High quality proposal feature generation for crowded pedestrian detection

作者：

Highlights：

• Dual-region feature generation is proposed to eliminate false positives among different instances for crowded pedestrian detection.

• The problem of pedestrian overlap is effectively dealt with by paired multi-instance prediction.

• Paired non-maximum suppression is utilized to overcome the shortcomings of the original NMS and solve the problems of false and missed detection.

摘要

•Dual-region feature generation is proposed to eliminate false positives among different instances for crowded pedestrian detection.•The problem of pedestrian overlap is effectively dealt with by paired multi-instance prediction.•Paired non-maximum suppression is utilized to overcome the shortcomings of the original NMS and solve the problems of false and missed detection.

论文关键词：Crowded pedestrian,Pedestrian detection,Visible proposal,Feature fusion,Paired prediction

论文评审过程：Received 24 September 2021, Revised 11 February 2022, Accepted 21 February 2022, Available online 28 February 2022, Version of Record 5 April 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108605