Risky Action Recognition in Lane Change Video Clips using Deep Spatiotemporal Networks with Segmentation Mask Transfer
Open accessThe DOI matches its Crossref record; 36 of the 40 references it cites were found in a registry or an index.
Abstract
Advanced driver assistance and automated driving systems rely on risk estimation modules to predict and avoid dangerous situations. Current methods use expensive sensor setups and complex processing pipelines, limiting their availability and robustness. To address these issues, we introduce a novel deep learning based driving risk assessment framework for classifying dangerous lane change behavior in short video clips captured by a monocular camera. First, semantic segmentation masks were generated from individual video frames with a pre-trained Mask R-CNN model. Then, frames overlayed with these masks were fed into a time distributed CNN-LSTM network with a final softmax classification layer. This network was trained on a semi-naturalistic lane change dataset with annotated risk labels. A comprehensive comparison of state-of-the-art pre-trained feature extractors was carried out to find the best network layout and training strategy. The best result, with a 0.937 AUC score, was obtained with the proposed framework. Our code and trained models are available open-source.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 3×
“The third best result was obtained with an FT+LSTM architecture which used ResNet50 [39] as its backbone model.”
MethodVery Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 3×
“815 FbF FT VGG19 [34] ImageNET [31] 144.”
BackgroundImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 1×
“Developments in machine learning, particularly through deep convolutional neural networks (CNNs), significantly increased object detection capabilities [3] and made reliable object tracking achievable [4].”
BackgroundFaster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
ImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 4×
“815 FbF FT VGG19 [34] ImageNET [31] 144.”
BackgroundMethodInfluentialMicrosoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460
He, K., Gkioxari, G., Dollar, P., & Girshick, R. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.322 · Cited by 29,137
Pan, S. J., & Yang, Q. (2010) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2009.191 · Cited by 23,387 · cited 1×
“Inductive transfer learning refers to the case where the source and target tasks are different, whereas in transductive transfer learning, the domain changes while the task remains the same [25].”
BackgroundXception: Deep Learning with Depthwise Separable Convolutions
Chollet, F. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.195 · Cited by 20,380 · cited 2×
“761 FT+LSTM Xception [38] ImageNET 23 50 0.”
MethodMobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., & Adam, H. (2017) · arXiv · doi:10.48550/arxiv.1704.04861 · Cited by 19,015 · cited 1×
“VGG19 [34], MobileNet [35], InceptionResNet [36], NasNET [37], Xception”
Background
Research works printed but not linked: 2
- grayson, 2003, Risk hazard perception and perceived control
- held, 2016, Learning to track at 100 fps with deep regression networks, European Conference on Computer Vision, 749
Cited by
A Survey of Autonomous Driving: Common Practices and Emerging Technologies
Yurtsever, E., Lambert, J., Carballo, A., & Takeda, K. (2020) · IEEE Access · doi:10.1109/access.2020.2983149 · Cited by 1,889
“We employed an open-source1 deep spatiotemporal video-based risk detection framework [197] to assess the image sequences shown in this figure.”
MethodSelf-citationA Review of Vision-Based Traffic Semantic Understanding in ITSs
Chen, J., Wang, Q., Cheng, H. H., Peng, W., & Xu, W. (2022) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2022.3182410 · Cited by 173
“For example, Yurtsever et al. [187] sent video frames marked with a vehicle semantic segmentation mask into a CNN-LSTM network.”
BackgroundScene-Graph Augmented Data-Driven Risk Assessment of Autonomous Vehicle Decisions
Yu, S.-Y., Malawade, A. V., Muthirayan, D. R. P., Khargonekar, P. P., & Faruque, M. A. A. (2021) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2021.3074854 · Cited by 147
Song, D., Zhao, J. W., Zhu, B., Han, J., & Jia, S. (2024) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2024.3409874 · Cited by 88
“To display only a forward driving video through an ordinary display may affect the reliability of SDRLs [12], [26], [27], [28].”
BackgroundMethodTo Explain or Not to Explain: A Study on the Necessity of Explanations for Autonomous VehiclesPreprint
Shen, Y., Jiang, S., Chen, Y., & Driggs-Campbell, K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.11684 · Cited by 61
“asked ten participants to watch driving video clips and give a score (subjectively) for the risk of maneuver seen in the videos [33].”
