摘要
针对机动车行驶中驾驶员是否正确系戴安全带的检测问题,提出一种融合目标检测和语义分割的新型Unet-Canny网络的安全带检测方法。该方法需要用SSD网络定位驾驶员位置,将所采集信息输入Unet-Canny网络后对安全带图像进行分割。新型Unet-Canny网络是在Unet网络上添加Res-Canny模块后得到的。结合Canny算子和残差结构搭建的Res-Canny模块能够让网络在训练中增强对安全带边缘特征信息的提取能力,从而提高图像的分割效果。利用数据集对Unet-Canny网络的训练实验表明:新型Unet-Canny网络比FCN、PSPNet、SegNet、Unet等网络的检测效果好;使用所提出方法检测安全带的精确率比用Unet网络提升了3.4个百分点。
To solve at the problem of small and irregular objects in driver’s safety belt detection,a new method of safety belt detection based on Unet-Canny network combining object detection and semantic segmentation was proposed.In this method,SSD network is used to locate the driver’s position,then the collected information is input into Unet-Canny network to segment the seat belt image.The new Unet-Canny network is based on Unet network with Res-Canny module.Delete which is built by combining Canny operator and residual structure.This module can enhance the extraction ability of seat belt edge features in training,thus improving the image segmentation effect.The experimental results of Unet-Canny network training by using data sets show that the enhanced Unet-Canny network has better accuracy and recall rate than Unet network,as well as FCN,PSPNet,SegNet and Unet network.The accuracy of using the proposed algorithm to detect seat belt is 3.4 percentage points higher than that of using Unet network.
作者
彭方达
宋长明
李阳
王浩
PENG Fangda;SONG Changming;LI Yang;WANG Hao(School of Science,Zhongyuan University of Technology,Zhengzhou 450007,China)
出处
《中原工学院学报》
CAS
2022年第4期22-30,共9页
Journal of Zhongyuan University of Technology
基金
国家自然科学基金项目(12171438)。