扫地机器人L型墙+高反物品场景测试数据

扫地机器人在工作时时常会遇到一些特殊场景，比如L型墙场景且有高反物品存在的场景，基于视觉的扫地机在此场景中获取的深度数据常有边界不清晰且多径干扰强等组合负责问题，如何能在此场景中准确定位与避障是个难点也是重点，在此场景，扫地机深度视觉模组获取的深度数据的精度就显得十分重要。搭建扫地机真值系统，用真值相机和目标深度模组进行不同材质、光线、距离下L型墙+高反物品场景的数据采集，获取L形墙+高反物品场景的数据，并对数据进行处理得到校正后的数据用于目标深度模组的校正，提升深度模组在扫地机L型墙+高反物品场景下的测量精度，以解决扫地机定位与避障问题。采扫地机器人真值系统采集不同材质、光线、距离下的L型墙+高反物品场景的数据，包括原始深度相机数据、真值相机采集的点云数据；对原始深度相机数据进行系统误差去除；对真值相机采集的L型墙+高反物品场景点云数据进行修复，通过平面拟合算法进行L型墙+高反物品场景的墙面校正、再通过滤波算法对点云进行平滑去噪，得到修复后的真值相机点云；将修复后的真值相机点云与深度相机进行点云配准，得到校正后的L型墙+高反物品场景的点云数据——校正后数据laser-x、laser-y、laser-z和校正后数据tof-x、tof-y、tof-z。

Floor sweeping robots frequently encounter special scenarios during operation, such as the L-shaped wall scenario with high-reflectivity objects. For vision-based sweeping robots, the depth data acquired in this scenario often suffer from combined complex issues including blurry boundaries and strong multipath interference. Achieving accurate positioning and obstacle avoidance in this scenario is both a key challenge and research priority, where the accuracy of depth data collected by the robot's depth vision module is extremely critical.
To address this problem, a ground-truth system for floor sweeping robots was constructed. The system collected data under the L-shaped wall + high-reflectivity objects scenario with different materials, lighting conditions and distances, using a ground-truth camera and the target depth module. The collected data includes raw depth camera data and point cloud data captured by the ground-truth camera. Then, the collected data was processed to obtain calibrated data for calibrating the target depth module, so as to improve the measurement accuracy of the depth module in the L-shaped wall + high-reflectivity objects scenario and solve the positioning and obstacle avoidance problems of the sweeping robot.
The ground-truth system collects data of the L-shaped wall + high-reflectivity objects scenario under different materials, lighting conditions and distances, including raw depth camera data and point cloud data acquired by the ground-truth camera. First, systematic errors were removed from the raw depth camera data. Then, the point cloud data collected by the ground-truth camera was repaired: wall calibration for the L-shaped wall + high-reflectivity objects scenario was performed via the plane fitting algorithm, and then the point cloud was smoothed and denoised using the filtering algorithm to obtain the repaired ground-truth camera point cloud. Subsequently, point cloud registration was conducted between the repaired ground-truth camera point cloud and the depth camera data to obtain calibrated point cloud data of the L-shaped wall + high-reflectivity objects scenario, including calibrated data laser-x, laser-y, laser-z and tof-x, tof-y, tof-z.

搜集汇总

数据集介绍

特点

该数据集为扫地机器人在L型墙和高反物品场景下的测试数据，包含1122条记录，每年更新一次，旨在通过真值系统和算法处理提升深度模组的测量精度，解决复杂环境中的定位与避障问题。