一种基于监控视频的有效的行人自动跟踪方法  

Effective automatic pedestrian tracking method based on surveillance video

作　　者：宗智嵩 王夏黎[1] 张明辉[1] 安毅生[1] 

机构地区：[1]长安大学信息工程学院,陕西西安710064

出　　处：《微型机与应用》2017年第12期46-49,共4页Microcomputer & Its Applications

基　　金：中国博士后科学基金资助项目(2012M521729)

摘　　要：行人跟踪是智能视频监控领域一个重要的研究内容,受到广泛关注。在实际应用中,对视频中行人跟踪处理的实时性和准确性都有很高要求,因此,如何在场景复杂多变的视频图像中完成对多行人自动、快速而准确的跟踪,是行人跟踪方法的研究热点。提出了一种有效的行人自动跟踪方法,该方法通过Ada Boost+Haar的检测框架以及SURF特征点匹配,改进原始的在线Boosting跟踪算法,解决了跟踪过程中由于目标尺度变化造成的跟踪误差问题并实现了对视频图像中多行人自动检测跟踪。使用校园内实际监控视频进行实验,实验结果证明该方法有较好的跟踪效果。Pedestrian tracking is an important research content in the field of intelligent video surveillance, which is being widely concerned. In practice, the pedestrian tracking in video processing has a high requirement of real-time and accuracy. Therefore, how to complete automat- ic, fast and accurate pedestrians tracking in the complex video scene is a hot spot. In this paper, an effective method of pedestrian tracking is proposed. This method combined with AdaBoost ＋ Haar detection framework and improved online Boosting algorithm with the SURF feature matching solves the error caused by multi-scale window and realizes the automatic detection and tracking of multi pedestrians in video images. Experiments with actual surveillance videos on campus show that the proposed method has a better processing effect.

关 键 词：智能监控视频 运动目标检测 多目标跟踪 ADABOOST 在线Boosting 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]