Backgroundroadscene2vec: A tool for extracting and embedding road scene-graphs
Malawade, A. V., Yu, S.-Y., Hsu, B., Kaeley, H., Karra, A., & Faruque, M. A. A. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.108245 · Cited by 66
Review of Graph-Based Hazardous Event Detection Methods for Autonomous Driving Systems
Xiao, D., Dianati, M., Geiger, W. G., & Woodman, R. (2023) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2023.3240104 · Cited by 60
“In response, deep learning models have been proposed that utilise both convolutional neural networks (CNNs) and recurrent neural networks (RNNs) [9], [10], [11] to detect hazardous events from dashboard video footage.”
BackgroundLearning From Interaction-Enhanced Scene Graph for Pedestrian Collision Risk Assessment
Liu, X., Zhou, Y., & Gou, C. (2023) · IEEE Transactions on Intelligent Vehicles · doi:10.1109/tiv.2023.3309274 · Cited by 49
“Several works have focused on the collision risk assessment methods for traffic scenarios [24], [29], [31], [33], [47], [48], [49].”
BackgroundMethodInfluentialIntegrating Deep Reinforcement Learning with Model-based Path Planners for Automated Driving
Yurtsever, E., Capito, L., Redmill, K., & Ozgune, U. (2020) · IEEE Intelligent Vehicles Symposium · doi:10.1109/iv47402.2020.9304735 · Cited by 47
“The pipe usually continues with scene understanding [12], assessment [13], planning [14] and finally ends with motor control.”
BackgroundSelf-citationPredicting perceived risk of traffic scenes using computer vision
de Winter, J., Hoogmoed, J., Stapel, J., Dodou, D., & Bazilinskyy, P. (2023) · Transportation Research Part F Traffic Psychology and Behaviour · doi:10.1016/j.trf.2023.01.014 · Cited by 36
“Other research shows that various other factors contribute to perceived risk, such as road maintenance, infrastructural issues, brake lights, but also overtaking maneuvers or lateral maneuvers of other vehicles (Bazilinskyy et al., 2020; Yurtsever et al., 2019).”
BackgroundInfluentialWang, J., Malawade, A. V., Zhou, J., Yu, S.-Y., & Faruque, M. A. A. (2024) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv57701.2024.00732 · Cited by 71
““None” graph extraction refers to SOTA DL-based model [41] without using graphs, where the downstream task processes raw image data using a CNN and an LSTM model.”
MethodInfluentialReview of Accident Detection Methods Using Dashcam Videos for Autonomous Driving Vehicles
Rocky, A., Wu, Q. M. J., & Zhang, W. (2024) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2024.3354852 · Cited by 37
“For instance, [10] used Mask R-CNN for semantic segmentation and then used a CNN-LSTM network to classify dangerous lane change detection based on the frame overlays with these masks.”
MethodZhang, Z., Lu, C., Cui, G., Meng, X., Gong, C. Y., & Gong, J. (2024) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2024.3356658 · Cited by 25
“To directly learn risk patterns from data, risk assessment models based on supervised learning have been proposed [27], [28], [29], where history scenario information is the input, and the risk score (or label) is the output.”
BackgroundIzquierdo, R., Quintanar, A., Lorenzo, J., Daza, I. G., Parra, I., Llorca, D. F., & Sotelo, M. Á. (2021) · IEEE Access · doi:10.1109/access.2021.3106692 · Cited by 19
DROID: Driver-Centric Risk Object IDentification
Li, C., Chan, S. H., & Chen, Y. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3294305 · Cited by 24
“Existingworks[22],[64], [65],[66]formulate subjective risk assessment as risk-level classification, which requires human annotators to label the risk level of a scenario (e.g., lane change).”
BackgroundA Review of UTDrive Studies: Learning Driver Behavior From Naturalistic Driving Data
Liu, Y., & Hansen, J. H. L. (2021) · IEEE Open Journal of Intelligent Transportation Systems · doi:10.1109/ojits.2021.3109039 · Cited by 20
Self-citationDistributionally Robust CVaR-Based Safety Filtering for Motion Planning in Uncertain Environments
Safaoui, S., & Summers, T. H. (2024) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra57147.2024.10611276 · Cited by 21
HazardVLM: A Video Language Model for Real-Time Hazard Description in Automated Driving Systems
Xiao, D., Dianati, M., Jennings, P., & Woodman, R. (2024) · IEEE Transactions on Intelligent Vehicles · doi:10.1109/tiv.2024.3451350 · Cited by 20
“As input, networks take video data from onboard cameras that record scenes [13],[14],[15],[16],[17],[18].”
Liu, Z., & Xiang, Q. (2025) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2025.3547969 · Cited by 9
“Yurtsever et al. [38] recruited participants to watch driving videos from Japanese urban roads and used a Likert scale to score driving risks.”
