基于多帧特征融合的视频中夜间行人检测方法  

A Night Pedestrian Detection Method in Video Based on Multi-frame Feature Fusion

作　　者：王宇凡 张姗姗 WANG Yufan;ZHANG Shanshan(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机与数字工程》2023年第2期368-371,共4页Computer & Digital Engineering

摘　　要：近年来,深度卷积神经网络在行人检测领域取得了良好的成果,视频行人检测是行人检测的一个分支。论文针对视频中夜间行人检测提出一种基于多帧特征融合的离线检测方法,利用行人跟踪获取时序特征进行特征融合,在小尺度行人和被遮挡行人的检测上性能有所提升。In recent years,deep convolutional neural networks have achieved good results in the field of pedestrian detection.Video pedestrian detection is a branch of pedestrian detection.This paper proposes an offline detection method based on multi-frame feature fusion for pedestrian detection in video at night.The pedestrian tracking is used to obtain features of time sequence in feature fusion,which improves the performance of object detection in small and occluded set.

关 键 词：视频目标检测 行人检测 特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]