Path Normalization for Traffic Surveillance Video Retrieval

In this paper we present a novel method based on path normalization for classification in the traffic surveillance videos. Extracting the low level feature vectors in various sizes and recording all spatial-temporal information without fixed sampling rate are the main reason in this normalization. The normalized feature vectors are used for unsupervised learning and since most people of society have legal traffic behaviors system can extract the necessary knowledge automatically to detect illegal behavior. In the proposed structure, decision making for these behaviors is based on spatial-temporal features. The experimental results show high accuracy in trajectories classification using path normalization.