Pedestrian and cyclist detection based on deep neural network fast R-CNN

Kelong Wang,Wei Zhou

doi:10.1177/1729881419829651

Abstract

In this article, a unified joint detection framework for pedestrian and cyclist is established to realize the joint detection of pedestrian and cyclist targets. Based on the target detection of fast regional convolution neural network, a deep neural network model suitable for pedestrian and cyclist detection is established. Experiments for poor detection results for small-sized targets and complex and changeable background environment; various network improvement schemes such as difficult case extraction, multilayer feature fusion, and multitarget candidate region input were designed to improve detection and to solve the problems of frequent false detections and missed detections in pedestrian and cyclist target detection. Results of experimental verification of the pedestrian and cyclist database established in Beijing’s urban traffic environment showed that the proposed joint detection method for pedestrians and cyclists can realize the stable tracking of joint detection and clearly distinguish different target categories. Therefore, an important basis for the behavior decision of intelligent vehicles is provided.

Highlights

The rapid development of intelligent driving technology has improved road traffic safety and urban traffic congestion
Li et al designed a cyclist detection method based on improved histograms of oriented gradients (HOG)
To solve frequent false detections and missed detections of pedestrians and cyclists, poor detection results of small-sized targets, and the complex and changeable background environment, this article presents the following main contributions: (1) a difficult case extraction method is designed based on the fast regional convolutional neural network, (2) a multilayer feature fusion method is designed, (3) an improved algorithm of depth network model is designed for multitarget candidate region input, and (4) a unified method of pedestrian and cyclist joint superscript detection and classification is constructed

Summary

Introduction

The rapid development of intelligent driving technology has improved road traffic safety and urban traffic congestion. To solve frequent false detections and missed detections of pedestrians and cyclists, poor detection results of small-sized targets, and the complex and changeable background environment, this article presents the following main contributions: (1) a difficult case extraction method is designed based on the fast regional convolutional neural network, (2) a multilayer feature fusion method is designed, (3) an improved algorithm of depth network model is designed for multitarget candidate region input, and (4) a unified method of pedestrian and cyclist joint superscript detection and classification is constructed. The results of this method are better than those of the traditional fast R-CNN method with a small increase in training time Drawing on this idea, this study designs a corresponding difficult case extraction network structure for fast R-CNN target detection by replacing the two shared full connection layers and output layers with an original full connection layer and output layer. RPN’s multitask loss function is shown in the following equation

N reg i

Experimental results and analysis

Conclusion and future work