Abstract: Aiming at the problem of night vehicle video detection and vehicle traffic statistics, an improved algorithm based on video image processing to extract headlights is proposed. By analyzing the characteristics of night vehicle video, the gradient filtering method is used to eliminate the interference of the ground reflected light to the lamp, and the image enhancement is realized. The watershed segmentation algorithm and the histogram double peak method are combined to extract the information of the car headlights. By using the matching principle, a new matching algorithm and tracking algorithm are designed to realize the matching and tracking of the lights. Finally, the statistical functions of vehicle detection and traffic flow are realized.

Key words: vehicle detection, gradient filtering, watershed segmentation, lamp pairing, headlight tracking

摘要： 针对夜间车辆视频检测和车流量统计的难题，提出了一种改进的基于视频图像处理提取车前灯的算法。通过分析夜间车辆视频的特点，利用梯度滤波法消除地面反射光对车灯的干扰，实现图像增强，并将分水岭分割算法和直方图双峰法相结合提取车前灯的信息。利用车灯配对匹配原则设计了一种新的匹配算法和跟踪算法实现车灯的配对与跟踪，最终准确地实现了车辆检测和车流量的统计功能。

关键词: 车辆检测, 梯度滤波, 分水岭分割, 车灯配对, 车灯跟踪