CoChat
Sign Up
Cited by 21Autonomous Vehicle Technology and SafetyImage Enhancement TechniquesAdvanced Vision and Imaging

A Review of Detection and Removal of Raindrops in Automotive Vision Systems

Open access
Hamzeh, Yazan, Rawashdeh, Samir A.·2021·Journal of Imaging
DOI 10.3390/jimaging7030052PMC PMC8321291
Open this paper in CoChat — free accountOpen PDF

Abstract

Research on the effect of adverse weather conditions on the performance of vision-based algorithms for automotive tasks has had significant interest. It is generally accepted that adverse weather conditions reduce the quality of captured images and have a detrimental effect on the performance of algorithms that rely on these images. Rain is a common and significant source of image quality degradation. Adherent rain on a vehicle’s windshield in the camera’s field of view causes distortion that affects a wide range of essential automotive perception tasks, such as object recognition, traffic sign recognition, localization, mapping, and other advanced driver assist systems (ADAS) and self-driving features. As rain is a common occurrence and as these systems are safety-critical, algorithm reliability in the presence of rain and potential countermeasures must be well understood. This survey paper describes the main techniques for detecting and removing adherent raindrops from images that accumulate on the protective cover of cameras.

AI snapshot

Written from the paper’s full text.

TL;DR

This survey reviews detection and removal of adherent raindrops on automotive camera covers and windshields, covering classical image-processing methods and deep-learning/GAN approaches. It concludes that adherent raindrops are harder to handle than falling rain streaks, classical methods degrade under heavy rain or complex scenes, and CNN/GAN methods are promising though PSNR/SSIM benchmarking may be unreliable.

Key findings

  • Adherent raindrop detection and removal is more challenging than falling-rain detection/removal because raindrops persist over many frames, have irregular shapes and sizes, appear blurry due to closeness to the image plane, and occlude larger image areas.
  • Classical methods show varied reported performance: a pan-tilt camera approach removed 92.6% of noises; Liao et al. achieved PSNR above 30 dB for light raindrop concentrations but processing took 732–972 ms; Ishizuka and Onoguchi achieved 10 frames/s daytime and 30 frames/s nighttime detection; eMSER precision/recall ranged from (0.8, 0.38) to (0.68, 0.5).
  • The Bézier-curve raindrop model reduced error by up to three orders of magnitude versus a spherical model, but practical use is limited because the relation between raindrop volume and diameter is nonlinear and, with inclination angle, not monotonic.
  • Deep-learning/GAN methods such as DeRaindrop, Quan et al., Hoa et al., Peng et al., and Ren et al. generally outperformed classical or CNN baselines on PSNR/SSIM in the reviewed comparisons, but the paper cautions that PSNR and SSIM may not be the best benchmarking choice.

Methods

  • Design: narrative literature survey/review, not an experimental study.
  • Data/sample: synthesizes prior research on raindrop detection, raindrop modeling, and raindrop removal for automotive and surveillance applications; no new dataset or primary experiments are reported.
  • Main analysis: organizes the field into detection and rain-intensity estimation, adherent raindrop models, non-automotive removal techniques, classical automotive detection/removal, and deep-learning/CNN/GAN methods, with summary tables and conclusions.

Limitations

  • Stated in paper: most reviewed algorithms performed poorly under heavy rain conditions or fast-changing scenes with many moving objects; complex detection algorithms can introduce unacceptable real-time latency; PSNR/SSIM may not be the best evaluation metrics, and reported marginal improvements may fall within statistical margin of error.
  • Stated in paper: several reviewed methods have practical constraints, including requiring the camera to focus on the windshield, accurate rotation-angle information, lane markings or buildings for segmentation, or non-real-time post-processing.
  • Evident from text: as a narrative review, it reports no new experimental validation or standardized benchmark; the summary tables (Tables 3–6) are referenced but not reproduced in the provided text, so exact cross-method quantitative comparisons cannot be verified from the text alone.

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

Integrity signals

Registry

Crossref · 10.3390/jimaging7030052

Cites

71 of 71 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

6 of 115 cited sources checked · 3 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Eigen et al. proposed a system to remove dirt and raindrop noise from a single image using the NN architecture originally developed by LeCun et al.

    “Eigen et al. [46] proposed a system to remove dirt and raindrop noise from a single image, using the NN architecture originally developed by LeCun et al. [47].”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [46] Eigen, D., Krishnan, D., and Fergus, R. (2013, January 1–8). Restoring an Image Taken through a Window Covered with Dirt or Rain. Proceedings of the 2013 IEEE International Conference on Computer Vision, Sydney, NSW, Australia. source
    • cites [47] LeCun, 1998, Gradient-based learning applied to document recognition, Proc. IEEE source
  • Premisecitation

    Hoa et al.'s proposed algorithm performed better than all other algorithms in terms of SSIM and PSNR scores.

    “The quantitative results showed that Hoa et al. proposed algorithm performed better than all other algorithms in terms of SSIM and PSNR scores.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [56] Hao, Z., You, S., Li, Y., Li, K., and Lu, F. (2019, January 27–28). Learning from synthetic photorealistic raindrop for single image raindrop removal. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea. source

      Section 5.1 reports that the complete model outperformed the compared methods on both PSNR and SSIM on the synthetic test set. The claim omits that setting: Section 5.2 says its quantitative results on the real-world dataset were on par with Qian’s method, not better than every comparator. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Premisecitation

    Hoa et al.'s proposed algorithm generated de-rained images with the least artifacts compared to other benchmarked methods.

