Real-time Detection of Abnormal Vehicle Events with Multi-Feature over Highway Surveillance Video

This paper introduces a framework of real-time abnormal vehicle event detection with multi-feature over highway high-definition surveillance video. The framework is composed of two parts: multi-feature extraction and abnormity detection. In multi-feature extraction, a fast constrained Delaunay triangulation (CDT) algorithm based on constrained-edge priority is presented to instead of complicated segmentation algorithms. After calibrating manually to extract the actual driveways from surveillance video sequence, localizing vehicle regions and tracking via detection of vehicle regions to extract static features and motional features in monitor area, multi-feature vectors are created for each vehicle. In abnormity detection, a method of adaptive detection modeling of vehicle events (ADMVE) is introduced. A Semi-supervised Mixture of Gaussian Hidden Markov Model is trained with the multi-feature vectors for each video segment. The normal model is trained by supervised mode with manual labeling, and becomes more accurate via adaptation iteration. The abnormal models are trained through the adapted Bayesian learning with unsupervised mode. Finally, experiments using real video sequence are performed to verify the proposed method.