and Gerhard Rigoll. 2014. The tum gait from audio, image and depth (gaid) database: Multimodal recognition of subjects and traits. Journal of Visual Communication and Image Representation 25, 1 (2014), 195-206. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given
    § 5 CONCLUSIONS AND FUTURE DIRECTIONS · p. 28

    Regarding the datasets, one can clearly notice the efforts towards a detection independent from gender, angle in the scene, clothes and shoes, carry of bags and other items, among other common issues.

    “Regarding such datasets, one can clearly notice the efforts towards an detection independent from gender, angle in the scene, clothes and shoes, carry of bags and other itens, among other common issues.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation
    § 5 CONCLUSIONS AND FUTURE DIRECTIONS · p. 29

    There is an increasing demand for hybrid approaches composed of both gait and complementary biometric characteristics, such as face and ear.

    “as well as hybrid approaches composed of both gait and complementary biometric characteristics, such as face and ear [64].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [64] [64] Haider Mehraj and Ajaz Hussain Mir. 2020. Feature vector extraction and optimisation for multimodal biometrics employing face, ear and gait utilising artificial neural networks. International Journal of Cloud Computing 9, 2-3 (2020), 131-149. - source

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given
    § 5 CONCLUSIONS AND FUTURE DIRECTIONS · p. 29

    The presented datasets do not introduce more than one person per video, either for training or validation purposes.

    “The presented datasets do not introduce more than one person per video, either for training or validation purposes.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § 1 INTRODUCTION · p. 3

    The keywords employed in the search were 'gait recognition using deep learning technique', with deep learning technique replaced by the architectures themselves.

    “Regarding the selection of the reviewed papers, the keywords employed in the search were 'gait recognition using deep learning technique ', such that deep learning technique was replaced by the architectures themselves, e.g., convolutional neural networks, LSTM, and so on.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 1 INTRODUCTION · p. 2

    Gait recognition methods are commonly cheaper than other approaches because they do not require sophisticated equipment for data extraction.

    “Besides, since it does not require sophisticated equipment for data extraction, these methods are commonly cheaper than other approaches, majorly due to the popularization of surveillance systems and the advent of cellphones equipped with accelerometers, which transformed the burden of extracting data signals into a straightforward task.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Gait recognition models do not require high-resolution images and specialized equipment for proper identification, unlike iris and fingerprints.

  • Probably not a substantial claim: may not pass S4 (based).

    Study design and conductnone given
    § 1 INTRODUCTION · p. 3

    The studies considered in this research were published five years ago at most.

    “Finally, the studies considered in this research were published five years ago at most.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given
    § 1 INTRODUCTION · p. 3

    Another filter considered was the work's innovation, i.e., papers with similar techniques and architectures were discarded.

    “Another filter considered was the work's innovation, i.e., papers with similar techniques and architectures were discarded.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § 3.5 Deep Belief Networks · p. 15

    Fernandes et al. employed DBNs to support gait assessment in diagnosis of Parkinson's and movement disorder diseases, confirming that DBN-based approaches are suitable for the task.

    “Regarding gait recognition, Fernandes et al. [21] employed DBNs to support gait assessment in the diagnosis of Parkinson's and movement disorder diseases. The authors employed wearable sensors to extract features from the subjects and performed a comparative classification analysis of parkinsonian gait, confirming that DBN-based approaches are suitable for the task.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] [21] Carlos Fernandes, Luís Fonseca, Flora Ferreira, Miguel Gago, Lus Costa, Nuno Sousa, Carlos Ferreira, Joao Gama, Wolfram Erlhagen, and Estela Bicho. 2018. Artificial neural networks classification of patients with parkinsonism based on gait. In 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) . IEEE, 2024-2030. - source

  • May combine two claims.

    Interpretationreasoning
    § 3.7 Discussion · p. 17

    Most gait recognition methods using deep learning have not yet explored confusing context approaches, and therefore are not explicitly able to separate the underlying structure of gait data into significant disjoint variables.

    “However, most gait recognition methods available using deep learning have not yet explored such approaches. Therefore, they are not explicitly able to separate the underlying structure of gait data in the form of significant disjoint variables.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given
    § 2.3 Gait Recognition · p. 11

    Techniques proposed for gait recognition can be divided into two main groups: template-based and non-template-based methods.

    “Moreover, techniques proposed for gait recognition can be divided into two main groups, i.e., template- and non-template-based methods.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.1 Convolutional Neural Networks · p. 12

    Wu et al. proposed the first work using CNNs for cross-view gait-based human identification, obtaining gains above 10% over previous results.