Xiao, D., Dianati, M., Jennings, P., & Woodman, R. (2024) · IEEE Transactions on Intelligent Vehicles · doi:10.1109/tiv.2024.3392589 · Cited by 8
Scene-Graph Embedding for Robust Autonomous Vehicle Perception
Yu, S.-Y., Malawade, A. V., & Faruque, M. A. A. (2023) · Springer eBooks · doi:10.1007/978-3-031-28016-0_18 · Cited by 6
Liu, D., Li, D., Ding, H., Cao, Y., & Gao, K. (2025) · Transportation Research Part C Emerging Technologies · doi:10.1016/j.trc.2025.105270 · Cited by 9
Hazards Prioritization With Cognitive Attention Maps for Supporting Driving Decision-Making
Huang, Y., & Wang, X. (2024) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2024.3413675 · Cited by 6
“Meanwhile, though remarkable studies have been proposed for risky scenarios or behaviors classification [6], [7], [8] in specific driving manoeuvres, collision prediction [9] or intervention [10], road anomaly detection [11], eye fixation prediction in critical or accident scenarios [12], and…”
BackgroundPundlik, S. J., Choe, S., Baker, P., Lee, C.-Y., Madi, N. A., Bowers, A. R., & Luo, G. (2026) · Scientific Reports · doi:10.1038/s41598-026-53710-1 · Cited by 3
A survey of video human behaviour recognition Methodologies in the Perspective of Spatial-Temporal
Wang, Z., Zheng, Y., Liu, Z., & Li, Y. (2022) · International Conference on Intelligent Technology and Embedded Systems (ICITES) · doi:10.1109/icites56274.2022.9943587 · Cited by 3
“During the research work, as well as the original propose of Mask R-CNN, the mask means to render the detection object or behaviours with color information[58].”
MethodDynamic Attention-Enhanced Spatio-Temporal Network for Pedestrian Collision Risk Assessment
Gao, H., Wang, B., Liu, X., Zhou, Y., & Gou, C. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-8792-0_15 · Cited by 2
Monocular Vision-Based Prediction of Cut-In Manoeuvres with LSTM Networks
Nalçakan, Y., & Baştanlar, Y. (2023) · Communications in computer and information science · doi:10.1007/978-3-031-40395-8_8 · Cited by 2
Edge Feature-Enhanced Network for Collision Risk Assessment Using Traffic Scene Graphs
Liu, X., Zhou, Y., Ye, Y., & Gou, C. (2024) · IEEE Intelligent Transportation Systems Magazine · doi:10.1109/mits.2024.3458393 · Cited by 2
“…and 1043-carla-sg datasets using the fivefold cross-validation method, comparing our approach with six state-of-the-art methods, including an image data-based method [27] and graph data-based methods [32], [50], [54] (MR-GIN+LSTM-attn, MR-GCN+mean, MR-GCN+LSTM-last, MR-GCN+LSTM-attn, and PCRA).”
A Surrogate Approach for Real-Time Stress Assessment of Urban Drivers Using Driving Video Data
Wan, S., He, J. Q., Wu, X., Ye, Y., Qin, P., & Fang, Z. (2025) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2025.3538647 · Cited by 1
“Previous studies have shown that a driver’s cognitive process involves understanding both the global and spatiotemporal contexts of driving scenarios [15], [24].”
InfluentialHuang, Y., Berrío, J. S., Shan, M., & Worrall, S. (2026) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2026.115230 · Cited by 1
Manas, K., Keser, M., & Knoll, A. (2025) · IEEE International Automated Vehicle Validation Conference (IAVVC) · doi:10.1109/iavvc61942.2025.11219621 · Cited by 1
“Empirical results demonstrate significant performance gains, with HKTSG improving accuracy over CNN-LSTM [10] by 14.7% on the IESG dataset and out-performing PCRA [11] by 2-3% on both IESG and Non-IESG datasets [11].”
InfluentialLearning From Selective Attention for Pedestrian Collision Risk Assessment
Gao, H., Wang, B., Huang, Z., Liu, X., & Gou, C. (2026) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2026.3677481 · Cited by 1
“Techniques such as spatiotemporal 3D CNN networks [39] and CNN-LSTM networks [40] have demonstrated initial success by leveraging deep learning to predict collision risks.”
