MP-RAD Dataset for the paper, "Detection of Road Accidents using Synthetically Generated Multi-Perspective Accident Videos", published in IEEE Transactions on Intelligent Transportation Systems, 2022. Paper LINK
Road accidents are often caused by short abnormal events, including traffic violations, abrupt changes in vehicular motion, driver fatigue, etc. Observing an accident event from the proper camera perspective plays a crucial role in detecting accidents. However, capturing such abnormal events from a limited camera perspective may not be possible. We present a deep learning framework to analyze the accident events recorded from multiple perspectives. First, we estimate feature similarity in videos recorded from multiple perspectives. We then divided the video samples into high and low-feature similarity groups. Next, we extract spatio-temporal features from each group using two-branch DCNNs and fuse them using a rank-based weighted average pooling strategy followed by classification. We present a new road accident video dataset (MP-RAD), where each accident event is synthetically generated and captured from five independent camera perspectives using a computer gaming platform.
Video Samples: Contain a few video samples from MP-RAD.visil/: Implementation of ViSiL: Fine-grained spatio-temporal video similarity learning network.Feature_Extraction_WA_Pooling.py: Extract features, weighted average poolingvisil.ipynb: Notebook for generating feature similarity-based confusion matrix
- Place a single accident video (all five angles) into the query.txt file
- Obtain the feature similarity-based confusion matrix using
visil.ipynb - Generate rank sequence as per the sorted values
- Set the
kvalue inFeature_Extraction_WA_Pooling.py - Extract and save video features
- Construct a classifier network according to the description provided in the Classifier section or see
Classifier.py - Train the network with the provided setting
The final classifier model is a 3-layer MLP, where the number of units is 512, 32, and 1, respectively, regularized by dropout with a probability of 0.6 at the last layer. We have trained the model with a learning rate of 0.01 for 350 iterations using the Adagrad optimizer.
If you find our code, dataset, or any related artefacts useful in your research, please consider citing our work MP-RAD IEEE ITS 2022.