一种深度图像特征的建筑物内行人检测方法  

Pedestrian Detection Method in Buildings Based on Depth Image Feature

作　　者：刘剑[1] 徐萌 赵悦[1] 张锐[1] 高恩阳[1] LIU Jian, XU Meng, ZHAO Yue, ZHANG Rui, GAO En-yang(Faculty of Information and Control Engineering, Shenyang Jianzhu University, Shenyang 110168, China)

机构地区：[1]沈阳建筑大学信息与控制工程学院,沈阳110168

出　　处：《小型微型计算机系统》2018年第4期852-858,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61272253)资助;辽宁省自然科学基金项目(201602616)资助;辽宁省教育厅科学研究项目(L2015443)资助;住建部项目(2015-K2-015)资助

摘　　要：针对传统建筑物内行人检测算法复杂背景遮挡、光照影响等导致的检测不准确、效率低等问题,提出一种基于深度差值及方向梯度特征的行人检测算法.利用Kinect采集图像,在深度图像中对深度差值及方向梯度进行计算,通过滑动窗口对整个深度图像进行特征提取,获得特征向量,并利用主成分分析法降维.最后利用随机森林选取分类能力较强的特征并进行分类,实现训练及检测.在不同背景及光照条件下进行检测实验,平均检测率达到87.89%,平均每帧检测时间为0.121s.将本方法与GEBCF(泛化和检测平衡共生特征)及FCF(滤波通道特征)算法对比,检测率分别提高0.92%、0.68%.实验结果表明本方法有效提高了行人检测的准确率,具有更高的检测效率,能快速、准确地检测行人.In order to solve the problems of inaccurate detection and low efficiency in traditional pedestrian detection algorithm in the building,which caused by occlusion,complex background and illumination.This paper proposesa method based on depth image feature for pedestrian detection.Weuse Kinect to get images,and calculatethe depth difference and the direction gradient in the depth images.The feature extraction of the whole depth image is carried out by sliding window,and the dimensionis reduced by the principal component analysis （PCA）.Finally,we use the random forest to select some features with strong classify ability as the final feature,to get the classifier and achieve training and testing.We carry out the test in different background and illumination conditions,the detection rate was 87.89%,and the average detection time was 0.121s per frame.At the same time,the method is compared with GEBCF algorithm and FCFalgorithm,the detection rate is improved by 0.92% and 0.68%,respectively.According to the experimental results,this method can effectively improve the accuracy of pedestrian detection,and has higher detection efficiency.It can detect pedestrians quickly and accurately.

关 键 词：行人检测 深度差值特征 深度方向梯度特征 主成分分析法 随机森林 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]