As the first step of vehicle detection and recognition system, how to quickly and accurately detect the vehicle in a picture is directly related to the subsequent vehicle application research. In order to improve the processing speed of vehicle detection, reduce the false alarm rate of detection, and get better results, the method is applied in real scene, this paper carried out in-depth research on this. Collect traffic and urban road surveillance videos as experimental data, of which 2000 were positive samples and 2000 were negative samples. Firstly, a vehicle image preprocessing is carried out on the collected experimental data, and the image feature is extracted based on gray image and improved AdaBoost algorithm, and then the image enhancement is realized by using multi-scale Retinex. Using this method, we can make the image processing accord with the nonlinear characteristics of the human eye to the brightness response, and avoid the distortion of the image directly processed by Fourier transform. In order to improve AdaBoost classifier, it is necessary to use local binary edge features and train the collected feature samples. In order to highlight the vehicle target and ignore the background, we need to use a selective graying way, which is based on the H component of HSV space. The experimental results show that the accuracy of AdaBoost classifier reaches 85.8%, the recall rate is 80.9%, and the comprehensive performance is very high, which can meet the performance requirements.

作为车辆检测识别系统的第一步，如何快速准确地检测出图片中的车辆，直接关系到后续的车辆应用研究。为了提高车辆检测的处理速度，降低检测误报率，获得更好的结果，将该方法应用于实际场景，本文对此进行了深入研究。收集交通和城市道路监控视频作为实验数据，其中2000个为正样本，2000个为负样本。首先对采集到的实验数据进行车辆图像预处理，基于灰度图像和改进的AdaBoost算法提取图像特征，然后利用多尺度Retinex实现图像增强。使用这种方法，我们可以使图像处理符合人眼对亮度响应的非线性特性，避免傅里叶变换直接处理的图像失真。为了改进 AdaBoost 分类器，需要使用局部二元边缘特征并对收集到的特征样本进行训练。为了突出车辆目标而忽略背景，我们需要使用一种基于HSV空间的H分量的选择性灰度化方式。实验结果表明，AdaBoost分类器的准确率达到85.8%，召回率为80.9%，综合性能非常高，可以满足性能要求。为了改进 AdaBoost 分类器，需要使用局部二元边缘特征并对收集到的特征样本进行训练。为了突出车辆目标而忽略背景，我们需要使用一种基于HSV空间的H分量的选择性灰度化方式。实验结果表明，AdaBoost分类器的准确率达到85.8%，召回率为80.9%，综合性能非常高，可以满足性能要求。为了改进 AdaBoost 分类器，需要使用局部二元边缘特征并对收集到的特征样本进行训练。为了突出车辆目标而忽略背景，我们需要使用一种基于HSV空间的H分量的选择性灰度化方式。实验结果表明，AdaBoost分类器的准确率达到85.8%，召回率为80.9%，综合性能非常高，可以满足性能要求。