Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (11): 141-149.DOI: 10.3778/j.issn.1002-8331.2011-0115

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Occluded Pedestrian Detection Based on Multi-Scale Context Information

ZHAO Shiyang, WANG Xiaofeng   

  1. College of Information Engineering, Shanghai Maritime University, Shanghai 201306, China
  • Online:2022-06-01 Published:2022-06-01



  1. 上海海事大学 信息工程学院,上海 201306

Abstract: Pedestrian detection in occluded scenes has always been a thorny problem in computer vision. In this case, due to the large difference in scale of occluded pedestrians and low visibility, it usually brings great challenges to detection. To solve this problem, this paper proposes a model structure for pedestrian occlusion detection, which improves the pedestrian detection method based on anchor-free. First, a structure for extracting multi-scale context information is designed. By cascading multiple convolutional layers with different dilation rates, using dense connections to achieve multi-scale feature sharing, the context information of each region is extracted to solve the occlusion problem. In addition, in order to improve the robustness of features, the multi-scale feature fusion is adaptive adjusted using the channel attention mechanism. Experimental results show that this method achieves 41.73% of MR'2 on the occlusion subset of Caltech pedestrian dataset, which is better than other contrast detectors.

Key words: pedestrian detection, multi-scale context, channel attention, anchor-free

摘要: 在遮挡场景下的行人检测一直是计算机视觉中的一个棘手问题,由于被遮挡的行人尺度差异大,可见率低,通常会给检测带来极大的挑战。针对这一问题,提出了一种针对行人遮挡检测的模型结构,对基于anchor-free的行人检测方法进行改进。设计了一种提取多尺度上下文信息的结构,通过级联多个不同扩张率的卷积层,使用密集连接实现多尺度特征共享,提取各个区域的上下文信息来解决遮挡问题。此外,为了提高特征的可分辨性,使用通道注意力机制对多尺度特征融合进行自适应的调整。实验结果表明,该方法在Caltech行人数据集的遮挡子集上实现了41.73%的MR?2,性能优于其他检测算法。

关键词: 行人检测, 多尺度上下文, 通道注意力, anchor-free