EV-IMO

EV-IMO数据集是由马里兰大学高级计算机研究所创建的，旨在为事件相机提供室内场景中的运动分割学习方法。该数据集包含32分钟的室内记录，涉及多达3个快速移动的物体，并提供了精确的像素级运动掩码、自我运动和深度真值。数据集通过VICON运动捕捉系统跟踪物体和相机，并利用3D扫描技术获取深度图和像素级物体掩码的真值。EV-IMO数据集特别适用于场景约束的机器人应用，如自主移动机器人的导航和障碍物避免。The EV-IMO dataset was created by the University of Maryland Institute for Advanced Computer Studies, aiming to support learning-based motion segmentation methods for event cameras in indoor scenes. This dataset contains 32 minutes of indoor recordings involving up to three rapidly moving objects, and provides precise pixel-wise motion masks, ego-motion, and depth ground truth. The dataset tracks objects and the camera via the VICON motion capture system, and leverages 3D scanning technology to acquire ground truth for depth maps and pixel-wise object masks. The EV-IMO dataset is particularly suitable for scene-constrained robotic applications, such as navigation and obstacle avoidance of autonomous mobile robots.

搜集汇总

背景与挑战

背景概述

EVIMO是一个专注于运动分割和自运动估计的事件相机数据集，包含EVIMO和EVIMO2两个子集，提供6自由度姿态、物体掩模、深度图等多模态数据，并通过Vicon动作捕捉和3D扫描技术确保数据质量。该数据集适用于在极端黑暗环境下研究快速运动物体的视觉算法。