Liu, Y., & Hansen, J. (2021) · SAE technical papers on CD-ROM/SAE technical paper series · doi:10.4271/2021-01-0189 · Cited by 1
Self-citationOperator Motion Comparison with Standard using Computer Vision and Deep Learning
Yadav, A. A., Kadam, Y. B., Agashe, S. D., & Sutaone, M. (2022) · International Conference on Signal and Information Processing · doi:10.1109/iconsip49665.2022.10007473 · Cited by 1
Modeling driving task-relevant attention for intelligent vehicles using triplet ranking
Withanawasam, J., & Kamijo, S. (2023) · Machine Vision and Applications · doi:10.1007/s00138-023-01437-8 · Cited by 1
“Recognition methods are proposed for specific traffic scenarios such as detecting potentially risky lane change behaviors [39, 40] and localizing regions where pedestrians may appear [15].”
Enhancing Driver Safety through Real-Time Feedback on Driving Behavior: A Deep Learning Approach
Thakrar, O., Ranka, P., Pai, V., Correia, S., & Karani, R. B. (2023) · International Conference on Advanced Computing Technologies and Applications (ICACTA) · doi:10.1109/icacta58201.2023.10392647 · Cited by 1
“The paper concludes that the proposed approach and architecture can provide a stronger and more reliable solution for improving safety not only for physical drivers but also in autonomous driving and advanced driver assistance systems by recognizing risky actions [17].”
BackgroundWithanawasam, J., Javanmardi, E., Gu, Y., & Kamijo, S. (2021) · International Journal of Automotive Engineering · doi:10.20485/jsaeijae.12.4_125 · Cited by 1
Development and Key Technologies of the Cognition system of Autonomous Driving
Zhang, Y., Feng, J., Huang, M., Liu, Q., & Fu, W. (2023) · Highlights in Science Engineering and Technology · doi:10.54097/hset.v56i.10835
Driver-Centric Predictive Risk Map Modeling via Deep Reinforcement Learning
Chen, D., Lu, C., & Gong, J. (2024) · China Automation Congress (CAC) · doi:10.1109/cac63892.2024.10864604
“Several driving risk assessment frameworks for classifying dangerous behavior have been designed using convolutional neural networks and long-short term memory networks to model risk in both spatial and temporal domains [8], [9].”
HGAT-CP: Heterogeneous Graph Attention Network for Collision Prediction in Autonomous Driving
Jiang, Y., Zhou, B., Li, Y., Wu, X., & Xiong, Z. (2025) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra55743.2025.11128557
“We evaluate our proposed method against four strong baseline models: CNN-LSTM [14], MRGIN [17], DPM [12], and MRGCN [7].”
Marion, A. (2022) · Knowledge UdeS (Institutional Deposit of the University of Sherbrooke)
CTAMNet: an efficient video recognition method on edge devices
Li, B., Zheng, C., Wang, Z., Zhou, Y., & Zhang, W. (2026) · Journal of Real-Time Image Processing · doi:10.1007/s11554-026-01909-w
“Another approach is to use 2D CNN with post-processing fusion [36, 38], but this approach loses low-level temporal information.”
Deep Spatio-Temporal Memory and Interaction Network for Traffic Incident Detection
Hu, W., Zhang, Y., & Li, H. (2026) · Sensors · doi:10.3390/s26154979
“Mask R-CNN was used for semantic segmentation; then, the convolutional neural network–long short-term memory (CNN-LSTM) network was used to detect and classify dangerous lane changes based on Mask frame coverage [9].”
Metadata
- Authors
- Yurtsever, E., Liu, Y., Lambert, J., Miyajima, C., Takeuchi, E., Takeda, K., & Hansen, J. H. L.
- Affiliations
- Nagoya University (JP)
- The University of Texas at Dallas (US)
- Daido University (JP)
- ORCID
- Yurtsever, Ekim
- Venue
- 2019 IEEE Intelligent Transportation Systems Conference (ITSC)
- Publisher
- IEEE
- Published
- 2019-10
- Pages
- 3100-3107
- OpenAlex
- W2990460756
- Cited by
- 40
- References
- 40
- Citations by year
- 2026: 3 · 2025: 5 · 2024: 11 · 2023: 9 · 2022: 4 · 2021: 5 · 2020: 3
- FWCI
- 2.99
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE International Conference on Intelligent Transportation Systems (ITSC) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Topics
- Physical Sciences › Engineering › Automotive Engineering › Autonomous Vehicle Technology and Safety
- Physical Sciences › Engineering › Safety, Risk, Reliability and Quality › Traffic and Road Safety
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- lane changing, driving risk assessment, semantic segmentation masks, Mask R-CNN, spatio-temporal networks, CNN-LSTM, dangerous driving detection, driver assistance systems, autonomous driving
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.
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.