Vision-based activities recognition by trajectory analysis for parking lot surveillance

This paper presents a novel event recognition framework in video surveillance system, particularly for parking lot environment. The proposed video surveillance system employs the adaptive Gaussian Mixture Model (GMM) and connected component analysis for background modeling and objects tracking. Spatial-temporal information of motion trajectories are extracted from video samples of known events to form representative feature vectors for event recognition purposes. An event is represented by feature vector that contains dynamic information of the motion trajectory and the contextual information of the tracked object. The event classification is accomplished by measuring the similarity of the extracted feature vector to the labeled definition of known events and analyzing the contextual information of the detected event. Experiments have been carried out on the live video stream captured by the outdoor camera, and the results have demonstrated great accuracy of the proposed event recognition algorithm.