全天候车辆实时检测系统

All-Weather Vehicle Real-Time Detection System

中文关键词:  车辆检测  混合高斯背景建模  直方图双峰阈值法  分水岭分割法

英文关键词:vehicle detection  Gaussian mixture background modeling  histogram bimodal threshold method  watershed segmentatio method

中文摘要:

      为解决全天候车辆检测技术精度不高的难题，提出了一种新型联合检测算法。通过在混合高斯背景建模法中加入膨胀、腐蚀算法，减少了视频车辆图像中的大量断点和随机噪声；利用最小面积法设置外接矩形框并通过不断调整矩形框的大小和位置，准确框选出视频图像中的目标车辆。由于夜间光线场景复杂，以能被明显区分的车前灯作为目标，通过直方图双峰阈值法和分水岭分割法将其从视频图像的背景中分离，并基于车前灯的形态参数，设置合适的车灯匹配条件，提出一种实时的车灯配对、车辆跟踪算法，大幅提高了夜间车辆检测的准确度。提出的新型联合车辆检测算法，有效地消除了由于外界光线环境改变造成的车辆检测误差，为全天候车辆检测提供了新思路，具有很强的针对性和重要的现实意义。

英文摘要:

      In order to solve the problem of low accuracy of all-weather vehicle detection technology,a new joint detection algorithm is proposed.By adding dilation and erosion algorithms to the Gaussian mixture background modeling,a large number of breakpoints and random noise in the video vehicle image are reduced.The circumscribed rectangular frame is set by using the minimum area method,and the size and position of the rectangular frame is continuously adjusted,the target vehicle in the video image is accurately selected.Due to the complexity of the nighttime light scene,the clearly distinguishable headlights are taken as the target,and they are separated from the background of the video image by the histogram bimodal threshold method and the watershed segmentation method,and appropriate matching conditions of the lights are set based on the morphological parameters of the headlight,and a real-time vehicle pairing and vehicle tracking algorithm is proposed,which greatly improves the accuracy of vehicle detection at night.The proposed new joint vehicle detection algorithm effectively eliminates vehicle detection errors caused by changes in the external light environment and provides new ideas for all-weather vehicle detection.It has strong pertinence and important practical significance.