Human visual tracking in surveillance video

Abstract

Visual surveillance in dynamic scenes, especially for human activities, is one of the current challenging research topics in computer vision. It is a key technology to fight against terrorism and crime to ensure public safety. The motivation of this thesis is to design an efficient human visual tracking system for video surveillance deployed in complex environments.

In video surveillance, detection of moving objects is the first step to analyze the video streams. And motion segmentation is one of popular approaches to do it. In this thesis, we propose a motion segmentation method to overcome the problem of motion blurring.

The task of human tracking is key to the effective use of more advanced technologies, like activity recognition and behavior understanding. However, human tracking routines often fail either due to human's arbitrary movements or occlusions by other objects. To overcome human's arbitrary movement, we propose a new Silhouette Chain Shift model for human detection and tracking. To track human under occlusions, firstly each frame is represented by a scene energy which consists of all the moving objects. Then the process of tracking is converted to a process of minimizing the proposed scene energy.

Findings from the thesis contribute to improve the performance of human visual tracking system and therefore improve security in areas under surveillance.