    “Regarding cross-view gait-based human identification, Wu et al. [116] claim to propose the first work using CNNs in the context of grimace recognition. The authors employed several network architectures and demonstrated the power of CNNs in the context of biometric-based identification, obtaining gains above 10% to previous results.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [116] [116] Zifeng Wu, Yongzhen Huang, Liang Wang, Xiaogang Wang, and Tieniu Tan. 2017. A comprehensive study on cross-view gait based human identification with deep cnns. IEEE transactions on pattern analysis and machine intelligence 39, 2 (2017), 209-226. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1 Convolutional Neural Networks · p. 12

    Wang et al. employed nonstandard periodic GEI approaches for gait recognition and data augmentation.

    “A similar work proposed by Wang et al. [113] employ nonstandard periodic GEI approaches for gait recognition and data augmentation.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [113] [113] Kejun Wang, Liangliang Liu, Yilong Lee, Xinnan Ding, and Junyu Lin. 2019. Nonstandard Periodic Gait Energy Image for Gait Recognition and Data Augmentation. In Chinese Conference on Pattern Recognition and Computer Vision (PRCV) . Springer, 197-208. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.7 Discussion · p. 16

    GEIs are a little sensitive to silhouette noise in individual pictures.

    “GEIs are a little sensitive to silhouette noise in individual pictures.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] [26] Ju Han and Bir Bhanu. 2016. Individual Recognition UsingGait Energy Image. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2016). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.7 Discussion · p. 16

    GEI represents human motion in a single image while preserving temporal information.

    “It represents human motion in a single image while preserving temporal information.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] [26] Ju Han and Bir Bhanu. 2016. Individual Recognition UsingGait Energy Image. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2016). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.7 Discussion · p. 16

    Period Energy Image is a multichannel gait template proposed to tackle view angle variation and GEI loss of temporal information.

    “The Period Energy Image, for instance, comprises a multichannel gait template proposed to tackle view angles' variation faced by cross-view and GEI loss of temporal information [28].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] [28] Yiwei He, Junping Zhang, Hongming Shan, and Liang Wang. 2018. Multi-task GANs for view-specific feature learning in gait recognition. IEEE Transactions on Information Forensics and Security 14, 1 (2018), 102-113. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationcitation
    § 3.7 Discussion · p. 16

    Most deep learning-based gait recognition approaches use image/video domain, but other data sources like accelerometers, gyroscope, sensor-based, and handcrafted features also provided impressive results in a considerable number of works.

    “Although most deep learning-based approaches for gait recognition comprise image/video domain, other data sources, such as accelerometers, gyroscope, sensor-based, and handcrafted features in general, also provided impressive results in a considerable number of works.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 3.7 Discussion · p. 16

    Cross-view-based gait recognition is a popular approach for different visual angles, requires multiple fully controlled cameras and cooperative environments, and visually normalizes gait characteristics before combination.

    “Similarly, cross-view-based gait recognition is a popular approach used to deal with different visual angles. The input type requires multiple fully controlled cameras and cooperative environments, thus being restrained to real scenarios. Moreover, it visually normalizes gait characteristics before performing any combination, which allows the model to learn the relationships among visual movements”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [20] [20] G. Ariyanto e MS Nixon. 2011. Model-based 3D gait biometrics. in IJCB (2011). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [109] [109] Jian Zhang Yi Ma W. Kusakunniran, Qiang Wu and Hongdong Li. 2013. A New View-Invariant Feature for Cross-View Gait Recognition. IEEE TIFS (2013), 1642-1653. - source
    • cites [123] [123] Yasuhiro Mukaigawa Tomio Echigo Yasushi Makihara, Ryusuke Sagawa and Yasushi Yagi. 2006. Gait Recognition Using a View Transformation Model in the Frequency Domain. In ECCV: v.3953 (2006), 151-163. - source

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given
    § 2.3 Gait Recognition · p. 11

    Deep learning-based approaches induced a paradigm shift in the field of gait recognition, obtaining paramount results in a variety of applications.

    “Despite the success obtained by the methods mentioned above, deep learning-based approaches induced a paradigm shift in the field of gait recognition, obtaining paramount results in a variety of applications.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.7 Discussion · p. 16

    Optical-flow provides relevant information regarding objects, spatial arrangements, and changes in such arrangements.

    “Another approach observed in some works comprises Optical-flow, which provides relevant information regarding the objects observed in the scene, their spatial arrangements, and the changes in sich arrangements [57].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [57] [57] Zhengping Luo, Tianqi Yang, and Yanjun Liu. 2016. Gait optical flow image decomposition for human recognition. In 2016 IEEE Information Technology, Networking, Electronic and Automation Control Conference . IEEE, 581-586. - source

  • May combine two claims.

