基于公路监控视频的车辆检测和识别

Vehicle detection and recognition based on highway monitoring video

针对目前基于公路监控视频的运动车辆检测和分类存在检测速度慢、分类效果差等问题,提出了一种高斯混合模型和AlexNet结合的检测分类算法。该方法首先用高斯混合模型对场景的背景建模,用当前帧图像减去背景图像得到运动的车辆,然后用AlexNet对已检测到的车辆进行分类。针对自采数据集过小的问题,采用数据扩充策略来扩充训练数据。实验结果表明,该方法检测速度可达到45 f/s,车辆检测精确度为94. 4%,召回率为88. 6%,均优于主流检测方法。

There are several problems in motion vehicle detection and classification based on highway surveillance video,such as slow detection speed and poor classification effect.To solve these problems,a detection classification algorithm combining Gaussian Mixture Model and AlexNet is proposed.Firstly, this method uses the Gaussian Mixture Model to build the background model of the scene,then subtracts the background model from the current frame image to get the moving vehicle.After that this method uses AlexNet to classify the detected vehicles.In response to the problem of too small data set,a data expansion strategy is used to enrich the training data.The experimental results show that the detection speed of the method can reach 45 f/s,the vehicle detection accuracy is 94.4%,and the recall rate is 88.6%,which is better than the mainstream detection methods.