摘要
Security surveillance is widely used in daily life.For nighttime or complicated monitoring environments,this article proposes an infrared pedestrian monitoring based on YOLOv3.In the original YOLOv3 network structure,two aspects of optimization were made.One was to optimize the scale in the residual structure,and the rich features of the deconvolution layer were added to the original residual structure.The other was to use the DenseNet network to enhance the features.The optimization of fusion ability and delivery ability effectively improves the detection ability for small targets,and the pedestrian detection performance based on infrared images.After comparative testing,compared with YOLOv3,the overall mean average precision is improved by 4.39%to 78.86%.
出处
《国际计算机前沿大会会议论文集》
2020年第1期506-517,共12页
International Conference of Pioneering Computer Scientists, Engineers and Educators（ICPCSEE）
基金
the Fundamental Research Funds for the Local Universities of Hei longjiang Province in 2018(Grant No.2018-KYYWF-1189)
Shanghai Aerospace Science and Technology Innovation Fund(Grand No.SAST2017-104).
