中文摘要：

为提高夜间行车的安全性,提出了一种基于单目视觉的夜间前方车辆检测方法.该方法利用最大类间方差法自动确定阈值,从背景中抽取出前车尾灯,并根据HSV颜色空间的颜色阈值剔除非尾灯目标,利用Kalman滤波方法将图像分为跟踪区域和检测区域,在两个区域内分别进行尾灯配对,根据尾灯对之间特征相似性的比较,剔除误检的车辆;跟踪区域中漏检的车辆,根据前一帧检测的车辆位置和正确抽取的尾灯来估计,以实现车辆检测.实验结果表明,该算法能准确检测夜间前方车辆,有效降低漏检率和误检率.

英文摘要：

To enhance night driving safety,a preceding vehicle detection algorithm for night time is proposed based on monocular vision.With the method,the taillight of preceding vehicle is extracted from background by automatic image thresholding based on Otsu algorithm,and the non-taillight objects are removed by the color threshold in HSV color space.The image region of interest is divided into tracking region and detecting region by using Kalman filtering.The taillight pairing algorithm was implemented in the two regions respectively and the falsely detected vehicles are discarded by comparing the similarity of taillight pair features.The missed vehicle in tracking region is estimated according to the vehicle position in preceding frame and one correct extracted taillight to complete vehicle detection.The experimental results show that the algorithm can accurately detect preceding vehicles at night with both false and miss detection rates effectively reduced.