    Interpretationreasoning
    § 3.7 Discussion · p. 16

    Most of these works address unsupervised deep learning approaches like Autoencoders and DBNs, which are usually more expressive over these data types, while CNNs are paramount for raw image/video files.

    “Most of these works address unsupervised deep learning approaches in the process, such as Autoencoders and DBNs, which usually are more expressive over these data types, while CNNs are paramount when dealing with raw image/video files.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 3.7 Discussion · p. 16

    GEI maintains static and dynamic characteristics of human walking and significantly reduces computational cost.

    “Therefore, GEI maintains the static and dynamic characteristics of human walking and significantly reduces image processing's computational cost.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.7 Discussion · p. 16

    GEI focuses on specific representations of human walk, which does not soften the context of vector images.

    “It focuses on specific representations of the human walk, which does not soften the context of vector images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] [26] Ju Han and Bir Bhanu. 2016. Individual Recognition UsingGait Energy Image. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2016). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.3 Recurrent Neural Networks · p. 14

    Potluri et al. employed LSTMs to detect gait abnormalities through a wearable sensor system, with experiments on ten healthy individuals simulating specific gait abnormalities.

    “A similar work proposed by Potluri et al. [76] employed LSTMs to detect gait abnormalities through a wearable sensor system. The experiments were conducted over a set of data extracted from ten healthy individuals, such that seven behave normally, while the three remaining simulate specific gait abnormalities, i.e., sensory ataxic, Parkinson's gait, and hemiplegic gait.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [76] [76] Sasanka Potluri, Srinivas Ravuri, Christian Diedrich, and Lutz Schega. 2019. Deep Learning based Gait Abnormality Detection using Wearable Sensor System. In 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) . IEEE, 3613-3619. - source

  • May combine two claims.

    Interpretationcitation
    § 3.7 Discussion · p. 16

    Capsule Networks are capable of extracting partial gait representations hierarchically, providing better results for cases with multiple orientations or overlapped objects.

    “Capsule Networks, for instance, are capable of extracting partial gait representations in a hierarchical fashion, providing better results for cases where individuals or objects are presented in the scene with multiple orientations or overlapped.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation
    § 3.7 Discussion · p. 16

    Other architectures also provide valuable contributions, performing better for several specific tasks.

    “Nevertheless, the other architectures also provide a valuable contribution to the field, performing better for several specific tasks.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given
    § 4 DATASETS · p. 19

    Gait biometry demands a reasonable amount of movement recordings for a subject, implying recording and generating multiple videos for each individual.

    “Gait biometry demands a reasonable amount of movement recordings for a subject, implying recording and generating multiple videos for each individual.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning
    § 3.7 Discussion · p. 16

    Recurrent Neural Networks are paramount for sequential data like videos, making them essential for gait recognition.

    “Similarly, Recurrent Neural Networks are paramount to deal with sequential data, such as videos, thus presenting themselves as an essential tool for this kind of gait recognition approach.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 3.1 Convolutional Neural Networks · p. 12

    Li et al. proposed DeepGait, which combines deep convolutional features and Joint Bayesian for video sensor-based gait representation, outperforming hand-crafted features and obtaining state-of-the-art results on OU-ISIR large population dataset.

    “Further, Li et al. [52] proposed the DeepGait, a model that combines deep convolutional features and Joint Bayesian for video sensor-based gait representation. The model outperformed hand-crafted features, such as GEI, Frequency-Domain Feature, and Gait Flow Image, obtaining state-of-theart results over OU-ISR large population dataset [59].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [52] [52] C. Li, X. Min, S. Sun, W. Lin, and Z. Tang. 2017. Deepgait: A learning deep convolutional representation for viewinvariant gait recognition using joint bayesian. Applied Sciences 7, 3 (2017), 210. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [59] [59] Y. Makihara, H. Mannami, A. Tsuji, M.A. Hossain, K. Sugiura, A. Mori, and Y. Yagi. 2012. The OU-ISIR Gait Database Comprising the Treadmill Dataset. IPSJ Trans. on Computer Vision and Applications 4 (Apr. 2012), 53-62. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 4 DATASETS · p. 19

    The eventual creation of a dataset without formal consent from each individual may bring about lawsuits.

    “The eventual creation of a dataset without formal consent from each individual may bring about lawsuits.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given
    § 5 CONCLUSIONS AND FUTURE DIRECTIONS · p. 28

    Gait systems do not require the individuals' willingness to be identified.

