A3D stands for AnAn Accident Detection, which is a brand new dataset about various traffic accidents. Identifying traffic violations and accident events in autonomous driving scenarios is crucial for autonomous driving and driver assistance systems. Common traffic accident videos often assume that the camera is fixed and the video has a static background, but this is unreasonable for on-board cameras. Therefore, the videos in this dataset are based on an unsupervised first-person video traffic accident detection method, which mainly detects anomalies by predicting the future positions of traffic participants. This dataset contains videos from YouTube and a .pkl file, including manually annotated video start/end times and anomaly start/end times.

Citation

If you find this repo is useful, please feel free to cite our papers: @inproceedings{yao2018egocentric, title={Egocentric Vision-based Future Vehicle Localization for Intelligent Driving Assistance Systems}, author={Yao, Yu and Xu, Mingze and Choi, Chiho and Crandall, David J and Atkins, Ella M and Dariush, Behzad}, journal={IEEE International Conference on Robotics and Automation (ICRA)}, year={2019} } @inproceedings{yao2019unsupervised, title={Unsupervised Traffic Accident Detection in First-Person Videos}, author={Yao, Yu and Xu, Mingze and Wang, Yuchen and Crandall, David J and Atkins, Ella M}, journal={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, year={2019} }