LD-ConGR

Ld-congr数据集包含10个手势类，其中三个是静态手势，七个是动态手势 (图1)。在LD-ConGR中总共收集了542个视频和44,887个手势实例。这些视频是从5个不同场景中的30个主题中收集的，并使用Kinect v4以第三视角捕获。每个视频都包含一个色彩流和一个深度流。两个流以30 fps同步记录，分辨率分别为1280x720和640x576。数据集按被试随机分为训练集和测试集 (训练对象23个，测试对象7个)。在图3中报告了手势实例的数量和手势持续时间的统计 (以帧为单位)。

The Ld-congr dataset comprises 10 gesture categories, including 3 static gestures and 7 dynamic gestures (see Figure 1). A total of 542 videos and 44,887 gesture instances were collected for the LD-ConGR dataset. These videos were captured from a third-person perspective using a Kinect v4 sensor, sourced from 30 subjects across 5 different scenarios. Each video contains both a color stream and a depth stream, which are synchronously recorded at 30 fps with resolutions of 1280×720 and 640×576, respectively. The dataset is randomly split into training and test sets based on subjects: 23 subjects are assigned to the training set, while 7 are allocated to the test set. Statistics on the number of gesture instances and gesture duration (in frames) are reported in Figure 3.

搜集汇总

背景与挑战

背景概述

LD-ConGR是一个用于长距离连续手势识别的大规模RGB-D视频数据集，包含10个手势类（3个静态和7个动态），共542个视频和44,887个手势实例。数据从30个主题在5个场景中采集，使用Kinect v4捕获色彩和深度流，帧率30 fps，分辨率分别为1280x720和640x576，并按被试分为训练集和测试集，适用于手势识别研究。