一种基于多特征和机器学习的分级行人检测方法

针对单幅图像中的行人检测问题,提出了基于自适应增强算法(Adaboost)和支持向量机(Support vector machine, SVM)的两级检测方法, 应用粗细结合的思想有效提高检测的精度.粗级行人检测器通过提取四方向特征(Four direction features, FDF)和GAB (Gentle Adaboost)级联训练得到,精密级行人检测器用熵梯度直方图(Entropy-histograms of oriented gradients, EHOG)作为特征, 通过支持向量机学习得到.本文提出的EHOG特征考虑到熵, 通过分布的混乱程度描述,具有分辨行人和类似人的物体能力. 实验结果表明,本文提出的EHOG、粗细结合的两级检测方法能准确地检测出复杂背景下不同姿势的直立行人, 检测精度优于以往Adaboost方法.

Two-stage Pedestrian Detection Based on Multiple Features and Machine Learning

A two-stage detection method based on Adaboost and support vector machine (SVM) is proposed for the pedestrian detection problem in a single image, which uses the combination of coarse level and fine level detection to improve the accuracy of the detector. The coarse level pedestrian detector makes use of the four direction features (FDF) and the gentle Adaboost (GAB) cascade training; the fine level pedestrian detector uses entropy-histograms of oriented gradients (EHOG) as features and the SVM as classifier. The proposed EHOG features considering entropy and the distribution of chaos have the ability to distinguish between the pedestrians and the objects similar to people. Experimental results show that the proposed two-stage pedestrian detection method with the combination of the coarse-fine level and EHOG feature can accurately detect upright bodies with different postures in the complex background, at the same time the precision is better than the classic Adaboost methods.