    “The proposed algorithm as well generated de-rained images with the least artifacts, as compared to other methods benchmarked.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [56] Hao, Z., You, S., Li, Y., Li, K., and Lu, F. (2019, January 27–28). Learning from synthetic photorealistic raindrop for single image raindrop removal. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea. source
  • Premisecitation

    Due to their irregular shapes, closeness to image sensors, and long temporal presence in captured image frames, adherent raindrop detection and removal problem is relatively harder than that of rain streaks detection and removal.

    “Due to their irregular shapes, closeness to image sensors, and long temporal presence in captured image frames, adherent raindrop detection and removal problem is relatively harder than that of rain streaks detection and removal [11,12].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [11] Fouad, E., Salma, A., and Abdelhak, E. (April, January 29). Modelisation of raindrops based on declivity principle. Proceedings of the 2016 13th International Conference on Computer Graphics, Imaging and Visualization (CGiV), Beni Mellal, Morocco. source
    • cites [12] Halimeh, J.C., and Roser, M. (2009, January 3–5). Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source
  • Interpretationreasoning

    Other characteristics such as raindrop shape and persistence in an image sequence are quite different, which makes models developed for falling raindrops detection and replacement not applicable for adherent raindrops on windshields.

    “Other characteristics such as raindrop shape and persistence in an image sequence are quite different, which makes models developed for falling raindrops detection and replacement not applicable for adherent raindrops on windshields.”
    Quote in the paperSubstantial claim
  • May combine two claims. 44
  • Classified as probably not a substantial claim. 13

  • May combine two claims.

    Premisecitation

    Hoa et al. developed a synthetic raindrop generator and used synthetically generated rainy scenes along with real rainy scenes to train their GAN-based raindrop removal algorithm.

    “Hoa et al. [56] tried to solve this problem by developing a synthetic raindrop generator and using the synthetically generated rainy scenes, along with real raindrop real rainy scenes to train their GAN-based raindrop removal algorithm.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [56] Hao, Z., You, S., Li, Y., Li, K., and Lu, F. (2019, January 27–28). Learning from synthetic photorealistic raindrop for single image raindrop removal. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea. source

      Section 3 describes generating photorealistic raindrops for a synthetic training dataset, and Section 5.2 describes training on a real-world raindrop dataset. But Section 4 describes the authors’ removal method as a three-subnetwork detection-and-reconstruction architecture, not a GAN; Section 2.2 attributes the GAN-based method to Qian et al. The paper also does not say the synthetic and real scenes were combined in one training set. (LLM, conf=0.90)

      Experimental — accuracy not yet measured
    • Rests on an assumption: A large enough set of clear/rained image pairs of the same scene is necessary to train a CNN-based raindrop removal system.

  • May combine two claims.

    Premisecitation

    When trained using Qian et al.'s dataset with only a rough mask, Hoa et al.'s proposed algorithm produced results on-par with Qian and better than all other methods.

    “Results showed that the proposed algorithm produced results that were on-par with Qian [49] and better than all other methods.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [56] Hao, Z., You, S., Li, Y., Li, K., and Lu, F. (2019, January 27–28). Learning from synthetic photorealistic raindrop for single image raindrop removal. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea. source

      Section 5.2 says the methods trained on Qian’s dataset using rough masks achieved quantitative results on par with Qian’s. The statement that the proposed method outperformed other methods appears in Section 5.1, but it concerns the complete model on a synthetic dataset, not the rough-mask model on Qian’s dataset. Page numbers are unavailable in our copy. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [49] Qian, R., Tan, R.T., Yang, W., Su, J., and Liu, J. (2018, January 18–22). Attentive generative adversarial network for raindrop removal from a single image. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA. source

  • May combine two claims.

    Premiseown data

    Qualitatively, Hoa et al.'s proposed algorithm, as well as Qian's and DID-MDN, removed most raindrops successfully.

    “Qualitatively, the proposed algorithm, as well as Qian’s and the DIID-MDN removed most raindrops successfully.”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [56] Hao, Z., You, S., Li, Y., Li, K., and Lu, F. (2019, January 27–28). Learning from synthetic photorealistic raindrop for single image raindrop removal. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea. source
    • cites [49] Qian, R., Tan, R.T., Yang, W., Su, J., and Liu, J. (2018, January 18–22). Attentive generative adversarial network for raindrop removal from a single image. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA. source
    • cites [60] Zhang, H., and Patel, V.M. (2018, January 18–22). Density-aware single image de-raining using a multi-stream dense network. Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, USA. source

  • May combine two claims.

    Premisecitation

    Roser and Moosmann's classifier showed a good rate of classification for the highway-only subset (2% error) but the classification rate got worse as the complexity of the driving environment increased.

    “Results showed a good rate of classification for the highway-only subset (2% error) but the classification rate got worse as the complexity of the driving environment increased.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] Röser, M., and Moosmann, F. (2008, January 4–6). Classification of weather situations on single color images. Proceedings of the 2008 IEEE Intelligent Vehicles Symposium, Eindhoven, The Netherlands. source

  • May combine two claims.

