Nghiên cứu ứng dụng kỹ thuật theo dõi đối tượng xây dựng hệ thống camera giám sát thông minh

Abstract

This paper presents some results of building intelligent surveillance camera systems using object tracking. Main steps of the object tracker include (i) keypoint tracking using optical flow, (ii) keypoint matching, and (iii) consensus-base voting. A novel algorithm to accelerate processing using pipeline technique on multicores systems has also been proposed. The algorithm divides the whole processing frame into 4 stages which are executed on 4 different threads. Synchronization of threads is realized producer – consumer model. The proposed method achieved a 3.3 times increased computational time compared to the original one. The surveillance system continuously tracks target object and gives a warning sound if the object disappears in a predefined interval. Experimental results show that the proposed method achieves very promising results.