    “e.g., gait systems do not require the individuals' willingness to be identified.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisereasoning
    § 4 DATASETS · p. 19

    The extraction and public distribution of biometric data require permission from each participant.

    “The extraction and public distribution of biometric data require permission from each participant.”
    Quote in the paperNot a substantial claim
52 of 52

  • The paper relies on this without support.

    Gait recognition is useful in surveillance systems or hazy environments where fingerprint and face are hard or impossible to distinguish.

    “Such an approach presented itself extremely useful in the contexts of surveillance systems or hazy environments monitoring, for instance, where unique elements usually employed for biometric identification such as the fingerprint and the face are hard or impossible to distinguish.”

  • The paper relies on this without support.

    Gait recognition models do not require high-resolution images and specialized equipment for proper identification, unlike iris and fingerprints.

    “Further, gait recognition models do not require high-resolution images and specialized equipment for proper identification, like iris and fingerprints, for instance.”

  • The paper relies on this without support.

    Gait recognition methods are independent of individual cooperation, while other methods require the analyzed person to collaborate with the identification system.

    “Furthermore, gait recognition methods are independent of individual cooperation, while other methods require the analyzed person collaborates with the identification system.”

  • The paper relies on this without support.

    Gait information is easily collected from distance, which is an enormous advantage, especially when identification is not assisted by the analyzed person.

    “Moreover, gait information is easily collected from distance, which stands for an enormous advantage regarding other techniques, especially when the identification is not assisted by the analyzed person, e.g., criminal investigations.”

  • The paper relies on this without support.

    Standard gait recognition methods often suffer from constraints and challenges such as viewing angle, large intra-class variations, occlusions, shadows, and locating body segments.

    “Standard gait recognition methods, i.e., which comprise data pre-processing and features extracted in a handcrafted fashion for further recognition, often suffer from several constraints and challenges imposed by the complexity of the task, such as viewing angle and large intra-class variations, occlusions, shadows, and locating the body segments, among others.”

  • The paper relies on this without support.

    Deep learning emerged as a revolutionary tool for image and sound processing, computer vision, and speech, overwhelmingly outperforming virtually any baseline established until then.

    “A new trend on machine learning, known as deep learning, emerged in the last years as a revolutionary tool to handle topics in image and sound processing, computer vision, and speech, overwhelmingly outperforming virtually any baseline established until then.”

  • The paper relies on this without support.

    Deep learning exempts the necessity of manually extracting representative features from experts and provides paramount results regarding gait recognition, surpassing existing challenges.

    “The new paradigm exempts the necessity of manually extract representative features from experts and also provides paramount results regarding gait recognition, surpassing existing challenges and opening room for further research.”

  • The paper relies on this without support.

    Standard machine learning and gait recognition strategies are usually constrained to hand-crafted features and limited capacity for learning intrinsic patterns in data.

    “Even though standard machine learning and gait recognition strategies provided relatively satisfactory results in the past years, such methods are usually constrained to hand-crafted features and limited capacity for learning intrinsic patterns in data.”

  • The paper relies on this without support.

    Deep learning-based approaches are an elegant solution for tackling image/video and sequential problems and a powerful tool for gait recognition.

    “In this context, deep learning-based approaches emerged as an elegant solution for tackling image/video and sequential problems, among others, also revealing themselves as a powerful tool for gait recognition.”

  • The paper relies on this without support.

    Biometric recognition methods depend on passive provision of personal biometric information and some rely on specialized equipment.

    “Although the methods' reliability and safety are confirmed by their success in banks and public governance systems, two main hindrances must be stressed: (i) they depend on the passive provision of personal biometric information, i.e., the person must provide or register the required information for the recognition; and (ii) some of these systems rely on specialized equipment.”

  • The paper relies on this without support.

    CNNs obtained outstanding results in various applications and won most benchmark challenges in recent years.

    “Such behavior is expected since CNNs obtained outstanding results in various applications and won most of the benchmark challenges in the last years.”

  • The paper relies on this without support.

    The surveyed literature is representative of the state of the art in deep learning-based gait recognition.

    “Concerning gait recognition's actual state-of-the-art scenario, one can consider two main approaches, i.e., literature- and representation-based.”

  • The paper relies on this without support.

    The classification of techniques into categories (CNN, RNN, etc.) is exhaustive and mutually exclusive enough for the survey's analysis.

    “Table 2 provides a summary of the surveyed studies presented in this section.”
13 of 13

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Table 3 repeats section 4.6 instead of identifying section 4.7

    Minor

    The dataset columns in Table 3 run from 4.1 to 4.10 but label two columns 4.6 and none 4.7, although section 4.7 is the Southampton dataset.

