基于图像处理的车辆识别系统设计

为了实时了解道路交通信息，及时处理交通事故，在一定程度上缓解交通事故频发的状况，设计了基于方向梯度直方图（HOG）特征与支持向量机（SVM）的车辆检测系统.在收集大量的车辆样本，构建正、负样本集之后，提取出每一个正、负样本的HOG特征向量并汇总，进而形成检测所用的SVM分类器模板.在视频检测过程中，提取视频中每帧图像的HOG特征送到训练好的分类器中与模板进行对比，并用矩形框标注检测出的车辆目标.利用实际道路监控视频进行车辆检测系统测试，结果表明，对于不同的路况、天气和光线下的道路环境，该算法都可以完成实时且准确的检测，有较强的实际场景应用能力.

Vehicle Detection System Based on Image Processing

To obtain traffic information and dispose of traffic accidents in time, and ultimately reduce accidents frequency to a certain extent, a vehicle detection system based on HOG and SVM was designed. Firstly, a large number of vehicle samples were collected to build positive and negative sample sets, and then HOG feature vectors of all these samples were extracted and summarized. Finally, the SVM classifier template was formed. In the video detection process, the HOG features of each frame in the video were extracted and compared with the template that has been trained, and the detected vehicle targets were tagged with a rectangular frame. The actual road monitoring video was adopted to test the vehicle detection system, and the results demonstrated that the algorithm designed shows real-time performance, accuracy, and applicability for different road conditions, different weather and different illumination.