CADP: A Novel Dataset for CCTV Traffic Camera based Accident Analysis
- CADP dataset provides samples for accident detection and forecasting type analysis
- Average length of videos in our dataset is 366 frames per video with longest video consisting of 554 frames
- Time to accident - duration from time 0 in video to onset of first accident in annotated videos is 3.69 seconds
|
|
|
|
|
Abstract
We present a novel dataset for traffic accidents analysis. Our aim is to resolve the lack of public data for research about automatic spatio-temporal annotations for traffic safety in the roads. Our Car Accident Detection and Prediction~(CADP) dataset consists of 1,416 video segments collected from YouTube, with 205 video segments have full spatio-temporal annotations. Through analysis of CADP dataset, we observed a significant degradation of object detection in pedestrian category in our dataset, due to the object sizes and complexity of the scenes. To this end, we propose to integrate the Augmented Context Mining (ACM) into the Faster R-CNN detector to complement the accuracy for small pedestrian detection. Our experiments indicated a considerable improvement in object detection accuracy +8.51% for CM and +6.20% for ACM. For person~(pedestrian) category, we observed significant improvements:~+46.45\% for CM and +45.22\% for ACM, compared to Faster R-CNN. Finally, we demonstrate the performance of accident forecasting in our dataset using our Imporved Faster R-CNN and the Accident LSTM architectures. We achieved an average 1.684 seconds in terms of Time-To-Accident measure with highest Average Precision is 47.25 %. We expect our dataset can serve as the starting point of a new research direction, which can grow incrementally in coming years.
Dataset Information
Dataset Comparison
Network Architecture
Paper
|
|
Citation |
Dataset Download
The dataset is available here 1. Dataset Release README - contains all necessary information Additional Information Links 2. Extracted Frames from video clips 3. Annotations json file 4. Duration information of videos 5. Annotation Guideline for Spatio Temporal Location Annotation Tasks Note: By downloading this dataset you agree to use this dataset for non-commercial and research based purposes and prior permission is required to be sought of authors for other use.
Slide Presentation
6. CADP Slide Presentation explaining the paper
Data Visualizations
Related Papers
Fu-Hsiang Chan, Yu-Ting Chen, Yu Xiang, Min Sun Anticipating accidents in dashcam videos Asian Conference on Computer Vision, 2016
Waqas Sultani, Chen Chen, Mubarak Shah Real-world Anomaly Detection in Surveillance Videos Preprint ArXiV, 2018
Other works on related to “Object Detection”
Shaoqing Ren, Kaiming He, Ross Girshick, Jian SunFaster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017
Tsung-Yi Lin, Piotr Dollar , Ross Girshick, Kaiming He , Bharath Hariharan, and Serge BelongieFeature pyramid networks for object detection IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2017
Contact
For questions/comments, contact any of us Tuan Nguyen Anh Jean Baptiste Lamare Ankit Shah