    Premisecitation

    Yan et al.'s HSV classifier performed better than KNN and AdaBoost.MH, but slightly lower than AdaBoost with one-vs-all.

    “The algorithm performed better than KNN and AdaBoost.MH, but it was slightly lower than AdaBoost with one-vs-all.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] Yan, X., Luo, Y., and Zheng, X. (2009, January 26–29). Weather recognition based on images captured by vision system in vehicle. Proceedings of the Advances in Neural Networks—6th International Symposium on Neural Networks (ISNN 2009), Wuhan, China. source

  • May combine two claims.

    Premisecitation

    Camera-based rain-sensing provides benefits over FIR-based ones, in terms of improved detection rate, glare reduction, and detection of other particles like dust or salt on the windshield.

    “Camera-based rain-sensing provides benefits over FIR-based ones, in terms of improved detection rate, glare reduction, and detection of other particles like dust or salt on the windshield.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [13] Gormer, S., Kummert, A., Park, S.-B., and Egbert, P. (2009, January 3–5). Vision-based rain sensing with an in-vehicle camera. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source
    • cites [14] Yan, X., Luo, Y., and Zheng, X. (2009, January 26–29). Weather recognition based on images captured by vision system in vehicle. Proceedings of the Advances in Neural Networks—6th International Symposium on Neural Networks (ISNN 2009), Wuhan, China. source

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

    Premisecitation

    Barnum et al. [5] observed a 15% to 30% increase in the number of feature points tracked successfully in the de-rained image sequence vs. rainy sequence.

    “A 15% to 30% increase in the number of feature points tracked successfully was observed in the de-rained image sequence vs. rainy sequence.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5] Barnum, P., Kanade, T., and Narasimhan, S.G. (2007, January 14–21). Spatio-temporal frequency analysis for removing rain and snow from videos. Proceedings of the Workshop on Photometric Analysis for Computer Vision (PACV), in Conjunction with ICCV, Rio de Janeiro, Brazil.

  • May combine two claims.

    Interpretationnone given

    Yan et al.'s algorithm is good for classifying rainy conditions but less appropriate for vehicle ADAS applications.

    “This algorithm is good for classifying rainy conditions but it is less appropriate for vehicle ADAS applications.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] Yan, X., Luo, Y., and Zheng, X. (2009, January 26–29). Weather recognition based on images captured by vision system in vehicle. Proceedings of the Advances in Neural Networks—6th International Symposium on Neural Networks (ISNN 2009), Wuhan, China. source

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

    Premisecitation

    In applications that use LIDAR and radar, rain attenuates the strength of the transmitted signals and introduces noise.

    “In applications that use LIDAR and radar, rain attenuates the strength of the transmitted signals and introduces noise [1].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] Garg, 2007, Vision and rain, Int. J. Comput. Vis. source

  • May combine two claims.

    Premisecitation

    The NIR detection region is relatively small, and system performance degrades with induced infrared beams from the outside environment.

    “The detection region is still relatively small, and system performance degrades with induced infrared beams from the outside environment.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [13] Gormer, S., Kummert, A., Park, S.-B., and Egbert, P. (2009, January 3–5). Vision-based rain sensing with an in-vehicle camera. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

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

    Premisecitation

    Chia et al. [4] observed that the number of feature points extracted by Harris and SURF detectors was reduced by 48% and 68%, respectively due to falling rain.

    “Chia et al. [4] observed that the number of feature points extracted by Harris and SURF detectors was reduced by 48% and 68%, respectively due to falling rain.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4] Chia, W.C., Yeong, L.S., Ch’Ng, S.I., Seng, K.P., and Ang, L.-M. (2014, January 26–28). The effect of rainfall on feature points extraction and image stitching. Proceedings of the 2014 International Conference on Information Science, Electronics and Electrical Engineering, Sapporo, Japan. source

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

    Premisecitation

    Garg and Nayar [1] broadly classify adverse weather conditions into steady (fog, mist, and haze) or dynamic (rain, snow, and hail).

    “Garg and Nayar [1] broadly classify adverse weather conditions into steady (fog, mist, and haze) or dynamic (rain, snow, and hail).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] Garg, 2007, Vision and rain, Int. J. Comput. Vis. source

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

    Interpretationnone given

    The camera-based rain detection system by Gormer et al. [13] is dedicated to raindrop detection and cannot be used for any other image-based applications, such as lane-keeping or traffic sign recognition.

    “This system is dedicated to raindrop detection and cannot be used for any other image-based applications, such as lane-keeping or traffic sign recognition.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [13] Gormer, S., Kummert, A., Park, S.-B., and Egbert, P. (2009, January 3–5). Vision-based rain sensing with an in-vehicle camera. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

  • May combine two claims.

    Interpretationcitation

    Practical usage of the Bézier curve method in a real-time image-based application is limited by the accuracy of estimating raindrop volume, which is related to raindrop diameter; this relation is not linear and, when inclination angle is considered, is not even monotonic.

