I+ Lab RadSet v1
收藏资源简介：
The existing public datasets often suffer from small data volumes, leading to insufficient training processes that result in severe overfitting and poor generalization performance. To address this issue, a radar dataset named RadSet is constructed. During the data acquisition phase, frequency modulated continuous wave (FMCW) radar system IWR1843 Boost manufactured by Texas Instruments (TI) was used. The RadSet dataset is collected by I+ Lab at Shandong University, covering a rectangular area 5 meters long in front of the radar and 4 meters wide, with the radar placed at a height of 120 cm above the ground. The volunteers execute the aforementioned activities at distances ranging between 1 to 5 meters from the radar. The RadSet captures six types of daily activities, including 1) bending; 2) fall; 3) jumping; 4) punching; 5) jogging; and 6) walking, performed by 10 volunteers. Bending refers to lowering the body to pick up an object from the ground while remaining in the same position. Fall describes a sudden loss of balance, resulting in a fall during walking or standing. Jumping involves propelling the body off the ground in place. Punching refers to throwing a punch within the entire radar field of view. Jogging refers to running at a moderate speed without a fixed direction. Walking refers to moving at a steady pace without a specific direction.
现有公开数据集普遍存在数据体量偏小的问题，易导致训练不充分，进而引发严重过拟合与泛化性能欠佳的缺陷。为解决该问题，本研究构建了一款名为RadSet的雷达数据集。在数据采集阶段，采用了德州仪器（Texas Instruments, TI）生产的调频连续波（Frequency Modulated Continuous Wave, FMCW）雷达系统IWR1843 Boost。RadSet数据集由山东大学I+实验室采集，采集范围为雷达前方长5米、宽4米的矩形区域，雷达架设高度为距地面120厘米。志愿者在距雷达1至5米的范围内开展上述六类日常活动。RadSet数据集共采集了10名志愿者完成的6类日常活动，具体包括：1）弯腰拾取（Bending）：保持原地不动，俯身从地面拾取物品；2）摔倒（Fall）：行走或站立过程中突发失衡并摔倒；3）原地跳跃（Jumping）：使身体腾空离地的动作；4）出拳（Punching）：在雷达全视场内完成挥拳动作；5）慢跑（Jogging）：以适中速度跑动且无固定方向；6）行走（Walking）：以稳定步速移动且无特定方向。
