Video surveillance system for tracking human activity

Abstract

Human activity tracking is a very useful and powerful function in video surveillance system. It is also a very important research area in video processing. The objectives of this project are to study the existing tracking algorithm and implement the kernel-based mean shift tracking algorithm. The advantages and disadvantages of the algorithms are discussed. Possible improvements for mean sift algorithm are also given. The proposed system was implemented in Matlab and Java. Both systems produce good recognition result when being tested on different data set. Although the Matlab version performed better, the Java version was implemented with a nice user interface, and the corresponding histogram of the image was shown in the system. Finally the results generated by the system are presented and discussed. In addition the suggestions to improve the system are also given.