    “Despite excellent representation of raindrop shape using this method, practical usage in a real-time image-based application is limited by the accuracy of estimating raindrop volume, which is related to raindrop diameter. This relation is not linear and, when the inclination angle is considered, is not even monotonic.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] Roser, M., Kurz, J., and Geiger, A. (2010). Realistic modeling of water droplets for monocular adherent raindrop recognition using Bézier curves. Asian Conference on Computer Vision, Springer. source

  • May combine two claims.

    Premiseown data

    Wu et al.'s proposed saliency-map algorithm performed better than Kurihata et al. [18] Eigendrops and Roser and Geiger [21] initial raindrop detection stage, both in terms of Precision and Recall metrics.

    “Results showed that the proposed algorithm performed better than the other two, both in terms of Precision and Recall metrics.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [35] Wu, Q., Zhang, W., and Kumar, B.V. (October, January 30). Raindrop detection and removal using salient visual features. Proceedings of the 2012 19th IEEE International Conference on Image Processing, Orlando, FL, USA. source
    • cites [18] Kurihata, H., Takahashi, T., Ide, I., Mekada, Y., Murase, H., Tamatsu, Y., and Miyahara, T. (2005, January 6–8). Rainy weather recognition from in-vehicle camera images for driver assistance. Proceedings of the IEEE Intelligent Vehicles Symposium, Las Vegas, NV, USA. source
    • cites [21] Roser, M., and Geiger, A. (October, January 27). Video-based raindrop detection for improved image registration. Proceedings of the in 2009 IEEE 12th International Conference on Computer Vision Workshops, ICCV Workshops, Kyoto, Japan. source

  • May combine two claims.

    Premisecitation

    Ishizuka and Onoguchi proposed two raindrop detection algorithms, one for daytime and one for nighttime.

    “Ishizuka and Onoguchi [37], proposed two algorithms for raindrop detection; one for daytime and the other for night.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [37] Ishizuka, J., and Onoguchi, K. (2016, January 24–26). Detection of raindrop with various shapes on a windshield. Proceedings of the 5th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2016), Roma, Italy. source

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

    Premisecitation

    Processing time was 10 frames per second for daytime and 30 frames per second for nighttime raindrop detection.

    “Processing time was 10 and 30 frames/second for daytime and nighttime raindrop detection, respectively.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [37] Ishizuka, J., and Onoguchi, K. (2016, January 24–26). Detection of raindrop with various shapes on a windshield. Proceedings of the 5th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2016), Roma, Italy. source

  • May combine two claims.

    Premisecitation

    In daytime tests, most raindrops were correctly detected, but some in the sky region were missed.

    “At daytime, results showed that most raindrops were correctly detected, but some ones in the sky region were missed.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [37] Ishizuka, J., and Onoguchi, K. (2016, January 24–26). Detection of raindrop with various shapes on a windshield. Proceedings of the 5th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2016), Roma, Italy. source

  • May combine two claims.

    Interpretationcitation

    Halimeh and Roser's RIGSEC model is cohesive and dynamic, and considers the effect of background variations on raindrop appearance at different areas of the captured image.

    “The RIGSEC model is a cohesive, dynamic model, that considers the effect of background variations on raindrop appearance, at different areas of the captured image.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] Halimeh, J.C., and Roser, M. (2009, January 3–5). Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

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

    Premisecitation

    Cord and Gimonet proposed an adherent raindrop detection algorithm that uses either background subtraction or a watershed approach.

    “Cord and Gimonet [38], proposed an adherent raindrop detection algorithm that uses either background subtraction or a watershed approach for identifying raindrop candidates.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [38] Cord, 2014, Detecting unfocused raindrops: In-vehicle multipurpose cameras, IEEE Robot. Autom. Mag. source

  • May combine two claims.

    Interpretationcitation

    Liao et al.'s overall processing time was between 732 and 972 ms, which might be too long for real-time image-based automotive applications.

    “The overall processing time was between 732 and 972 ms which might be too long for real-time image-based automotive applications.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [36] Liao, 2013, Video-based water drop detection and removal method for a moving vehicle, Inf. Technol. J. source

  • May combine two claims.

    Interpretationcitation

    To create Eigendrops, the camera needs to be focused on the windshield, which means it cannot be used for other vision-based automotive applications.

    “Furthermore, to achieve strong and distinctive characteristics that are necessary to create the Eigendrops, the camera needs to be focused on the windshield which means it cannot be used for other vision-based automotive applications.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] Kurihata, H., Takahashi, T., Ide, I., Mekada, Y., Murase, H., Tamatsu, Y., and Miyahara, T. (2005, January 6–8). Rainy weather recognition from in-vehicle camera images for driver assistance. Proceedings of the IEEE Intelligent Vehicles Symposium, Las Vegas, NV, USA. source

  • May combine two claims.

    Interpretationcitation

    Liao et al.'s technique relied on on-road lane markings and complex building arrangements for image segmentation and raindrop detection; it is not clear how the process would perform with the absence of road marks or lack of buildings.

    “Another drawback of the proposed technique is that it relied on on-road lane markings and complex building arrangements for image segmentation and raindrop detection. It is not clear how the process would perform with the absence of road marks or lack of buildings.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] Liao, 2013, Video-based water drop detection and removal method for a moving vehicle, Inf. Technol. J. source

  • May combine two claims.

