A Noval System Architecture for Multi Object Tracking Using Multiple Overlapping and Non-Overlapping Cameras

摘要

Typically, visual surveillance systems contain multiple cameras with overlapping and non-overlapping views used to improve an accuracy and the area coverage of the surveillance systems. The real-time and continuous capture of video data from cameras requires automatic analysis, object detection and tracking of objects before any malicious activity or event analysis can be performed on the video data. The existing systems rely on back- end database and servers to process video data from multiple cameras to track the objects. However, this paper presents a novel system architecture that allows peer-to-peer communication between the multiple cameras. Each camera is capable of tracking the object individually and equipped with processor, memory and communication medium. In addition, each camera only exchanges a small amount of data for consistent labelling of objects across the multiple cameras in real-time. This research work also presents a survey of multi-camera object/person tracking system. The goals of this research work are three-fold: i) serve as a guideline for researchers who are new to image/video processing and want to contribute to this research area, ii) provides a novel system architecture for consistently tracking of objects in multi-camera video surveillance systems, and iii) provides further research directions required into accuracy and quality-of-service assurance of video surveillance systems.