融合边缘和角点特征的实时车辆检测技术

Real-time Vehicle Detection Using Edge and Corner Feature

采用一种快速角点提取方法来提取角点特征,克服阴影干扰,并与边缘特征融合实现车辆检测.提出一种加入噪声自适应求取边缘二值化阈值的方法,有效抑止不同光照条件、路面强噪声,实现强边缘提取.利用车牌和通风栅格、车辆轮廓特征,解决拥堵检测问题.根据前后帧图像特征预估车速,极大减少了误触发.采用SVM进行有、无车分类.给出夜晚检测方法.实验表明本文算法较好的解决了车辆检测,特别是拥堵时的车辆检测.

A method combining features of corner and edge to detect vehicle is proposed. The detector of corner is a high-speed feature detector which can eliminate the disturbance of shadow. A new adaptive threshold obtained through adding noise into histogram of grayed edge map is to binarize edges. The new threshold extracts stronger edges and eliminates larger noise on different light condition and road surface effectively. Integrating vehicle＇s plate, radiator grill and outline, the proposed method to detect vehicles in congested condition is more accurate. Using the characteristics of neighboring frames to estimate the speed of a vehicle, the error rate of detection is decreased. SVM is used to class image frame into two classes. The paper also gives how to detect vehicle at night. Experimental results are given to demonstrate that the proposed techniques are effective and efficient for vehicle detection especially on congested road.