TUM monoVO

TUM monoVO 是用于评估单目视觉里程计 (VO) 和 SLAM 方法的跟踪精度的数据集。它包含 50 个真实世界序列，包含超过 100 分钟的视频，记录在不同的环境中——从狭窄的室内走廊到宽阔的室外场景。所有序列主要包含探索相机运动，在同一位置开始和结束：这允许通过从开始到结束的累积漂移来评估跟踪精度，而不需要完整序列的真实情况。与现有数据集相比，所有序列都经过光度校准：数据集创建者提供传感器报告的每一帧的曝光时间、相机响应函数和镜头衰减因子（渐晕）。

TUM monoVO is a dataset dedicated to evaluating the tracking accuracy of monocular visual odometry (VO) and SLAM methods. It comprises 50 real-world sequences with over 100 minutes of video footage, recorded across diverse environments ranging from narrow indoor corridors to expansive outdoor scenes. All sequences primarily feature exploratory camera motions, starting and ending at the same location: this allows for the assessment of tracking accuracy via the cumulative drift from start to finish, without requiring ground truth for the full sequence. Compared to existing datasets, all sequences have been photometrically calibrated: the dataset creators provide the exposure time, camera response function, and lens attenuation factor (vignetting) for each frame as reported by the sensor.

搜集汇总

背景与挑战

背景概述

TUM monoVO是一个用于评估单目视觉里程计和SLAM方法跟踪精度的数据集，包含50个真实世界序列（超过100分钟视频），覆盖室内外多种环境。其关键特点是所有序列都经过光度校准，提供每帧的曝光时间、相机响应函数等参数，便于精确分析。该数据集由慕尼黑工业大学于2016年发布，适用于计算机视觉研究。