主办：陕西省汽车工程学会
ISSN 1671-7988  CN 61-1394/TH
创刊：1976年
汽车实用技术

汽车实用技术 ›› 2022, Vol. 47 ›› Issue (5): 34-37.DOI: 10.16638/j.cnki.1671-7988.2022.005.008

• 智能网联汽车 • 上一篇    

基于改进 Faster-RCNN 算法的行人检测

贺艺斌，田圣哲，兰贵龙   

  1. 长安大学 汽车学院
  • 出版日期:2022-03-15 发布日期:2022-03-15
  • 通讯作者: 贺艺斌
  • 作者简介:贺艺斌（1995—），硕士研究生，研究方向为 智能驾驶，E-mail:2019122006@chd.edu.cn。

Pedestrian Detection Based on Improved Faster-RCNN Algorithm

HE Yibin, TIAN Shengzhe, LAN Guilong   

  1. School of Automobile, Chang’an University
  • Online:2022-03-15 Published:2022-03-15
  • Contact: HE Yibin

PDF

摘要： 行人检测是汽车智能化进程中十分重要的任务，而深度学习的蓬勃发展为其指明了新 的方向。文章通过对 Faster-RCNN 算法进行改进，将其应用于车载摄像头下的行人检测。将 特征提取网络由原来的 Vgg16 网络改为 ResNet50 网络，并增加了锚点个数，提高了算法对小 目标行人的检测效果。文章还引入了 FPN 金字塔结构，其更适用于多尺度的行人检测。最后 将改进后的算法在 Caltech 行人数据集上进行了试验，取得了较好的检测效果，mAP 达到了 95%。

关键词: 深度学习；Faster-RCNN；行人检测

Abstract: Pedestrian detection is a very important task in the process of automobile intelligence, and the vigorous development of deep learning has pointed out a new direction for it.This paper improves the Faster-RCNN algorithm and applies it to pedestrian detection based on vehicle cameras. In this paper, the feature extraction network is changed from the original Vgg16 network to the Resnet50 network, and the number of anchor points is increased, which improves the detection effect of the algorithm on small target pedestrians. This paper also introduces the FPN pyramid structure, which is more suitable for multi-scale pedestrian detection. Finally, the improved algorithm was tested on the Caltech pedestrian data set, and a good detection effect was achieved, and the mAP reached 95%.

Key words: Deep learning; Faster-RCNN; Pedestrian detection