    Premisecitation

    Wu et al.'s proposed algorithm successfully removed raindrops from images of the windshield at slow driving speed but did not perform as well at higher speeds, due to increased raindrop shape deviation from the assumed circular shape.

    “The Proposed algorithm also successfully removed raindrops from images of the windshield at slow driving speed but did not perform as well at higher speeds, due to increased raindrop shape deviation from the assumed circular shape.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [35] Wu, Q., Zhang, W., and Kumar, B.V. (October, January 30). Raindrop detection and removal using salient visual features. Proceedings of the 2012 19th IEEE International Conference on Image Processing, Orlando, FL, USA. source
    • The authors state this as an assumption: Raindrops have a circular shape.

  • May combine two claims.

    Interpretationcitation

    The pan-tilt system requires an accurate measure of rotation angle at all times; otherwise, rotational calculations and corrections will be skewed and system performance degrades.

    “The main drawback of this system is that an accurate measure of rotation angle must be available at all times. Otherwise, all rotational calculations and corrections will be skewed and system performance degrades.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [29] Yamashita, A., Kaneko, T., and Miura, K. (May, January 26). A virtual wiper-restoration of deteriorated images by using a pan-tilt camera. Proceedings of the IEEE International Conference on Robotics and Automation, New Orleans, LA, USA.

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

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

    Interpretationreasoning

    The enhanced pan-tilt system still relies on a rotating camera; if camera rotation is required solely for detection and removing raindrops, the added cost and complexity may be prohibitive for most automotive applications.

    “This system still relies on a rotating camera which is acceptable if it is the type of camera used in the surveillance system. If, however, camera rotation is required solely for detection and removing raindrops, then the added cost and complexity may be prohibitive for most automotive applications.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [30] Yamashita, A., Harada, T., Kaneko, T., and Miura, K.T. (October, January 28). Removal of adherent noises from images of dynamic scenes by using a pan-tilt camera. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Sendai, Japan.

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

  • May combine two claims.

    Interpretationreasoning

    Roser and Moosmann's system does not detect individual raindrops on vehicle windshield, which makes it less appropriate for ADAS and autonomous driving applications.

    “Roser and Moosmann’s system does not detect individual raindrops on vehicle windshield which makes it less appropriate for ADAS and autonomous driving applications.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] Röser, M., and Moosmann, F. (2008, January 4–6). Classification of weather situations on single color images. Proceedings of the 2008 IEEE Intelligent Vehicles Symposium, Eindhoven, The Netherlands. source

  • May combine two claims.

    Interpretationcitation

    The spatio-temporal algorithm requires collecting and post-processing a sequence of image frames over time, which makes it not appropriate for real-time automotive applications.

    “This algorithm requires collecting and post-processing a sequence of image frames over time which makes it not appropriate for real-time automotive applications.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [33] Yamashita, A., Fukuchi, I., Kaneko, T., and Miura, K.T. (2008, January 19–23). Removal of adherent noises from image sequences by spatio-temporal image processing. Proceedings of the IEEE International Conference on Robotics and Automation, Pasadena, CA, USA. source

  • May combine two claims.

    Interpretationnone given

    Kurihata et al.'s Eigendrops model accounts for all variations of raindrop shape and size, and the effect of image background on raindrop appearance.

    “This model does account for all variations of raindrop shape and size, and the effect of image background on raindrop appearance.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] Kurihata, H., Takahashi, T., Ide, I., Mekada, Y., Murase, H., Tamatsu, Y., and Miyahara, T. (2005, January 6–8). Rainy weather recognition from in-vehicle camera images for driver assistance. Proceedings of the IEEE Intelligent Vehicles Symposium, Las Vegas, NV, USA. source

  • May combine two claims.

    Premisecitation

    Liao et al.'s process assumed a straight driving scenario, and it was difficult to identify raindrop patches when the vehicle was turning.

    “In addition, the process assumed a straight driving scenario, and Liao et al. observed that it was difficult to identify raindrop patches when the vehicle was turning.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] Liao, 2013, Video-based water drop detection and removal method for a moving vehicle, Inf. Technol. J. source
    • The authors state this as an assumption: A straight driving scenario.

  • May combine two claims.

    Interpretationcitation

    The RIGSEC model is relatively complex and computationally expensive.

    “It is, however, relatively complex and computationally expensive.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] Halimeh, J.C., and Roser, M. (2009, January 3–5). Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

  • May combine two claims.

    Premisecitation

    Cord and Aubert's system showed good results detecting and classifying weather conditions, with minimum misclassifications observed soon after the windshield wiper wipe event.

    “The system showed good results detecting and classifying weather conditions, with minimum misclassifications observed soon after the windshield wiper wipe event.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [17] Cord, A., and Aubert, D. (2011, January 5–9). Towards rain detection through use of in-vehicle multipurpose cameras. Proceedings of the 2011 IEEE Intelligent Vehicles Symposium (IV), Baden-Baden, Germany. source

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

    Premisecitation

    Sugimoto et al. used the eMSER algorithm to detect potential raindrop blobs.

