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.

Fast Pedestrian Detection Method Based on BING

A Novel Approach to Design the Fast Pedestrian Detection for Video Surveillance System

Multi-scale Human Detection Based on Window Gradient Potential Energy with Partial Occlusion Handling

Fast Pedestrian Detection And Tracking Based On Vibe Combined Hog-Svm Scheme

Pedestrian Detection Method Based on Improved YOLOv5s for Densely Occluded Scenarios

Real-time Pedestrian Crossing Lights Detection Algorithm for the Visually Impaired

Pedestrian detection based on modified dynamic background using gaussian mixture models and hog-svm detection

Pedestrian detection using improved Histogram of Oriented Gradients

Pedestrian Detection in Complex Scene Using Full Binary Tree Classifiers Based on Locally Assembled Binary Haar-Like Features

Pedestrian Detection Optimization Algorithm Based on Low-Altitude UAV

Effective Pedestrian Detection Using Center-symmetric Local Binary/Trinary Patterns

Fast Pedestrian Detection Based on Multiple Instance Hierarchical Hog Matrices

An Efficient Pedestrian Detection for Realtime Surveillance Systems Based on Modified YOLOv3

Pedestrian Detection Based on Center, Temperature, Scale and Ratio Prediction in Thermal Imagery

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

Color Self-Similarity Feature Based Real-Time Pedestrian Detection

Pyramid Center-Symmetric Local Binary/Trinary Patterns for Effective Pedestrian Detection

Fast moving pedestrian detection based on motion segmentation and new motion features

Pedestrian detection in video of outdoor condition

Efficient And Real-Time Pedestrian Detection At Night-Time Environments

Pedestrian Detection Based on an Assembly of Sub-image Detectors