Multi-model cooperation based self organization multiple cameras system for robust moving object detection

A self-organization multi-model cameras system for robust moving object detection is introduced. The system has three modules: automatic registration, cooperative motion detection and camera parameters change detection. The registration module automatically register visual camera and IR camera based on co-motion statistics. The cooperative motion detection produces precise results bases on cooperative processing mechanism we proposed in. Here registration and detection modules are combined, and detection results are fed back to registration module for detecting the change of camera parameters. By this way, the system can be adaptive to environment change and camera parameter change. We demonstrate the performance of the method on visual/IR synchronous video sequences. Experimental results show that the multi-model camera system carries out precise and robust camera registration and moving object detection.

I. Introduction

Moving object detection is the foundation of many video processing tasks, such as target tracking, target recognition, video compression. While robust and precise motion detection is a difficult problem. As multi-camera system [1] [2] [3] becomes a hot issue of computer vision, multi-sensor information cooperative processing provides an effective way for precise detection.