    “Sugimoto et al. [43] used Extended Maximally Stable External Region (eMSER) algorithm to detect potential raindrop blobs.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [43] Sugimoto, M., Kakiuchi, N., Ozaki, N., and Sugawara, R. (2012, January 16–19). A novel technique for raindrop detection on a car windshield using geometric-photometric model. Proceedings of the 15th International IEEE Conference on Intelligent Transportation Systems, Anchorage, AK, USA. source

  • May combine two claims.

    Premisecitation

    Real rain results showed that the proposed system successfully removed real raindrops but started failing with accumulation of water on the glass plate.

    “Real rain results showed that the proposed system successfully removed real raindrops from the captured images, but it started failing with the accumulation of water on the glass plate.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [46] Eigen, D., Krishnan, D., and Fergus, R. (2013, January 1–8). Restoring an Image Taken through a Window Covered with Dirt or Rain. Proceedings of the 2013 IEEE International Conference on Computer Vision, Sydney, NSW, Australia. source

  • May combine two claims.

    Premisecitation

    Further experiments showed that matching the constructed environment within a small ROI made raindrop positions converge to the optimum and CC values converge to MaxCorr values.

    “Further experiments showed that matching the constructed environment within a small ROI, raindrop positions converge to the optimum, and CC values converge to MaxCorr values as well.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] Halimeh, J.C., and Roser, M. (2009, January 3–5). Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

  • May combine two claims.

    Premiseown data

    Synthesized rain results showed that the proposed system outperformed the MLP system in removing more rain droplets and maintaining non-rain areas intact.

    “Synthesized rain that the proposed system outperformed the MLP system, in terms of removing more rain droplets and maintain non-rain areas of the image intact.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [46] Eigen, D., Krishnan, D., and Fergus, R. (2013, January 1–8). Restoring an Image Taken through a Window Covered with Dirt or Rain. Proceedings of the 2013 IEEE International Conference on Computer Vision, Sydney, NSW, Australia. source
    • cites [44] Burger, H.C., Schuler, C.J., and Harmeling, S. (2012, January 16–21). Image denoising: Can plain neural networks compete with BM3D?. Proceedings of the 2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, RI, USA. source

  • May combine two claims.

    Premisecitation

    Of the three evaluated, SURF performed worse and eMSER showed the best results, with precision and recall rates ranging from (0.8, 0.38) to (0.68, 0.5).

    “Of the three evaluated, SURF performed worse and eMSER showed the best results, with rates for (precision, recall) ranging from (0.8, 0.38) to (0.68, 0.5), respectively.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [43] Sugimoto, M., Kakiuchi, N., Ozaki, N., and Sugawara, R. (2012, January 16–19). A novel technique for raindrop detection on a car windshield using geometric-photometric model. Proceedings of the 15th International IEEE Conference on Intelligent Transportation Systems, Anchorage, AK, USA. source

  • May combine two claims.

    Premisecitation

    FastRIGSEC performed better than SURF and BLUR but fell behind BLUR + RIGSEC in terms of maximum precision achievable.

    “FastRIGSEC performed better than SURF and BLUR but fall behind (BLUR + RIGSEC) in terms of maximum precision achievable.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] Roser, M., and Geiger, A. (October, January 27). Video-based raindrop detection for improved image registration. Proceedings of the in 2009 IEEE 12th International Conference on Computer Vision Workshops, ICCV Workshops, Kyoto, Japan. source

  • May combine two claims.

    Premisecitation

    The proposed system failed to recognize large dirt noises, out-of-shape dirt, and dirt over areas not represented in the training set.

    “The proposed system failed to recognize large dirt noises and the ones that are out of shape, as well as dirt over areas that were not represented in the training set (e.g., bright orange cones).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [46] Eigen, D., Krishnan, D., and Fergus, R. (2013, January 1–8). Restoring an Image Taken through a Window Covered with Dirt or Rain. Proceedings of the 2013 IEEE International Conference on Computer Vision, Sydney, NSW, Australia. source

  • May combine two claims.

    Premisecitation

    Testing with snow and scratches resulted in near-zero PSNR, demonstrating that the system learned to remove dirt noise only.

    “Testing with other types of sensitized noise, namely snow, and scratches on the image surface resulted in the system described by Eigen et al. producing near-zero PSNR (peak signal to noise ratio) which demonstrates that the system learned to remove dirt noise only.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [46] Eigen, D., Krishnan, D., and Fergus, R. (2013, January 1–8). Restoring an Image Taken through a Window Covered with Dirt or Rain. Proceedings of the 2013 IEEE International Conference on Computer Vision, Sydney, NSW, Australia. source

  • Classified as probably not a substantial claim.

    Premisecitation

    Halimeh and Roser attributed the low CC results to the raindrop lens effect and blurriness of raindrop appearance due to camera focus near infinity.

    “Halimeh and Roser attributed this to the raindrop lens effect and blurriness of raindrop appearance, due to camera focus near infinity.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [12] Halimeh, J.C., and Roser, M. (2009, January 3–5). Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

  • May combine two claims.

    Premisecitation

    In image restoration, fastRIGSEC performed well but fell slightly behind BLUR + RIGSEC, especially with an increase in raindrop density.

    “In terms of image restoration, fastRIGSEC performed well but fall slightly behind (BLUR + RIGSEC), especially with an increase in raindrop density.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] Roser, M., and Geiger, A. (October, January 27). Video-based raindrop detection for improved image registration. Proceedings of the in 2009 IEEE 12th International Conference on Computer Vision Workshops, ICCV Workshops, Kyoto, Japan. source

  • May combine two claims.

