摘要
为解决智能视频监控系统下中远距离的行人检测问题,设计并实现了一个基于目标形状特征的检测系统;提出了4种较为有效的旨在描述人体形状局部特性的矩形特征,引入了对传统Adaboost算法的改进措施,减小了计算量,优化了系统结构;试验表明,该系统具有较高的检测率和处理速度,对阴影、雨天、枝叶晃动等监控场景中常见的噪声干扰不敏感,在P43·0GHz的PC上处理320×240的视频序列可达到15帧/s以上。
To solve the problem of pedestrian detection in far-field intelligent video surveillance system, this paper provides a detection system based on object＇s appearance characteristics. Four types of rectangle filters which efficiently describe the local characteristics of people＇s appearance are given and some modified mothed to adaboost algorithm are introduced, which decrease the computation time and optimize the system configuration. The result of experiments show the system has a high detection rate and realize real-time processing. The common noise disturbing which exist in the surveillance scenes such as shadows, raining, leaves blowing in the wind have little effect on the detection results, the system runs the 320 240 pixel video image sequence at above 15 frames/second on a 3.0 GHz P4 processor.
出处
《计算机测量与控制》
CSCD
2006年第11期1462-1465,共4页
Computer Measurement &Control
基金
国家自然科学基金资助项目(60372085)
陕西省自然科学研究计划项目(2003K06-G15)
关键词
行人检测
ADABOOST算法
矩形特征
智能视频监控
pedestrian detection
Adaboost algorithm
rectangle filters
intelligent video surveillance system
