Real-time Object Detection and Tracking for Video Surveillance

Abstract

In view of application in smart visual surveillance systems this paper presents a new method of object detection and tracking in a surveillance scenario. The paper proposes robust object detection and tracking mechanism in which the background subtraction uses parallel processed Kernel Density Estimation (KDE) and the object tracking uses spatial color models of the detected object for tracking. The models of newly detected objects are stored in a reference list. Models of object detected in the next frame are compared with the reference models to track the object in the new frame. The spatial dimension introduced make the algorithm performs very well and fast. The system has been implemented both in indoor and outdoor environments and was found not only functionally okay but very computationally efficient in terms of processing time. The proposed system also is capable of detecting collision and object merging. One major areas the system contributes is the fast processing which is required by surveillance systems.