转运机器人感知数据
收藏资源简介：
精准的环境感知能力是转运机器人实现安全运行的前提条件，而当前环境感知研究多依赖实验室模拟场景数据，与实际工业场景存在较大差异，导致算法在实际应用中检测精度与鲁棒性不足。其资源来源于同一涂装厂房南区场景，与前述定位数据集采集环境保持一致，确保了数据的场景真实性与关联性。采集设备为转运机器人搭载的万集单线激光雷达wj_719与Intel RealSense双目相机D455，两种传感器协同工作，可实现对环境的全方位感知。为模拟实际工业场景中的障碍物分布，在机器人运行轨道周边及对接区域布置了多种类型的障碍物，包括不同尺寸的工业工件、模拟作业人员的人形模型、转运托盘等，障碍物摆放位置涵盖机器人运行轨迹的侧面、前方及斜前方等关键区域，覆盖不同障碍物密度与尺寸工况。产生方法为通过ROS系统搭建多源感知数据同步采集框架，实时同步采集激光雷达的点云数据与双目相机的RGB图像、深度图像数据，采集频率均稳定在10Hz。采集完成后，利用PCL库对激光点云数据进行初步预处理，包括点云去畸变、地面点剔除、点云聚类等操作；利用OpenCV库对图像数据进行预处理，包括图像去噪、灰度化、阈值分割等操作。同时，组织专业技术人员对预处理后的感知数据进行人工标注，标注内容包括障碍物的类别、边界框位置、障碍物中心坐标等关键信息，标注精度误差≤0.5cm。主要内容包含激光雷达原始点云数据与预处理后点云数据、双目相机原始RGB图像与深度图像数据、预处理后的图像数据、障碍物人工标注信息、传感器内参与外参数据等。数据格式包含ROS标准.bag格式、.bmp格式，数据量为20.63GB，包含上千组不同障碍物场景下的感知数据序列。该数据集有效支撑了障碍物感知范围≥50m、检测距离误差≤0.2m的技术指标验证，为障碍物检测算法训练与优化、点云-图像配准技术研发、多源感知融合算法研究提供了覆盖不同障碍物密度、类型、尺寸的全面数据，显著提升了转运机器人在复杂工业环境中的安全运行能力与环境适应性。
Accurate environmental perception is a prerequisite for the safe operation of transfer robots. However, most current environmental perception studies rely on laboratory simulated scene data, which differs significantly from real industrial scenarios, resulting in insufficient detection accuracy and robustness of algorithms in practical applications. This dataset is sourced from the southern area of the same painting workshop, which is consistent with the acquisition environment of the previously mentioned positioning dataset, ensuring the authenticity and relevance of the data scenarios. The acquisition equipment includes the Wanji single-line LiDAR WJ-719 and Intel RealSense D455 binocular camera mounted on the transfer robot. The two sensors work collaboratively to achieve all-round perception of the environment. To simulate the obstacle distribution in real industrial scenarios, various types of obstacles are arranged around the robot's operating track and docking areas, including industrial workpieces of different sizes, humanoid models simulating workers, transfer pallets, etc. The placement positions of obstacles cover key areas such as the side, front, and oblique front of the robot's movement trajectory, covering working conditions with different obstacle densities and sizes. The data acquisition framework is built via the ROS system to achieve synchronous collection of multi-source perception data. Point cloud data from the LiDAR, RGB images and depth images from the binocular camera are collected synchronously in real time, with the acquisition frequency stabilized at 10 Hz. After data collection, the LiDAR point cloud data is preliminarily preprocessed using the PCL library, including operations such as point cloud distortion removal, ground point removal, and point cloud clustering; the image data is preprocessed using the OpenCV library, including operations such as image denoising, grayscale conversion, and threshold segmentation. Meanwhile, professional technicians are organized to manually annotate the preprocessed perception data. The annotation content includes key information such as obstacle categories, bounding box positions, and obstacle center coordinates, with the annotation accuracy error ≤ 0.5 cm. The main content of the dataset includes original point cloud data and preprocessed point cloud data from the LiDAR, original RGB images and depth image data from the binocular camera, preprocessed image data, manual annotation information of obstacles, sensor intrinsic and extrinsic parameter data, etc. The data formats include the ROS standard .bag format and .bmp format. The total data volume is 20.63 GB, containing thousands of sets of perception data sequences under different obstacle scenarios. This dataset effectively supports the verification of technical indicators such as obstacle perception range ≥50 m and detection distance error ≤0.2 m. It provides comprehensive data covering different obstacle densities, types, and sizes for the training and optimization of obstacle detection algorithms, the research and development of point-cloud-image registration technology, and the research of multi-source perception fusion algorithms, significantly improving the safe operation capability and environmental adaptability of transfer robots in complex industrial environments.
