Vehicle Tracking Based On Computer Vision

Target tracking technology has been proposed for several decades, after long time of research and development, it has become a very important technology in today’s society, also it plays a huge role in daily life and military application fields. With the continuous development of computer vision technology, the detection and tracking of vehicle based on computer vision is an active research area in the intelligent transportation systems, it is the fundamental of traffic statistics and behavior analysis. The research of the detection and tracking of multiple targets in the dynamic backgrounds remains open, many scientific research institutions and colleges invested a lot of time and energy for the research of this field. This paper mainly studies the detection and tracking of moving targets in the traffic scene, which includes foreground segmentation, vehicle detection, Kalman filter and multi-target data association. This technology has a broad prospect of application in the intelligent transportation systems.The research content of this article is as follows:1、Motion vehicle detection:First of all, it introduces several kinds of commonly used foreground segmentation algorithm, this paper uses a dynamic background modeling technique based on the simplified KDE (Kernel Density Estimation). It can effectively segment moving foreground from the complex backgrounds. The detection of moving vehicles is accurate and effective after noise reduction and connected domain labeling, which makes a good preparation for target tracking.2、Vehicle tracking:The study of single target tracking based on Kalman filter is investigated. This article proposes a feature extraction technique based on rgI color histogram, which outperforms HSV (Hue, Saturation and Value) spatial moments with respect to the locating of tracking. It can achieve target tracking in the complex environment. The experimental results of tracking using various features verify the superiority of rgI color histogram, it can serve as a very effective features for the description of moving targets.3、Multi-target data association:The extraction of global features for the detected multiple targets and the data association based on feature matching are investigated. A similarity function is established and an optimization is performed on the multiple measurement values. The value with the highest probability is selected as the matching result after compared with threshold. The multi-target data association is essential for the multi-target tracking.The experimental results indicate that, the proposed tracking algorithm is real-time and robust, and can accurately tracking multiple vehicles in real traffic scenes.