Abstract:With the development of self-driving technology, pedestrian detection models are becoming more useful and important. Although object detection models based on Convolutional Neural Networks have achieved favorable detection results, due to the occlusion in crowded scenarios, dense pedestrian detection is still a challenging task compared with generic object detection. For a CNN-based object detector, an appropriate label assignment method is significant. In this paper, we propose an effective training sample allocation method called Occlusion-Prediction aware Label Assignment (OPLA) to improve the performance of the detector in crowded scenarios. For those Ground Truths (GTs) without severe occlusion, we directly assign prediction boxes with the best matching quality as the corresponding positive samples. In terms of GTs in heavily crowded areas, we fine-tune the matching quality based on visible ratio to enhance the training sample selection priority of severe occluded GTs. In addition, we reveal the deficiency of currently popular post-processing algorithms, Non-Maximum Suppression (NMS) and Soft-NMS for dense pedestrian detection, and propose Hierarchical-NMS (H-NMS) whose strategy diverges with different Intersection over Union (IoU) levels. Experimental results on Crowdhuman, the largest dense pedestrian detection dataset at present, show our approach optimizes the Average Precision (AP) and Miss Rate (MR) of Fully Convolutional One Stage Object Detection (FCOS) by 4.05% and 5.35% and outperforms many state-of-the-art detectors, respectively. Besides, our method does not bring any increase in calculation during inference.

Detection of partially occluded pedestrians by an enhanced cascade detector

Pedestrian Detection Method Based on Improved YOLOv5s for Densely Occluded Scenarios

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

A Study On Occluded Pedestrian Detection Based On Block-Based Features And Ensemble Classifier

Pedestrian Partial Occlusion Detection Method Based on Improved YOLO Network Structure

Pedestrian Detection Based on Merged Cascade Classifier.

Cascade Features Based Method for Pedestrian Detection in Street Scene

Solving Occlusion Problem in Pedestrian Detection by Constructing Discriminative Part Layers

Occluded Pedestrian Detection Algorithm Based on Attention Mechanism

A cascaded classifier for pedestrian detection

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

Improved pedestrian detection with peer AdaBoost cascade

Learning Hierarchical Graph for Occluded Pedestrian Detection

PedHunter: Occlusion Robust Pedestrian Detector in Crowded Scenes

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

An Integrated Deep Learning Framework for Occluded Pedestrian Tracking.

Cascade Transformer Decoder Based Occluded Pedestrian Detection With Dynamic Deformable Convolution and Gaussian Projection Channel Attention Mechanism

An overview of fast pedestrian detection: feature selection and cascade framework of boosted features

A Boosted Multi-Task Model for Pedestrian Detection with Occlusion Handling.

A Pedestrian Detection Method Based on Hierarchical Tree Cascade Classification at Nighttime

A high accuracy pedestrian detection system combining a cascade AdaBoost detector and random vector functional-link net.