  • Table 3 omits silhouettes for CMU MoBo

    Minor

    Table 3 leaves the silhouette entry blank for dataset 4.1, CMU MoBo, while its description explicitly says silhouette masks are provided.

  • A 42-subject database is said to contain a 54-subject subset

    Minor

    The KY4D description says the database has 42 subjects and is divided into three subsets, but describes one of those subsets as having 54 subjects.

  • Figure 7 assigns column 10 to two different conditions

    Minor

    The caption assigns column 10 to both time + backpack and time + coating shoes, despite presenting them as distinct variations.

  • Table 3 omits overground walking for OU-ISIR

    Minor

    Table 3 leaves the overground-walking entry blank for dataset 4.5, OU-ISIR, although its Speed Transition Dataset A includes walking on the ground.

5 of 5

Cites

Flagged first, then by citations

Research works printed but not linked: 12

  1. Ala Abdulhakim Alariki, Muqadas Faiz, Sanaullah Balagh, and Christine Murray. 2018. A review of finger-vein biometric recognition. In Proceedings of the World Congress on Engineering and Computer Science.
  2. Ruud M. Bolle, Jonathan H. Connell, Sharath Pankanti, Nalini K. Ratha, and Andrew W. Senior. 2013. Guide to Biometrics. Springer Science & Business Media.
  3. 10.5555/2969033.2969125 · DOI
  4. Ju Han and Bir Bhanu. 2016. Individual recognition usinggait energy image. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2016).
  5. 10.5555/2999134.2999257 · DOI
  6. 10.5555/1795765 · DOI
  7. 10.5555/1965992.1966027 · DOI
  8. M. Nixon, J. Carter, J. Shutler, and M. Grant. 2001. Experimental plan for automatic gait recognition. University of Southampton, Southampton, UK (2001).
  9. 10.5555/1949767.1949795 · DOI
  10. M. Roder, L. A. Passos, L. C. F. Ribeiro, B. C. Benato, A. L. Falcão, and J. P. Papa. In Press. Intestinal parasites classification using deep belief networks. In Proceedings of the 19th International Conference on Artificial Intelligence and Soft Computing (ICAISC). IEEE, 242–251.
  11. 10.5555/3294996.3295142 · DOI
  12. R. Abu Zneit, Ziad AlQadi, and M. Abu Zalata. 2017. A methodology to create a fingerprint for RGB color image. Int. J. Comput. Sci. Mob. Comput. 16, 1 (2017), 205–212.

Cited by

Sorted by citations

Related

Authors
Filipi Gonçalves dos Santos, C., Oliveira, D. D. S., A. Passos, L., Gonçalves Pires, R., Felipe Silva Santos, D., Pascotti Valem, L., P. Moreira, T., Cleison S. Santana, M., Roder, M., Paulo Papa, J., & Colombo, D.
Affiliations
  • Federal Institute of São Carlos - UFSCar, Brazil and Eldorado Research Institute, São Paulo, Brazil
  • Universidade Estadual Paulista (Unesp) (BR)
  • São Paulo State University - UNESP, Bauru, São Paulo, Brazil
  • Petrobras (Brazil) (BR)
ORCID
Venue
ACM Computing Surveys
Publisher
Association for Computing Machinery (ACM)
Published
2022-01-18
Volume
55
Issue
2
Pages
1-34
ISSN
  • 0360-0300 (ISSN-L)
  • 1557-7341
DOI
10.1145/3490235
OpenAlex
W4205268692
Cited by
108
References
137
Citations by year
2026: 11 · 2025: 32 · 2024: 26 · 2023: 28 · 2022: 11
FWCI
6.55
Type
journal article
Language
en
Open access
Licence
www.acm.org/publications/policies/copyright_policy#Background (vor)
Locations
  • ACM Computing Surveys (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University) (cc-by)
Funders
  • São Paulo Research Foundation (#2013/07375-0, #2014/12236-1, #2019/ 07665-4, and #2020/12101-0)
  • Brazilian National Council for Research and Development (#307066/2017-7 and #427968/2018-6)
  • Eldorado Research Institute
  • Petroleo Brasileiro S.A. (#2017/ 00285-6)
Topics
  • Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
  • Physical Sciences › Computer Science › Signal Processing › Biometric Identification and Security
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Human Pose and Action Recognition
Keywords
gait recognition, deep learning, biometric identification, unauthorized access detection, feature extraction, gait dataset, gait classification, wearable accessories

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-17.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-17.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review