基于CENTRIST方法与显著图分割的行人检测方法  

Pedestrian Detection Based on CENTRIST Method and Saliency Segmentation

作　　者：李磊[1,2] 张博[2,3] 

机构地区：[1]上海市公安局青浦分局,上海201700 [2]上海交通大学电子工程系图像通信与网络工程研究所 [3]上海交通大学上海市数字媒体处理与传输重点实验室,上海200240

出　　处：《电视技术》2015年第7期96-99,共4页Video Engineering

摘　　要：在智能监控系统中,行人是最为关键的目标对象。监控系统可根据当前捕获的行人检测结果,触发跟踪系统持续观察兴趣目标,从而给出兴趣目标的行为与状态信息。考虑到行人检测结果直接影响跟踪系统的输出,基于CENTRIST(Census Transform Histogram)方法的行人检测结果,采用显著图分割技术,将包含行人的前景区域与检测框内的背景区域分离开来,使得跟踪系统能够根据行人的主体运动部分做出准确判断,有效地缓解了背景区域以及行人局部运动(如手、脚运动)对跟踪结果的干扰。实验结果表明,该行人检测方法在提高后续跟踪模块准确率的同时,又能适用于实时性要求较高的智能视频监控系统。Pedestrians are the most important objects in intelligent surveillance tasks. When a pedestrian is detected, tracking mod- ule of the surveillance system will be triggered. Therefore, activity analysis and status description of the pedestrian can be provided when he/she is detected and tracked. Considering that the tracking pertbrmance is directly affected by the initial detection result, in this paper, a pedestrian detection method based on CENTRIST algorithm and saliency segmentation is proposed, which alleviates the interference in tracking incurred by background regions and local motion of human＇ s hands and feet. The background region and the main moving part of a pedestrian in the CENTRIST detection bounding box can be efficiently separated by the saliency segmen- tation method. Therefore, the tracking performance can be improved effectively since only the global motion of a pedestrianis com- puted. The experimental results show that the proposed method achieves promising tracking performance and shows applicability for a real-ti,ne intelligent surveillance system.

关 键 词：智能监控 行人检测 显著图分割 

分 类 号：TN919.8[电子电信—通信与信息系统]