WEAR：可穿戴运动活动识别数据集

一、简介 WEAR 是面向可穿戴与自我中心活动识别的户外运动数据集，发表于 IMWUT 期刊 2024 年相关刊期，专为视觉和惯性人类活动识别（HAR）研究构建；填补了同时包含自我中心视频与惯性传感器数据的户外运动数据集空白，由 22 名参与者在 11 个不同户外地点完成 18 种锻炼活动，同步采集惯性加速度数据与自我中心视频数据，两种模态数据具备互补的预测优势，适配户外复杂场景的活动识别研究

1. Introduction
The WEAR dataset is an outdoor sports dataset tailored for wearable and egocentric activity recognition research. Published in a 2024 issue of the *IMWUT* journal, it is specifically constructed for studies on visual and inertial human activity recognition (HAR). It fills the critical gap in existing outdoor sports datasets that simultaneously provide both egocentric video and inertial sensor data. A total of 18 exercise activities were performed by 22 participants across 11 distinct outdoor locations, with synchronized inertial acceleration data and egocentric video data collected during data acquisition. The two modalities feature complementary predictive strengths, making the dataset well-suited for activity recognition research in complex outdoor scenarios.

搜集汇总

背景与挑战

背景概述

WEAR数据集是一个面向可穿戴与自我中心活动识别的户外运动数据集，发表于2024年IMWUT期刊，填补了同时包含自我中心视频与惯性传感器数据的户外运动数据集空白。它由22名参与者在11个不同户外地点完成18种锻炼活动，同步采集惯性加速度和视频数据，两种模态数据互补，适用于户外复杂场景的人类活动识别研究，支持单模态/多模态模型训练和时间作用定位等算法开发。