基于卷积神经网络预处理的Hog特征车辆检测算法

Vehicle Detection Algorithm Based on Convolutional Neural Network Preprocessing and Histogram of Oriented Gradient

传统方法中,通过计算图像Hog特征,构建支持向量机进行分类,对车辆全身的检测率欠佳。提出一种基于卷积神经网络预处理的Hog特征算法进行车辆全身检测,基于以上方法,利用卷积神经网络自主学习特点,对待检测图像进行预处理,提取出有效的边缘特征图。结果表明,该方法识别率优于传统方法以及普通CNN架构。

In traditional method, through calculating the image characteristics of Hog, and constructing the support vector machine to recognize vehicle, the detection rate of the method, aiming to the whole body of the car, is not good. Proposes an algorithm based on convolutional neuralnetwork to preprocess the image, based on the above method, uses the self-learning characteristics of the convolutional neural network, topreprocess the undetected image and get the effective edge feature map. The result shows that the detection rate of this method is betterthan the traditional method and the ordinary convolutional neural network architecture.