基于多特征融合的行人检测跟踪统计方法  

Pedestrian Detection and Tracking Count Method Based on Multi-feature Fusion

作　　者：史涛[1] 程方[1] 

机构地区：[1]重庆邮电大学通信网与测试技术重点实验室,重庆400065

出　　处：《电视技术》2014年第19期179-183,共5页Video Engineering

基　　金：国家科技重大专项(2012ZX03005008)

摘　　要：为了避免传统方法行人过线统计的不足,提高视频中人数统计的适用性和有效性,提出了一种基于多特征融合的行人检测跟踪统计方法。首先,采用垂直拍摄的方式获取视频,利用Mean-Shift分割算法分割图像,根据发色信息和头部轮廓特征识别出人头目标区域;其次使用融合多特征的匹配算法对人头进行匹配跟踪;最后通过运动目标轨迹分析估算出监控区域内的人数。实验结果表明提出的算法在保证准确率的前提下,扩大了人数统计的适用性。A novel method of pedestrian detection and tracking count is presented based on multi-feature fusion, and it can avoid the deficiency that the statistical pedestrian crossed the line of traditional methods to improve applicability and effectiveness of people counting in video sequence. Firstly, it used vertical shot way to get video and used the Mean-Shift segmentation algorithm to segment a video sequence. Head target area could be identified ac- cording to hair color clustering information and head silhouette feature. Secondly, a mulfi-feature fusion match algorithm is used for head matching and tracking. Lastly, the number of people within the monitored area could be analyzed and estimated by moving target trajectory. Experimental results show that the proposed algorithm expand the applicability of the statistical pedestrian on the condition that the algorithm guarantees the accuracy.

关 键 词：均值偏移 人头识别 匹配跟踪 目标链 轨迹分析 人数统计 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]