Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (21): 214-221.DOI: 10.3778/j.issn.1002-8331.2206-0315

• Graphics and Image Processing • Previous Articles     Next Articles

Dense Pedestrian Detection with Iterative Faster R-CNN

HE Yuzhe, XU Guangmei, HE Ning, YU Haigang, ZHANG Ren, YAN Kang   

  1. 1.Beijing Key Laboratory of Information Service Engineering, Beijing Union University, Beijing 100101, China
    2.College of Smart City, Beijing Union University, Beijing 100101, China
  • Online:2023-11-01 Published:2023-11-01

迭代Faster R-CNN的密集行人检测

贺宇哲,徐光美,何宁,于海港,张人,晏康   

  1. 1.北京联合大学 北京市信息服务工程重点实验室,北京 100101
    2.北京联合大学 智慧城市学院,北京 100101

Abstract: Pedestrian detection uses computer vision technology to determine whether there are pedestrians in the image or video sequence and give accurate positioning. In this paper, a dense pedestrian detection model based on iterative Faster R-CNN is proposed to solve the common pedestrian occlusion problem in dense scenes. An IterDet iteration scheme is used to improve Faster R-CNN, which effectively solves the problem of choosing a balance between precision and recall for the NMS and its improvement. At the same time, the recursive pyramid structure is used to further enhance the feature extraction ability of the model. This paper trains and validates on the challenging WiderPerson and CrowdHuman datasets. The experimental results show that compared with Faster R-CNN, the model in this paper significantly improves the precision and recall, but also significantly reduces the mMR. Especially on the WiderPerson dataset, the recall, precision, and mMR has reached SOTA results of 97.65%, 91.29%, and 40.43%, respectively.

Key words: pedestrian detection , dense scenes, occlusion problem, Faster R-CNN, iterative scheme

摘要: 行人检测是利用计算机视觉技术判断图像或者视频序列中是否存在行人并给予精确定位。针对行人检测在密集场景下普遍存在的行人间遮挡问题,提出基于迭代Faster R-CNN的密集行人检测模型,利用一种IterDet迭代方案对Faster R-CNN进行改进,有效解决非极大值抑制(NMS)算法及其改进在选择精确度和召回率之间平衡点的难题。同时利用递归金字塔结构(RFP)进一步增强模型提取特征能力。在具有挑战性的WiderPerson和CrowdHuman数据集上进行训练和验证,实验结果表明,该模型相比Faster R-CNN在精度和召回率显著提升的同时,漏检率也明显降低。尤其在WiderPerson数据集上召回率、精度、漏检率等性能指标分别达到了97.65%、91.29%、40.43%的SOTA结果。

关键词: 行人检测, 密集场景, 遮挡问题, Faster R-CNN, 迭代方案