改进的视频车辆背景差分检测

The Video Vehicle Detection Based on Improved Background Difference Method

实时交通信息检测在智能交通系统中起着重要的作用,视频车辆检测是交通信息检测的一种重要手段。背景差分算法因其灵活性和准确性,成为基于视频的运动目标实时检测的一种常用方法,传统背景差分算法仅仅强调对二维图像的处理,尤其强调对图像分割和目标跟踪。该文对传统背景差分算法进行了改进,提出一种基于虚拟线框的车辆视频检测算法。该算法核心思想通过在每个车道设置两个虚拟线框来检测交通流参数,虚拟线框的输出信号源于背景差分。该方法只需对虚拟线框内的图像区域进行处理,从而并且避开了在视频图像中进行复杂的车辆特征提取与跟踪,减少了运算量,降低了运算负荷。经测试算法的处理速度为25帧/秒,车辆识别精度约为88%。

The collection of real-time traffic data plays a critical role in the intelligent transport system, and video-based detectionis an important part in traveler information systems. Background difference method has become common means in real-time mo-tion detection because of its flexibility and veracity. Traditional background difference method is mainly based on 2-dimensionalimage processing, especially on image division and vehicle tracking. Improved the traditional background difference method, an al-gorithm of vehicle detection which is based on virtual frame is proposed in the paper. The main idea of this algorithm is based onthe lane, each lane can have two virtual-frame to detect its traffic parameters. Each virtual-frame's output signals mainly derivefrom the background difference. This method is only processing small area within virtual-frame and avoiding vehicle tracking in 2-dimensional image, hence the time cost of calculation and the computation burthen is reduced. The experiment tells us that thespeed of the algorithm is 25 fps, the precision is about 88%.