Abstract:Pedestrian detection has achieved significant progress in recent years. Though promising results have been obtained on standard pedestrians, it remains challenging to detect pedestrians in various occlusion situations. In this paper, we propose Progressive Refinement Network (PRNet) , a novel single-stage detector for occluded pedestrian detection. Inspired by human’s progressive process on annotating occluded pedestrians, PRNet perform sequential refinement in a single-stage detection framework by three phases: finding anchors of visible parts with high confidence, calibrating these anchors with a full body template derived from occlusion statistics, and adjusting the calibrated anchors to target full-body regions. Unlike conventional methods that utilize predefined anchors directly for full-body estimation, the proposed confidence-aware anchor calibration offers an adaptive anchor initialization for detection with occlusions, while helps reduce the gap between visible-part and full-body detection. We also propose an occlusion loss, which automatically up-weights heavily occluded pedestrian samples. In addition, a Receptive Field Backfeed (RFB) module is introduced to diversify receptive fields in early layers that commonly fire only on visible parts or small-size full-body regions. To further learn generalized representations of pedestrians in different occlusion states, we propose to establish a new single-stage detector with dual-stream architecture namely PRNet++. An Easy-branch and a Hard-branch are designed to learn complementary representation that are more robust to various occlusions. Moreover, the generalization ability of models for other domains are critical in real-world applications, but it has not attracted too much attention. To address this issue, we introduce the unsupervised domain adaptation setting. To handle occlusion situations in unknown domain better, we especially design a Dynamic Iterative adaptation strategy and a Multi-experts adaptation strategy for our occlusion-aware detectors PRNet and PRNet++. Extensive supervised within-dataset experiments and unsupervised domain adaptation experiments were performed on CityPersons, Caltech, ECP-night, KITTI and INRIA datasets, which can validate the effectiveness of our proposed methods in occluded pedestrian detection.

Learning Hierarchical Graph for Occluded Pedestrian Detection

Towards Accurate Dense Pedestrian Detection Via Occlusion-Prediction Aware Label Assignment and Hierarchical-Nms.

Pedestrian Detection Method Based on Improved YOLOv5s for Densely Occluded Scenarios

Mask-Guided Attention Network for Occluded Pedestrian Detection

PedHunter: Occlusion Robust Pedestrian Detector in Crowded Scenes

Mask-Guided Attention Network and Occlusion-Sensitive Hard Example Mining for Occluded Pedestrian Detection

Solving Occlusion Problem in Pedestrian Detection by Constructing Discriminative Part Layers

PSC-Net: learning part spatial co-occurrence for occluded pedestrian detection

OADB-Net: an Occlusion-Aware Dual-Branch Network for Pedestrian Detection

Selective Kernel and Spatial Grouping Attention Network for Occluded Pedestrian Detection

MSAGNet: Multi-Stream Attribute-Guided Network for Occluded Pedestrian Detection

Coupled Network for Robust Pedestrian Detection With Gated Multi-Layer Feature Extraction and Deformable Occlusion Handling

Pedestrian Partial Occlusion Detection Method Based on Improved YOLO Network Structure

Pedestrian detection based on hierarchical co-occurrence model for occlusion handling

PRNet++: Learning Towards Generalized Occluded Pedestrian Detection Via Progressive Refinement Network

Progressive Refinement Network for Occluded Pedestrian Detection

Occluded Pedestrian Detection via Distribution-Based Mutual-Supervised Feature Learning

Detection of partially occluded pedestrians by an enhanced cascade detector

Occluded Pedestrian Detection Through Guided Attention in CNNs.

Online Multiple-Pedestrian Tracking with Detection-Pair-Based Graph Convolutional Networks

Learning Pixel-Level and Instance-Level Context-Aware Features for Pedestrian Detection in Crowds.