场景点云数据集

场景点云数据集主要面向机器人作业空间视觉目标重构与实时动态侵入检测的研究，针对其中的场景点云物体类别分割神经网络模型的训练数据需求，基于realsense d455深度相机对相应机器人作业场景的拍摄生成的点云，主要拍摄了机器人作业过程中行人进入或其他物体侵入时的场景点云数据，并对点云中各个点的实际物体所属类别进行标注从而形成数据集。同时对实际场景进行仿真模拟，产生仿真场景点云以补充实际场景数据稀少的缺陷。总体数据量达1.88GB。

This scene point cloud dataset is primarily developed to support research on visual target reconstruction and real-time dynamic intrusion detection within robot operation spaces. To fulfill the training data demands of neural network models for object category segmentation in scene point clouds, we collected point cloud data of corresponding robot operation scenarios using a Realsense D455 depth camera. The captured data mainly covers scenes where pedestrians enter or other objects intrude during robot operation processes, and each point in the point clouds is annotated with its corresponding actual object category to construct the dataset. Additionally, we performed simulations of the actual operational scenes to generate simulated scene point clouds, aiming to address the issue of insufficient real-world scene data. The total data volume of this dataset reaches 1.88 GB.

搜集汇总

背景与挑战

背景概述

场景点云数据集是一个面向机器人作业空间视觉目标重构与实时动态侵入检测研究的数据集，包含实际和仿真场景的点云数据，并进行了类别标注。数据集总数据量为1.88GB，由华南理工大学发布，属于国家重点研发计划项目的一部分。