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.

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

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

Multiple Pedestrians Tracking Based on Gaussian Probability Model

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

Research on Hierarchical Pedestrian Detection Based on SVM Classifier with Improved Kernel Function

Pedestrian detection using improved Histogram of Oriented Gradients

A Method for Vehicle Detection Based on Local Gradients Vector

Pedestrian Detection Using a Mixture Mask Model

Vision-Based Moving Objects Detection with Background Modeling

An Enhanced Histogram of Oriented Gradients for Pedestrian Detection

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

Pedestrian Detection by Learning a Mixture Mask Model and Its Implementation

Representation and Analysis of Pedestrian Crossing States Based on Video Tracking

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

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

Pedestrian Detection Using Hybrid Statistical Feature.

Pedestrian Detection Via Part-Based Topology Model

Motion Based Pedestrian Recognition

Pedestrian Detection Optimization Algorithm Based on Low-Altitude UAV

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

A Novel Method for Tracking Pedestrians from Real-Time Video.