Computer Vision for Driving Scene Understanding

Computer Vision for Driving Scene Understanding: from Autonomous Driving to Road Condition Assessment

With recent advances in machine/deep learning, computer vision techniques have been extensively applied for various driving scene understanding applications, ranging from autonomous driving to road condition assessment. This talk will first show a big picture of the SoTA computer vision algorithms applied for driving scene understanding. It will then introduce several accomplished driving scene understanding projects, including (a) 3-D information (disparity/depth, optical flow, surface normal, etc.) estimation, and (b) collision-free space, lane marking, road anomaly/damage detection, etc. The major contributions of these works have been published in top-tier conferences/journals. Finally, the talk will conclude with existing challenges and discuss possible future works.

Lecture slides: https://drive.google.com/file/d/11mE0i14QOH1o03nKktyGrNzbEr52e-wG

Talk is based on the speakers' papers:

3-D information acquisition:

• CoT-AMFlow: Adaptive Modulation Network with Co-Teaching Strategy for Unsupervised Optical Flow Estimation (CoRL 2020) - https://arxiv.org/abs/2011.02156
• PVStereo: Pyramid Voting Module for End-to-End Self-Supervised Stereo Matching - https://www.ruirangerfan.com/pdf/ral2021_wang.pdf
• ATG-PVD: Ticketing Parking Violations on a Drone (ECCV 2020 workshop) - https://www.ruirangerfan.com/pdf/eccvw2020_wang.pdf
  https://sites.google.com/view/atg-pvd/home
• Three-Filters-to-Normal: An Accurate and Ultrafast Surface Normal Estimator (RA-L and ICRA'21) - https://arxiv.org/abs/2005.08165
  git:
  https://github.com/ruirangerfan/Three-Filters-to-Normal

Lane Marking Detection:

• Multiple Lane Detection Algorithm Based on Novel Dense Vanishing Point Estimation - https://www.ruirangerfan.com/pdf/tits2016_umar.pdf

Freespace & Road Anomaly Detection:

• SNE-RoadSeg: Incorporating Surface Normal Information into Semantic Segmentation for Accurate Freespace Detection (ECCV 2020) -
  https://arxiv.org/abs/2008.11351
• Dynamic Fusion Module Evolves Drivable Area and Road Anomaly Detection: A Benchmark and Algorithms - https://www.ruirangerfan.com/pdf/tcyb2021_wang.pdf
• Learning Collision-Free Space Detection from Stereo Images: Homography Matrix Brings Better Data Augmentation - https://arxiv.org/abs/2012.07890
  git: https://github.com/hlwang1124/SNE-RoadSeg

Road Condition Assessment:
papers:

• Road Surface 3D Reconstruction Based on Dense Subpixel Disparity Map Estimation (T-IP) - https://www.ruirangerfan.com/pdf/tip2018_fan.pdf
• Real-Time Dense Stereo Embedded in A UAV for Road Inspection - https://arxiv.org/abs/1904.06017
• We Learn Better Road Pothole Detection: From Attention Aggregation to Adversarial Domain Adaptation - https://www.ruirangerfan.com/pdf/eccvw2020_fan.pdf
• Road Damage Detection Based on Unsupervised Disparity Map Segmentation (T-ITS) - https://www.ruirangerfan.com/pdf/tits2019_fan.pdf
• Rethinking Road Surface 3D Reconstruction and Pothole Detection: From Perspective Transformation to Disparity Map Segmentation - https://arxiv.org/abs/2012.10802
  gits: https://github.com/ruirangerfan/road_surface_3d_reconstruction_datasets
  https://github.com/ruirangerfan/unsupervised_disparity_map_segmentation
  https://github.com/ruirangerfan/rethinking_road_reconstruction_pothole_detection

Presenter BIO:

Dr. Rui Ranger Fan received his B.Eng. Degree from the Harbin Institute of Technology and his Ph.D. degree from the University of Bristol. Rui is currently a research professor at Tongji University. Rui is also the General Chair of the Autonomous Vehicle Vision (AVVision) Community.
Rui’s research interests include computer vision, machine learning, robotics, and image processing.
More information about Rui can be found at www.ruirangerfan.com.