基于改进SSD算法的骑车人识别研究

A cyclist identification study based on improved SSD

在智能网联汽车蓬勃发展的大背景下,目标识别作为智能驾驶的关键技术能够提高公路环境的安全性,本文采用SSD算法对公路环境下骑车人检测识别技术进行研究,发现SSD算法的小目标检测效果和平均识别精度都不太理想,故通过参考YOLOv3算法的跨层链接思想,在网络上引入FPN结构,进而提高识别效果。在TDCB数据集上的实验结果表明,平均检测精度和对小目标检测效果均有所提高,精度上提高约为2.2%,检测速度虽略微减缓,仍符合实际应用需求,改进后的SSD算法对提高公路环境下骑车人安全有着重要意义。

Under the background of the fast development of intelligent networked vehicles,target recognition,as a key technology of intelligent driving,can improve the safety of highway environment. In this paper,SSD algorithm is used to study the cyclist detection and recognition technology in highway environment. It is found that the small target detection and average recognition accuracy of SSD algorithm are not ideal. Therefore,by referring to the cross layer link idea of YOLOv3 algorithm,this paper introduces FPN structure into the network to improve the recognition effect. The experimental results on TDCB dataset show that the average detection accuracy and small target detection effect are improved,and the accuracy is improved by about2.2%. Although the detection speed is slightly slowed down,it still meets the requirements of practical application and the improved SSD algorithm is of great significance to improve the safety of cyclists in highway environment.