摘要
为了提高前方车辆检测的准确率和效率,提出了一种改进的多分辨率下的多特征提取的方向梯度直方图(HOG)特征融合算法。首先将样本扩缩为分辨率不同的图像,后转为YUV色彩空间,其次加权融合Y,U,V多通道方向梯度直方图(HOG)形成训练特征,最后采用支持向量机(SVM)对融合后的训练特征车辆分类器训练和检测。实验表明,该算法比传统HOG提取特征算法车辆检测率更高,效率高达98.92%,并且在不同天气状况下均有良好的检测效果和鲁棒性。
To improve the accuracy and efficiency of vehicle detection, a multi-feature extractive histogram of oriented gradient (HOG) fusion algorithm is proposed. Firstly, the sample pattern is expanded and narrowed to images with different resolutions and converted to YUV color space, followed by weighted fusion Y, U, V multi-channel histogram of oriented gradient (HOG) training features, Finally, support vector machine (SVM) is used to train and detect the trained features. The experiment show-s that this algorithm has a higher detection rate than the traditional HOG extraction feature algorithm, and the efficiency is as high as 98.92%, and have good detection effect and robustness under different weather conditions.
作者
马龙
刘胜
Ma Long;Liu Sheng(School of Mechanical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《计算机时代》
2018年第10期42-45,共4页
Computer Era