Video analysis for traffic anomaly detection using support vector machines

In this paper we present a video-based traffic surveillance system which analyzes the video footage and uses the trajectories of the vehicles to detect any anomalous vehicle behavior at a traffic intersection. The trajectory analysis is done using support vector machines (SVMs). We also discuss the trajectory representation and trajectory filtering methods for increasing the accuracy of detection. To validate the proposed algorithms, we use data collected from a small scale testbed, which allows us to generate various training and testing data. This capability makes it possible to study how the different levels of variation in the training data impact the performance of the SVM classification.