基于改进RPN深度网络的端到端的监控场景行人检测研究  

End-to-end monitoring scene pedestrian detection based on improved RPN

作　　者：杨东明[1,2] 徐士彪 孟维亮[3] 葛水英[1] 杨真[1,2] 张晓鹏[3] 

机构地区：[1]中国科学院文献情报中心,北京100090 [2]中国科学院大学,北京100090 [3]中国科学院自动化研究所,北京100090

出　　处：《中国体视学与图像分析》2017年第2期209-215,共7页Chinese Journal of Stereology and Image Analysis

基　　金：国家科技支撑计划(No.2012BAH42F02);国家自然科学基金重点项目(61332017);国家自然科学基金面上项目(61671451;61571046)

摘　　要：监控场景下的行人检测相比其他场景,具有人流量大,遮挡程度高的特点。针对该场景,本文提出基于RPN深度网络的端到端检测方案。一方面改进了RPN深度网络,结合自行设计的深度卷积网络,获得了针对行人检测的深度网络。另一方面通过引入头肩模型进一步提高了针对该场景行人检测的检测性能,同时提升了检测实时性,最终实现了端到端检测。实验表明,本文方法有效提升了监控场景行人检测的检测性能和实时性,降低了检测漏检率。Compared to other scenes, pedestrian detection in monitoring scenes gets larger flow of people and higher degree of occlusion. This paper presents an end - to - end detection scheme based on region proposal network （ RPN）. On the one hand, we improved RPN by combining with a deep convolution net- work designed by ourselves to obtain a new deep neural network for pedestrian detection. On the other hand, we introduced head - shoulders model to improved detection performance for pedestrian detection at monitoring scenes, which improves the detection speed. Eventually, we achieved end - to - end detec- tion. The experimental results show that the method effectively improved the detection performance by re- ducing the missing rate and speeding up the detection.

关 键 词：监控 行人检测 深度学习 RPN网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]