Performance analysis of a surveillance system to detect and track vehicles using Haar cascaded classifiers and optical flow method

Abstract

This paper presents the real time vehicle detection and tracking system, based on data, collected from a single camera. In this system, vehicles are detected by using Haar Feature-based Cascaded Classifier on static images, extracted from the video file. The advantage of this classifier is that, it uses floating numbers in computations and hence, 20% more accuracy can be achieved in comparison to other classifiers and features of classifiers such as LBP (Local Binary Pattern). Tracking of the vehicles is carried out using Lucas-Kanade and Horn Schunk Optical Flow method because it performs better than other methods such as Morphological and Correlation Transformations. The proposed system consists of vehicle detection and tracking; and it is evaluated by using real data, collected from the route networks of Chittagong City of Bangladesh.