基于计算机视觉的车辆闯红灯行为检测研究

Research on Detection of Red Light Violation Based on Computer Vision

为了实现对车辆闯红灯行为的自动智能和准确检测。本文提出了一种基于计算机视觉的车辆闯红灯行为检测方法。首先,采用高清摄像头实时采集道路上的车道、车辆和信号灯的视频图像;接着,对视频进行图像信息预处理,定制虚拟检测线、交通灯检测区域、车道检测区域,将定制结果保存在存储单元中;然后用计算机视觉技术来检测车辆闯红灯行为,通过HSV颜色空间中的H分量来识别信号灯的状态,利用MHI算法来识别车辆状态;最后,根据交通法规来识别车辆闯红灯行为。实验结果表明,本文提出的检测算法方法简单、识别率高、实现成本低并具有较高的鲁棒性。

To automatically and accurately detect red-light violation, this paper presents a method of detection of red Light violation based on computer vision. Firstly, the image of traffic intersection is got by a monocular high-definition camera in real time. In order to detect the valid area, the image is pre-processed with the customization of stop line, traffic light areas, lane areas, and the result of customization is saved. Then, the state of traffic signals is recognized with H in HSV color model and the vehicles are detected by MHI. Finally, according to traffic laws and regulations, vehicle red-light violation will be identified. Experiment results show that this method has lightly computational load, low-cost, high recognition rate and robustness features.