A Pedestrian Detection Algorithm Based on Improved R-FCN
XU Xiaolong
HE Xiaojia
FANG Yun
YU Xiaochun
YAN Yuxiang
Abstract：An improved pedestrian detection algorithm is proposed to solve the problems of misdetection of humanoid objects as pedestrians,omission of small-scale pedestrians and omission of multiple pedestrians overlapping in existing pedestrian detec-tion algorithms.Based on R-FCN framework and combined with the special needs of pedestrian detection,the appropriate trunk net-work is selected to make the pedestrian features extracted more distinguishability.Deformable convolution is introduced to improve the feature extraction ability of irregular pedestrian areas and reduce the probability of misdetecting human-like objects as pedestri-ans.Through the fusion of multipath detection results,the sensitivity of the detection framework to pedestrians of different scales is increased.The detection anchor frame is improved and the generation quality of anchor frame is improved.The non-maximum sup-pression algorithm is improved to improve the problem of pedestrian omission in overlapping areas.Experimental results show that the accuracy of the proposed algorithm on Caltech data set is 10.02%higher than that of the original algorithm.
Keywords：pedestrian detectionfusion of multipath detection resultsdeformable convolutionnon-maximum suppression algorithm
Publication Date：2024-11-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 3185-3190,3284 )