VOID (Visual Odometry with Inertial and Depth)

数据集是使用Intel RealSense D435i摄像机收集的，该摄像机被配置为在400Hz下产生同步的加速度计和陀螺仪测量，以及在30Hz下产生同步的VGA大小 (640x480) RGB和深度流。深度帧是使用有源立体声获取的，并使用传感器工厂校准与RGB帧对齐。所有的测量都有时间戳。

数据集总共包含56个序列，包括室内和室外，具有挑战性的运动。典型场景包括教室，办公室，楼梯间，实验室和花园。在56个序列中，48个序列 (大约47k帧) 被指定用于训练，8个序列被指定用于测试，我们从中采样800个帧以构建测试集。每个序列将稀疏深度图集中在三个密度级别，1500，500和150点，对应于VGA大小的0.5%，0.15% 和0.05%。

This dataset was collected using an Intel RealSense D435i camera, which is configured to generate synchronized accelerometer and gyroscope measurements at 400Hz, as well as synchronized VGA-resolution (640×480) RGB and depth streams at 30Hz. Depth frames were acquired via active stereo and aligned with RGB frames using the sensor's factory calibration. All measurements are timestamped.

The dataset contains a total of 56 sequences covering indoor and outdoor environments with challenging motions. Typical scenarios include classrooms, offices, stairwells, laboratories, and gardens. Out of the 56 sequences, 48 (approximately 47,000 frames) are designated for training, while 8 are designated for testing, from which 800 frames are sampled to construct the test set. Each sequence provides sparse depth maps at three density levels: 1500, 500, and 150 points, corresponding to 0.5%, 0.15%, and 0.05% of the VGA-sized frame, respectively.

搜集汇总

背景与挑战

背景概述

VOID数据集由加州大学洛杉矶分校于2019年发布，使用Intel RealSense D435i摄像机采集，包含56个室内外序列的同步RGB、深度和惯性测量数据，主要用于视觉惯性里程计和深度补全任务。