Car tracking in video surveillance

Abstract

Technology has reached a stage where mounting cameras to capture video imagery is quite affordable, but finding available human resources to sit and watch the imagery is very expensive, due to the continuous monitoring that is required. The massive amount of data involved makes it unfeasible to guarantee attentive monitoring by human operators for long periods of time due to monotony and fatigue. Therefore, to assist human operators with identification of important events in videos, an intelligent visual surveillance system can be used. The goal of such implementations is not just to use cameras in the place of human eyes, but mainly to achieve the entire surveillance task as automatically as possible. There are three basic key steps in object tracking, which are detection of interesting moving objects, tracking of such objects from frame to frame, and the analysis of object tracks to so that their behaviour can be recognized. In this project, a tracking algorithm is designed to identify moving cars from a video sequence which is captured by a stationary camera. The algorithm is divided into two main parts, motion detection and tracking. Initially, the system detects any motion present in the scene by using the background subtraction technique. Finally, it keeps track of each moving car and associates the detections which correspond to the same car over a period of time together, and displays a car identification number for each track. The system was tested with different test sequences, and values for the thresholds used in the algorithm were selected experimentally. Tests showed that the algorithm can track cars from a stationary camera effectively.