摘要
提高大型停车场中的停车位检测精度和实时性具有重要意义。介绍了在基于深度学习框架tensorflow下搭建包括基础网络和辅助网络的网络结构。基础网络是Resnet网络,用于提取图像特征信息和图像分类信息;辅助网络是多尺度特征检测网络,用于提取不同尺度的特征图。最后通过非极大值抑制算法筛除重复检测框,得到停车位检测最佳位置。实验结果表明,该网络mAp值为81%,fps为32,与SSD、YOLO、Faster R-cnn相比,mAp值分别提高为2%,4.6%,0.5%,fps值分别提高为2,4,24,有效提高检测精度和实时性。
It is of great significance to improve the detection accuracy and real-time performance of parking spaces in large parking lots.Introduces the construction of a network structure including basic network and auxiliary network under tensorflow based on deep learning framework.The basic network is a Resnet network for extracting image feature information and image classification information.The auxiliary network is a multi-scale feature detection network for extracting feature maps of different scales.Finally,the non-maximum suppression algorithm is used to screen out the duplicate detection frame to get the best position for parking space detection.The experimental results show that the network mAp value is 81%,fps is 32.Compared with SSD,YOLO,and Faster R-cnn,the mAp value is increased to 2%,4.6%,0.5%,and the fps value is increased to 2,4,24,effectively improve detection accuracy and real-time.
作者
王马成
黎海涛
Wang Macheng;Li Haitao(Faculty of Information technology,Beijing University of technology,Beijing 100124,China)
出处
《电子测量技术》
2019年第21期105-108,共4页
Electronic Measurement Technology
