A Novel Video Processing Algorithm for Accurately Vehicle Detection and Tracking

Abstract

In this paper, a modified vehicle detection algorithm has been presented according to three based frame difference background subtraction followed by Kalman Filter (KF) to track moving vehicles. The presented vehicles detection and recognition methodology addresses the following issues: obtaining accurate vehicles detection with appropriate frame noise restoration, solving incomplete detection problems for multiple moving vehicles, increasing recognition performance, and reducing calculation time. Furthermore, the trajectory of each moving vehicle in the successive video frames has been measured and allocated. The proposed methodology has been tested on videos which are discriminated with respect to lighting conditions, camera resolution, and vehicle density. The average absolute error between predicted and detected vehicles has been utilized in order to estimate both detection and tracking accuracy of interest. Experimental results for multiple moving vehicle detection and tracking in three separate surveillance videos have achieved high accuracy (95.51%, 94.714%, and 95.719%) for test videos 1, 2, and 3 respectively.