摘要
借鉴目标识别领域的快速区域卷积神经网络(Fast RCNN),提出了基于Faster RCNN的行人检测方法,利用CNN提取图像特征,通过聚类和构建区域建议网络(RPN)提取可能含有行人的区域,利用检测网络对目标区域进行判别和分类,并在INRIA数据集中进行了测试验证。实验结果表明:相比基于可变形的组件模型(DPM)的行人检测方法,提出的方法,在测试集上检测准确度达到92. 7%,相比现有的其他方法,其检测效果更好。
Using Fast region-based convolutional neural network(Fast RCNN) for reference,propose pedestrian detection method based on Faster RCNN. in target recognition field,. Image features are extracted by CNN. A region proposal network(RPN)is built up to extract regions that might contain pedestrians combined with K-means cluster analysis. And the region is identified and classified by detection network. The method is tested in the INRIA dataset. Experimental results show that compared with the method of pedestrian detection based on DMP, the proposed method achieves the accuracy of 92. 7 %,it performs better,compared with other algorithms.
作者
张汇
杜煜
宁淑荣
张永华
杨硕
杜晨
ZHANG Hui;DU Yu;NING Shu-rong;ZHANG Yong-hua;YANG Shuo;DU Chen(Smart City College of Beijing Union University,Beijing 100101,China;College of Robotics of Beijing Union University,Beijing 100101,China;Beijing Key Laboratory of Information Service Engineering,Beijing 100101,China)
出处
《传感器与微系统》
CSCD
2019年第2期147-149,15,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金重大研究计划项目(91420202)
关键词
快速区域卷积神经网络
区域建议网络
行人检测
深度学习
Fast region-based convolutional neural network ( Fast RCNN)
region proposal network ( RPN)
pedestrian detection
deep learning