Anomaly Detection on Video by Detecting and Tracking Feature Points

Abstract

There is a well-known problem of video sequence analysis when it is necessary to identify and localize areas of abnormal movement of objects. This is necessary to attract the attention of the operator in the process of work or when analyzing the archive of video recordings. One of the solutions is based on tracklet analysis using short segments of the object’s trajectory that characterize its movement over a certain period of time, and then analysis of activity in various areas of the frame. Since the construction of the tracklet and trajectory is based on the optical flow, the quality and performance of the algorithm significantly depend on the choice and configuration of methods for detecting and tracking feature points. We have analyzed various combinations of these methods using the examples of test videos of normal and abnormal activity in a pedestrian zone. The necessity of a preliminary analysis of the methods used when setting up a video surveillance system to solve specific tasks is shown. Suitable combinations of methods are proposed.

Keywords

• anomaly detection
• tracklet analysis
• feature points detection
• feature points tracking
• computer vision