    Premiseown data

    Quantitative results showed that Quan et al.'s algorithm outperformed the other benchmarked algorithms in PSNR and SSIM, though it fell slightly behind Qian et al.'s algorithm for PSNR measurement of luminance.

    “Quantitative results showed that the algorithm proposed by Quan et al. outperformed the other benchmarked algorithms in terms of PSNR and SSIM results, though it fell slightly behind Qian et al. algorithm for PSNR measurement of luminance.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [53] Quan, Y., Deng, S., Chen, Y., and Ji, H. (November, January 27). Deep learning for seeing through window with raindrops. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea.

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

    • cites [55] Qian, R., Tan, R.T., Yang, W., Su, J., and Liu, J. (2021, January 20). rui1996/DeRaindrop. Available online: https://github.com/rui1996/DeRaindrop.

      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).

    Interpretationcitation

    Processing speed was the main drawback of Cord and Gimonet's algorithm.

    “Processing speed was the main drawback of this algorithm.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [38] Cord, 2014, Detecting unfocused raindrops: In-vehicle multipurpose cameras, IEEE Robot. Autom. Mag. source

  • May combine two claims.

    Premisecitation

    MaxCorr results were good, but CC results were low.

    “Maxcorr results were also good but CC results were low.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [12] Halimeh, J.C., and Roser, M. (2009, January 3–5). Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation. Proceedings of the 2009 IEEE Intelligent Vehicles Symposium, Xi’an, China. source

  • May combine two claims.

    Premisecitation

    Real-life dirt results showed that the proposed system removed most dirt noise while the MLP left many more unremoved.

    “Real-life dirt results showed that the proposed system removed most dirt noise in the images, while the MLP left many more unremoved.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [46] Eigen, D., Krishnan, D., and Fergus, R. (2013, January 1–8). Restoring an Image Taken through a Window Covered with Dirt or Rain. Proceedings of the 2013 IEEE International Conference on Computer Vision, Sydney, NSW, Australia. source
    • cites [44] Burger, H.C., Schuler, C.J., and Harmeling, S. (2012, January 16–21). Image denoising: Can plain neural networks compete with BM3D?. Proceedings of the 2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, RI, USA. source

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

    Premisecitation

    FPPI stayed high, which was attributed to insufficient temporal filtering of 3 consecutive frames.

    “In terms of raindrop prediction, FFPI (a measure of false detection) stayed high, which was attributed to insufficient temporal filtering (3 consecutive frames).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [38] Cord, 2014, Detecting unfocused raindrops: In-vehicle multipurpose cameras, IEEE Robot. Autom. Mag. source

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

    Premisecitation

    Quan et al. proposed an unsupervised alternative approach using joint channel and shape-driven attentions to improve performance.

    “They proposed an alternative approach that is unsupervised and makes use of a joint channel and shape-driven attentions to improve performance.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [53] Quan, Y., Deng, S., Chen, Y., and Ji, H. (November, January 27). Deep learning for seeing through window with raindrops. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea.

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

  • May combine two claims.

    Premisecitation

    Peng et al.'s proposed solution slightly outperformed Qian's algorithm, with the added value of smaller system size.

    “The proposed solution also slightly outperformed Qian’s algorithm with the added value of smaller system size.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [49] Qian, R., Tan, R.T., Yang, W., Su, J., and Liu, J. (2018, January 18–22). Attentive generative adversarial network for raindrop removal from a single image. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA. source

  • May combine two claims.

    Premisecitation

    In a Fréchet Inception Distance (FID) test measuring temporal consistency of de-rained to clean videos, Alletto et al.'s proposed algorithm performed best and Vid2Vid performed worst.

    “Results showed that the proposed algorithm performed best and Vid2Vid performed worst.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [69] Wikipedia (2020, November 24). Fréchet Inception Distance. Available online: https://en.wikipedia.org/wiki/Fr%C3%A9chet_inception_distance.

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

  • May combine two claims.

    Premiseown data

    On the Anmok paired image dataset, Kim et al.'s proposed algorithm performed best, improving scores on all three metrics between distorted and restored images.

    “Kim et al. proposed algorithm performed best, improving scores on all three metrics between distorted and restored images.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premiseown data

    Quan et al.'s approach added noise and changed the color of some rain-free segments.

    “As a comparison, Quan et al. [53] approach added noise and changes the color of some rain-free segments.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [53] Quan, Y., Deng, S., Chen, Y., and Ji, H. (November, January 27). Deep learning for seeing through window with raindrops. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea.

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

  • May combine two claims.

    Premisecitation

    Using the Qian et al. [49] dataset for training and testing, both Attentive GAN [49] and Kim et al.'s proposed algorithm improved PSNR and SSIM scores of restored images versus distorted images.

    “Using Qian et al. [49] dataset for training and testing showed that both attentive GAN [49] and Kim et al. proposed algorithm improved quantitative score of PSNR and SSIM of restored image versus distorted one.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [49] Qian, R., Tan, R.T., Yang, W., Su, J., and Liu, J. (2018, January 18–22). Attentive generative adversarial network for raindrop removal from a single image. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA. source
53 of 53

  • The paper relies on this without support.

