基于改进HOG特征和SVM分类器的行人检测

本文首先研究了行人检测的方法和面临的难点,然后根据高斯平滑滤波器和双线性插值法对传统方向梯度直方图做了改进,并搭建了支持向量机模型,从而构建了基于改进HOG特征和SVM分类器的行人检测系统。实验结果显示,在明亮且无遮挡的场景下,矩形框精确地定位行人,在光照不足或存在轻微遮挡时可以大体定位到行人,表明该系统在明亮无遮挡的情况下有准确的结果,并且在昏暗和轻度遮挡下检测效果良好,最后通过对比可以得出,本文提出的方法总体效果优于传统HOG特征和SVM分类器的方法。

Pedestrian Detection Based on Improved HOG Feature and SVM Classifier

This paper first studies the pedestrian detection method and the difficulties it faces,then improves the traditional direction gradient histogram according to Gaussian smoothing filter and bilinear interpolation method,and builds the support vector machine model to construct the improved HOG feature.And the pedestrian detection system of the SVM classifier.The experimental results show that in a bright and unobstructed scene,the rectangular frame accurately locates the pedestrian,and when the brightness is not bright enough or there is slight occlusion,the pedestrian can be generally positioned,indicating that the system has accurate results in the case of bright unobstructed,and The detection effect is good under dim and light occlusion.It can be concluded that the overall effect of the proposed method is better than the traditional HOG feature and SVM classifier.