基于HOG特征和SVM的前向车辆识别方法

HOG-Feature and SVM Based Method for Forward Vehicle Recognition

为了解决汽车安全驾驶辅助系统中的前向车辆实时识别问题,提出了一种基于梯度方向直方图特征和支持向量机的前向车辆识别方法。通过分割提取车辆底部阴影特征生成假设区域,采用基于直方图分析的方法实现车辆底部阴影的准确分割,综合分析车底阴影的水平边缘特征和垂直边缘特征完成假设区域的生成;使用基于梯度方向直方图特征和支持向量机得到的车辆分类器对获得的车辆假设区域进行验证,剔除了假设区域中的非车辆区域。利用采集的道路视频对提出的方法进行了车辆识别实验,结果表明,该方法能够在不同光照条件下自适应地进行实时车辆识别,其中在正常光照下的识别率为96.52%,误识别率为3.59%。

A HOG-feature and SVM based method was proposed for real-time forward vehicle recognition in automotive safety driver assistant systems. The shadow underneath vehicle was segmented accurately by using histogram analysis method and the initial candidates were generated by combining horizontal and vertical edge feature of shadow. These ini- tial candidates were further verified by using a vehicle classifier based on the histogram of gradient and support vector machine. The experimental results show that the proposed method could adapt to different light conditions robustly. Specially, the proposed method has a recognition rate of 96.52 percent and a false rate of 3. 59 percent in normal light condition.