多尺度级联行人检测算法的研究与实现

Research and Realization of Pedestrian Detection Algorithm by Multi-scale Cascaded Features

行人检测算法是利用行人的特征结合分类器对图片中是否有行人进行判断的方法。文中基于传统的HOG行人特征检测方法以及Adaboost分类器思想,改进了行人检测算法。使用多尺度的HOG特征对图片的检测区域进行特征提取,并采用级联的Adaboost分类器结合对应尺度的特征进行分类判断,将判断结果输入下一级分类器中继续进行分类判断,最终实现区域内有无人的检测。实验结果表明多尺度下的级联分类器能够更加有效地筛选出行人区域,在计算时间小幅增加的情况下,很大地提高了检测精度。

Pedestrian detection algorithm is a method to decide whether there exists pedestrian or not in the picture by characters of pedes-trians combined with classifiers. In this paper,an improved pedestrian detection algorithm is advanced based on the traditional HOG method and Adaboost classifying idea. In this algorithm,use the multi-scale HOG features to extract the features in the detection region,then the cascaded Adaboost classifiers with multi-scale features are used to judge,getting its result input into the next level of classifier,and finally achieve the goal of pedestrian detection. The experimental result shows that the cascaded classifiers using multi-scale features performs better in pedestrian detection,which achieves higher detection precision at little cost of computing time.