基于球面投影的激光点云目标检测  

Point cloud data processing and target recognition based on spherical projection

作　　者：李加定 万若楠 孙小广 邓磊[2] Li Jiading;Wan Ruonan;Sun Xiaoguang;Deng Lei(School of Electronic and Information Engineering,Guangzhou City University of Technology,Guangzhou 510800,China;School of Computer Science and Engineering,Central South University,Changsha 410083,China)

机构地区：[1]广州城市理工学院电子信息工程学院,广州510800 [2]中南大学计算机学院,长沙410083

出　　处：《电子测量技术》2024年第8期93-99,共7页Electronic Measurement Technology

基　　金：国家自然科学基金(62272490);广东省普通高校特色创新项目(KJQN201800606)资助。

摘　　要：基于深度学习的激光点云目标检测已成为了一个重要的研究领域。本文采用球面投影和2D图像的SOTA深度学习网络,实现3D激光点云目标快速检测。首先,将KITTI数据集单帧3D点云经球面投影转换成一帧2D的RGB三通道图像,像平面的像素位置取决于点云的三维坐标,其R、G、B 3个通道灰度值取决于点云归一化后的反射强度、距离、高度。其次,分析了不同分辨率下球面投影的重叠分布情况和对图像质量的技术影响。最后,采用语义分割模型DeepLab-V3+网络,仿真结果表明:该方法在分割精确度和速度方面都具有良好的性能,应用价值较高。Laser point cloud target detection based on deep learning has become an important research field.This article uses a SOTA deep learning network based on spherical projection and 2D images to achieve rapid detection of 3D laser point cloud targets.Firstly,a single frame 3D point cloud from the Semantic KITTI data set is transformed into a 2D RGB three channel image through spherical projection.The pixel position of the image plane depends on the three-dimensional coordinates of the point cloud,and the grayscale values of the R,G,and B channels depend on the normalized reflection intensity,distance,and height of the point cloud.Secondly,the overlapping distribution of spherical projections at different resolutions and their technical impact on image quality were analyzed.Finally,using the semantic segmentation model DeepLab-V3+network,simulation results show that this method has good performance in segmentation accuracy and speed,and has high application value.This paper presents a method of license plate character recognition based on the combination of Zernike moment and wavelet transformation features.

关 键 词：激光雷达 激光点云 三维目标检测 球面投影 

分 类 号：TN958.98[电子电信—信号与信息处理] TP391.4[电子电信—信息与通信工程]