一种基于YOLOv5的双摄像头下目标物三维点云空间定位方法  

A 3D Point Cloud Spatial Localization Method Based on YOLOv5 Dual Camera

作　　者：蔺丽华[1] 沈树建 刘帅[1] 李元 付文旭 张汉瑾 LIN Lihua;SHEN Shujian;LIU Shuai;LI Yuan;FU Wenxu;ZHANG Hanjin(School of Communication and Information Technology,Xian University of Science and Technology,Xi'an 710600)

机构地区：[1]西安科技大学通信与信息工程学院,西安710600

出　　处：《计算机与数字工程》2024年第12期3623-3629,共7页Computer & Digital Engineering

基　　金：中国高校产学研创新基金项目(编号:2021KSA05005)资助。

摘　　要：随着定位技术的发展,高精度、低成本的定位问题已成为当前研究的热点。基于此,论文提出一种双摄像头目标物三维点云空间定位的方法。该方法使用YOLOv5对双摄像头下同一目标物进行目标识别后分别获取目标物像素坐标,再使用相机转换模型将像素坐标转换为世界坐标,使用最小二乘法求解三维坐标,对目标物体实现空间定位。最终用Colmap建立三维点云模型,实现目标物在三维点云中的三维可视化定位,提高人们对位置的感知能力。实验结果表明:该算法定位精度误差在X轴运动方向小于7 cm,在Y轴运动方向小于10 cm,在Z轴运动方向小于16 cm。获得了良好的空间定位效果。With the development of positioning technology,high-precision,low-cost,real-time positioning has become a hot topic in current research.Based on this,this paper proposes a three-dimensional point cloud spatial localization method for du-al-camera objects.In this method,YOLOv5 is used to identify the same object under dual cameras,and then the pixel coordinates of the object are obtained respectively.Then the camera conversion model is used to convert the pixel coordinates into world coordi-nates,and the least square method is used to solve the three-dimensional coordinates to achieve spatial positioning of the object.Fi-nally,Colmap is used to establish a 3D point cloud model to achieve 3D visual positioning of objects in the 3D point cloud and im-prove people's perception of position.The experimental results show that the positioning accuracy error of the algorithm is less than 7 cm in the X-axis motion direction,less than 10 cm in the Y-axis motion direction and less than 16 cm in the Z-axis motion direc-tion.Good spatial positioning effect is obtained.

关 键 词：空间定位 目标识别 最小二乘法 三维点云 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]