    Adverse weather conditions degrade the performance of many image and video-based algorithms used in the automotive domain.

    “Adverse weather conditions degrade the performance of many image and video-based algorithms used in the automotive domain.”

  • The paper relies on this without support.

    Rain is a common and significant source of image quality degradation.

    “Rain is a common and significant source of image quality degradation.”

  • The paper relies on this without support.

    Adherent rain on a vehicle’s windshield in the camera’s field of view causes distortion that affects a wide range of essential automotive perception tasks, such as object recognition, traffic sign recognition, localization, mapping, and other advanced driver assist systems (ADAS) and self-driving features.

    “Adherent rain on a vehicle’s windshield in the camera’s field of view causes distortion that affects a wide range of essential automotive perception tasks, such as object recognition, traffic sign recognition, localization, mapping, and other advanced driver assist systems (ADAS) and self-driving features.”

  • The paper relies on this without support.

    Raindrops in falling and adherent forms share some common characteristics, such as refraction properties of a raindrop, higher intensity of the drop compared to the background, and average raindrop size.

    “Raindrops in these two forms share some common characteristics, such as refraction properties of a raindrop, higher intensity of the drop compared to the background, and average raindrop size.”

  • The authors state this as an assumption.

    The raindrop-occluded section of an image would be clear in an image taken by another camera.

    “The main drawback of this system is that it assumes that the raindrop-occluded section of an image would be clear in an image taken by another camera.”

  • The authors state this as an assumption.

    The peak-valley appearance of raindrops holds.

    “The main issue with this algorithm is that the assumption of the peak-valley appearance of raindrops does not always hold.”

  • The authors state this as an assumption.

    Raindrops have a circular shape.

    “due to increased raindrop shape deviation from the assumed circular shape.”

  • The authors state this as an assumption.

    A straight driving scenario.

    “the process assumed a straight driving scenario”

  • The authors state this as an assumption.

    The MAP solution for refining Θ assumes independence of non-consecutive frames and normal distribution for the observation probability, and independence over translational and rotational parameters.

    “This assumes independence of non-consecutive frames and normal distribution for the observation probability PZi|Θi. Independence over translational and rotational parameters is also assumed”

  • The paper relies on this without support.

    A large enough set of clear/rained image pairs of the same scene is necessary to train a CNN-based raindrop removal system.

    “It is generally quite hard to capture a large enough set of clear/rained of the same scene, necessary to train the CNN-based system.”
10 of 10

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.

  • Spatial-attention formula uses B instead of the defined attention map Q

    Minor

    The text identifies the sigmoid-normalized map Q as the spatial significance used for recalibration, but the displayed recalibration formula multiplies by B, the map before sigmoid normalization.

  • Bézier-curve formula omits one of the stated control points

    Minor

    The description specifies n + 1 control points indexed from 0 through n, but the defining sum starts at 1 and therefore excludes P0.

  • Dice coefficient formula conflicts with its verbal definition

    Minor

    The stated formula uses twice the true positives in the numerator, whereas the accompanying definition says the numerator is the size of the union. Those are different quantities.

3 of 3

Cites

Flagged first, then by citations

8 of the 79 references couldn’t be matched to a known paper.

Research works printed but not linked: 3

  1. Stuppacher, I., and Supan, P. (2020, November 24). Rendering of Water Drops in Real-Time. In Proceedings of the Central European Seminar on Computer Graphics for Students. Available online: https://old.cescg.org/CESCG-2007/papers/Hagenberg-Stuppacher-Ines/cescg_StuppacherInes.pdf.
  2. Sousa, T. (2005). Chapter generic refraction simulation. GPU Gems 2: Programming Techniques for High-Performance Graphics and General-Purpose Computation, Addison-Wesley Professional.
  3. Kingma, D.P., and Ba, J. (2015, January 5–8). Adam: A method for stochastic optimization. Proceedings of the 3rd International Conference for Learning Representations (ICLR), San Diego, CA, USA.

Cited by

Sorted by citations

Related

Authors
Hamzeh, Y., & Rawashdeh, S. A.
Affiliations
  • Ford Motor Company (United States) (US)
  • University of Michigan–Dearborn (US)
Corresponding author
Hamzeh, Yazan
ORCID
Venue
Journal of Imaging
Publisher
MDPI AG
Published
2021-03-10
Volume
7
Issue
3
Pages
52
ISSN
2313-433X (ISSN-L)
DOI
10.3390/jimaging7030052
PMC
PMC8321291
OpenAlex
W3134932434
Semantic Scholar
79dd18cddb1350930c839911058a814e7aab0b90
Cited by
21
References
79
Citations by year
2026: 1 · 2025: 2 · 2024: 5 · 2023: 8 · 2022: 4 · 2021: 1
FWCI
1.44
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Imaging (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Physical Sciences › Engineering › Automotive Engineering › Autonomous Vehicle Technology and Safety
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Enhancement Techniques
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Vision and Imaging
Keywords
raindrop detection, raindrop removal, automotive vision, adverse weather, image degradation, driver assistance systems, autonomous driving, object recognition, traffic sign recognition, localization and mapping

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 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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