当前位置>主页 > 期刊在线 > 信息技术 >

信息技术22年15期

基于双目立体视觉的实时测速系统设计与实现
李登览¹,宋晓炜² ,范存辉¹,朱鲁晓¹
(1. 中原工学院 电子信息学院,河南 郑州 450007;2. 开封大学,河南 开封 475004)

摘  要:针对遥控车速度竞赛,进行了基于双目立体视觉的实时测速系统的设计。遥控车实时测速系统搭载于 Jetson Xavier NX,根据双目立体视觉的视差原理,借助 YOLOv5s 目标检测算法实现了复杂环境下的遥控车检测、距离测量、速度测定、实时监控等功能。为了提高系统的检测效率,通过网络剪枝对 YOLOv5s 目标检测算法进行了模型压缩,同时使用 TensorRT 进行了模型加速。实验结果表明,该设备能够稳定运行,对遥控车的检测具有良好的鲁棒性和检测精度。


关键词:双目立体视觉;目标检测;实时测速;模型压缩



DOI:10.19850/j.cnki.2096-4706.2022.15.005


基金项目: 中原科技创新领军人才(214200510013);河南省高校重点科研项目(21A510016,21A520052);留学人员科研资助和创业启动项目(HRSS2021[36]);校内重大项目成果培育计划(K2020ZDPY02)


中图分类号:TP391.4                                     文献标识码:A                                        文章编号:2096-4706(2022)15-0020-04

Design and Implementation of Real-time Speed Measurement System Based on Binocular Stereo Vision

LI Denglan¹, SONG Xiaowei², FAN Cunhui¹, ZHU Luxiao¹

(1.School of Electronic Information, Zhongyuan University of Technology, Zhengzhou 450007, China; 2. Kaifeng University, Kaifeng 475004, China)

Abstract: A real-time speed measuring system based on binocular stereo vision is designed for the speed race of remote-controlled vehicles. The remote-controlled vehicle real-time speed measurement system is mounted on the Jetson Xavier NX. Based on the parallax principle of binocular stereo vision, the remote-controlled vehicle detection, distance measurement, speed measurement, real-time monitoring and other functions under complex environment are realized with the help of YOLOv5s target detection algorithm. In order to improve the detection efficiency of the system, Model compression of YOLOv5s target detection algorithm is performed by network pruning, and the model is accelerated by TensorRT. The experimental results show that the equipment can operate stably and has good robustness and detection accuracy for the detection of remote-controlled vehicles.

Keywords: binocular stereo vision; target detection; real-time speed measurement; model compression


参考文献:

[1] 王文成,李健.基于单片机的电机转速测量系统的设计 [J].仪表技术与传感器,2011(8):70-72.

[2] ZAIDI S,ANSARI M S,ASLAM A,et al. A Survey of Modern Deep Learning based Object Detection Models [J].IEEE Access,2019,7:128837-128868.

[3] BOCHKOVSKIY A , WANG C Y , L I A O H . YOLOv4: Optimal Speed and Accuracy of Object Detection [J/ OL].arXiv:2004.10934 [cs.CV].[2022-05-30].https://arxiv.org/ abs/2004.10934.

[4] 李梦龙 . 基于双目立体视觉的车辆速度轨迹智能测量 [D].郑州:中原工学院,2019.

[5] 刘艳,李腾飞 . 对张正友相机标定法的改进研究 [J]. 光学技术,2014,40(6):565-570.


作者简介:李登览(1995—),男,汉族,河南新乡人,硕士研究生在读,研究方向:计算机视觉;宋晓炜(1978—),男,汉族,山西大同人,教授,博士,研究方向:图像处理、计算机视觉;范存辉(1997—),男,汉族,山东滨州人,硕士研究生在读,研究方向:计算机视觉;朱鲁晓(1997—),女,汉族,河南平顶山人,硕士研究生在读,研究方向:计算机视觉。