Accident Recognition via 3D CNNs for Automated Traffic Monitoring in Smart Cities
Source of Publication
Advances in Intelligent Systems and Computing
© 2020, Springer Nature Switzerland AG. Automatic recognition of road accidents in traffic videos can improve road safety. Smart cities can deploy accident recognition systems to promote urban traffic safety and efficiency. This work reviews existing approaches for automatic accident detection and highlights a number of challenges that make accident detection a difficult task. Furthermore, we propose to implement a 3D Convolutional Neural Network (CNN) based accident detection system. We customize a video game to generate road traffic video data in a variety of weather and lighting conditions. The generated data is preprocessed using optical flow method and injected with noise to focus only on motion and introduce further variations in the data, respectively. The resulting data is used to train the model, which was then tested on real-life traffic videos from YouTube. The experiments demonstrate that the performance of the proposed algorithm is comparable to that of the existing models, but unlike them, it is not dependent on a large volume of real-life video data for training and does not require manual tuning of any thresholds.
3D convolutional neural networks, Accident recognition, Computer vision, Deep learning, Machine learning
Bortnikov, Mikhail; Khan, Adil; Khattak, Asad Masood; and Ahmad, Muhammad, "Accident Recognition via 3D CNNs for Automated Traffic Monitoring in Smart Cities" (2020). All Works. 329.
Indexed in Scopus