Xiaojing Yuan - Surveillance System, UNR

A Distributed Surveillance System

Xiaojing Yuan and Zehang Sun
Department of Computer Science, UNR

The Project:
Main
Overview
Methodology
Results
Future work
Publications
Acknowledgement

We have developed a distributed visual surveillance system, and the system demonstrates good performance under unconstrained outdoor environment. The basic structure of the system is a LAN, with server sitting in the security office and Camera Units distributed in different observation areas. The current system detects Vehicle/Pedestrian in the monitoring area. It can also be trained to detect other type of intrusions easily, because it is a trainable system. We have defined our own protocol to make the communication more efficient. The intrusion detection program consists of two steps: background subtraction based hypothesis generation and appearance-based hypothesis verification. The two-step strategy mitigates the difficulties faced by all background subtraction methods to appearance-based hypothesis generation step. In the hypothesis generation step, powerful pattern classification approach, utilizing redundant statistical Gabor filter features and SVMs, can screen out the false hypotheses very easily.

For future work, we plan to investigate more extensively the problem of information fusion from multiple cameras, lower the detection error, especially the FP using bootstrapping method, and upgrade the system with the capability of human activity recognition.