基于智能视频分析的行人检测、跟踪系统
首发时间:2017-05-31
摘要:随着国家"开放小区"系统工程的全面建设,大量摄像头和监控节点投入使用,加重了监控人员的负担,带来围绕智慧社区的安防监控的改革升级,其中小区人物跟踪、行为预警分析凸显出重要地位。因此,论文采用智能视频分析的方法进行行人检测、跟踪的功能研究开发。采用梯度方向直方图(Histogram of Oriented Gradient,HOG)+支持向量机(SupportVector Machine,SVM)的思路设计行人检测功能,通过难例样本训练与非极大值抑制(Non-Maximum Suppression,NMS)矩形框融合机制提升了分类器的性能,实际场景下测试的平均检测速率41.66ms/帧,检测平均准确率达到99.49%,平均召回率为72.21%,检测平均综合效果达到83.67%;设计融合颜色直方图特征的粒子滤波行人跟踪方案,经实验验证实际场景下,平均跟踪处理速度保持在6.5-8.5ms/帧,满足实时需求。同时设计开发了具有交互界面的系统客户端,研究内容对于小区视频监控中行人预警分析具有一定的实用价值。
关键词: 智能视频分析 行人检测跟踪 HOG特征 SVM分类 粒子滤波
For information in English, please click here
Pedestrian Detection and Tracking System Based on Intelligent Video Analysis
Abstract:Withthe national systematic projects of "Open District"vigorously constructed,lots of cameras and monitoring nodes are put into use,and a large number of camera and monitor node came into use, thus increased the burden on the monitoring staff, bringed the reform and upgrading of the security monitoring around smart community, among which pedestrian tracking and behavioral warning analysis highlight the important position.Therefore, the paper proposed a system about pedestrian detection and trackingbased on intelligent video analysis. Method of HOG+SVM is used to designed function of pedestrian detection, by training difficulties and using NMS rectangular frame fusion mechanism to improve the performance of the classifier, the average detection rate in the actual scenes is 41.66 ms/frame, the average accuracy rate of the detection achieves 99.49%, the average recall rate is 72.21% and the average combined effect of the detection achieves 83.67%; A tracking scheme is designed to track pedestrians based on particle filter with color feature histogram, system tests showed thatthe average tracking speed in the actual scenes maintain at 6.5-8.5 ms/frame,is able to meet the real-time tracking demands.At the same time, the system client with interactive interface is designed and developed, the research of the paper is of certain practical value to behavioral warning analysis in community video surveillance.
Keywords: Intelligent video Analysis Pedestrian Detection and Tracking HOG Features SVM Classifier Particle Filter
论文图表:
引用
No.4734905120268114****
同行评议
共计0人参与
勘误表
基于智能视频分析的行人检测、跟踪系统
评论
全部评论0/1000