A VIDEO-BASED VEHICLE DETECTION AND CLASSIFICATION SYSTEM USING CASCADE HAAR CLASSIFIER

Abstract

Vehicle classification data are imperative inputs for traffic operation, pavement design, and transportation planning. Automated traffic data collection and surveying can be a great tool in site selection, engineering, and more. A vigorous and real-time implementation is a challenge, however, especially in cases of high occlusion and a cluttered background. All these facts illustrate that vehicle data are extremely important for the rigorous analysis of traffic safety, traffic pollution, and flow characteristics. Unfortunately, most traffic sensors such as single-loop detectors presently in place cannot directly measure vehicle volumes. Though dualloop detectors can measure classified vehicle volumes, a few of them is used in the current transportation systems to meet the practical needs.
Seeing that traffic surveillance cameras have been increasingly deployed for monitoring traffic status on dominant roadways, effective utilization of these cameras for vehicle data collection is of practical significance. An important step towards attaining automated roadway monitoring capabilities is to detect vehicles in videos. The challenges lie in being able to reliably and rapidly detect multiple small objects of interest against a cluttered background which usually consists of trees and buildings, and then classify them as the light vehicle or heavy vehicle based on their  pixels length. To this end, we present a concept of traffic monitoring system. The proposed approach can disclose and classify vehicles using the uncelebrated video images. This prototype system is capable to use the uncelebrated surveillance cameras for real-time traffic data collection. The system also counts the number of cars passing in either direction in
each frame. The car detection and classification is done using a cascade of Haar features