Compressed sensing based object detection and tracking system using measurement selection process for wireless visual sensor networks

Abstract

Video surveillance applications in wireless visual sensor networks (WVSN) attract a lot of attention in recent years which demands higher performance with less complexity. Efficient and simple moving object detection and tracking (MODT) system is presented in this paper targetting the video surveillance application in WVSN. The main contribution of this paper is to develop a system that can perform both object detection and tracking with less complexity. This MODT system adopts compressed sensing (CS) to perform the background subtraction on compressive measurements and the subtracted measurements undergoes a measurement selection process (MSP) to extract the foreground measurements. MSP aims at extracting the minimum number of measurements that can yield higher detection accuracy. The detected object is tracked using the kalman filtering approach. Initially the centroid of the object is extracted from the binary image with the help of contour tracing which is then given as input to the Kalman filter to track the objects in the video. The performance of the MODT system is evaluated using parameters such as percentage of reduction in samples, energy complexity, detection and tracking accuracy. From the results it is evident that MODT system can achieve higher detection and tracking accuracy by reducing the measurements to more than 83 %. Also, the significance of the extracted measurements is observed by analyzing the detection accuracy which is around 0.88.