基于逻辑回归的车辆检测方法研究

Study on vehicle detection method based on logistic regression

针对高速公路追尾事故频发而导致自动驾驶系统需要提高车辆检测的实时性问题,本文提出了一种基于逻辑回归的车辆检测方法。首先,通过对包含车辆与非车辆图像的训练集提取HOG特征来训练逻辑回归分类模型以获得好的回归系数;然后,采用滑动窗口机制对截取的视频帧提取HOG特征并利用过训练好的逻辑回归模型进行检测,并结合非极大值抑制技术去除了多余的检测窗口;通过实验,在车辆图像的测试集上该模型的识别正确率达到了96.23%,在视频帧上的车辆检测效果显示该方法可满足实时性要求。

Frequent rear-end accidents on the highway makes the automatic driving system have to improve the real-time performanee of the vehiele deteetion, and a vehiele deteetion method based on Logistie Regression is proposed to overeome this problem in this paper. First of all, extraeting the HOG features of the dataset eonsist of vehiele and non-vehiele pietures to train the Logistie Regression Model （LRM） so as to obtain a good regression eoeffieient; then using the sliding window meehanism to extraet the HOG features of the intereept video frame and applying the well-trained LRM for deteetion, meanwhile, eombining with the non-maximal value suppression teehnique to remove the redundant deteetion window. Experiments on the vehiele image test dataset show that the vehiele identifieation aeeuraey of the LRM reaehes 96.23% and the performanee of vehiele deteetion on the video frame proves that the proposed method ean meet the real-time need.