An advanced on-line video tracking system

Abstract

This paper describes an on-line visual sensing system available to a security area. The sensing system capable of tracking and zooming functions detects an unexpected object and records the series of available extracted object images in the time varying environment. Two algorithms-image processing to detect the object motion and parallel pan/tilt/zoom control-are proposed for the sensing system. In the image processing, the object edge image is first abstracted by specific temporal differentiation of two successive edge images. The object size as a major information in zoom-lens control is determined by multi-stage eigenvalues of the real symmetric matrix in the abstracted image which is designed for low computational cost as well as noise detection. Secondly for a practical application, the organic path planning of parallel sensor control-three mechanical freedoms of pan, tilt and zoom-is established, and pipeline processing between image processing and sensor control is designed. Experimental results show that the proposed system is valid for numerous kinds of moving objects in the security zone.<>