基于YOLOv8目标检测的双目实时测距研究  

Research on Binocular Real-time Ranging Based on YOLOv8 Object Detection

作　　者：姜伟 JIANG Wei(Yangzhou Polytechnic College,Yangzhou 225009,China)

机构地区：[1]扬州职业大学,江苏扬州225009

出　　处：《扬州职业大学学报》2025年第1期30-34,共5页Journal of Yangzhou Polytechnic College

摘　　要：在双目测距的应用中,目标检测的精度决定了测距精度。为了选取合适的模型以减少计算量,分析部分超参数对训练结果的影响,选取YOLOv8n和YOLOv8m预训练模型,使用自定义数据集训练得到多个目标检测模型,分析了不同训练参数,如批次大小、迭代轮数和预训练权重等对训练过程和结果的影响。探讨如何在C#中使用ONNX Runtime部署YOLOv8检测模型,并基于双目视觉原理实现实时测距,该研究对部分自动化装置的目标检测和测距应用有参考价值。In the application of binocular ranging,the accuracy of object detection determines the precision of distance measurement.In order to select an appropriate model to reduce the computational load and analyze the influence of certain hyperparameters on the training results,the YOLOv8n and YOLOv8m pre-trained models are selected.Multiple object detection models are acquired through training with a custom data set,and the influence of different training parameters(such as batch size,number of iterations and pre-trained weights)on the training process and results is analyzed.This paper further discusses the methodology of deploying the YOLOv8 detection model via ONNX Runtime in C#and implement real-time distance measurement based on the binocular vision principle.This research offers great reference value for the application of object detection and ranging in certain automated devices.

关 键 词：YOLOv8目标检测 双目测距 实验分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]