Smart-Toy and Smart-Bedsheet Dataset Based on Pressure Mapping Smart Textile

For details about the two datasets, please read ''introduction_to_the_two_datasets.pdf".
Based on the pressure mapping smart textile, we have developed two applications: Smart-Toy and Smart-Bedsheet, as shown in Fig. 1 and Fig. 3. For the Smart-Toy, we were interested in common interactions with this plush toy in life (20 actions in total as shown in Fig. 2) and invited 10 participants (5 females and 5 males, aged 23∼35) to create the Smart-Toy dataset, the details of which are presented in section 2. For the SmartBedsheet, we used it to identify common sleeping postures (11 postures in total as shown in Fig. 4), and we invited 13 participants (4 females and 9 males, the weight ranges from 42kg to 110kg, and the height ranges from 155cm to 188cm) to create the Smart-Bedsheet dataset, the details of which will be introduced in section 3.
Welcome to visit our website: http://pplab.ustc.edu.cn/. If you have any questions about the two datasets, please feel free to contact us ([e-mail])

Funding
Our work is supported ”the National Natural Science Foundation of China” (Grant No.62072420) and ”the Fundamental Research Funds for the Central Universities” (Grant No.2150110020).

1. Smart-Toy dataset
A Smart-Toy was created as a 3D application. The majority of the toy’s surface is covered with a 23 × 16 textile sensor matrix, with a spatial resolution of 1.5cm × 1.5cm, as shown in Fig. 1b. The pressure sensitive textile is sewn onto the inner side of the plush toy’s skin, flipped, and folded together with the skin into a cuboid. The pressure distribution is digitalized by on-chip 12-bits ADCs in the dsPIC33F and transmitted wirelessly via Bluetooth to a smart phone at 41 frames/second. Ten healthy adults (5 females and 5 males, aged 23∼35) were invited to create the play interaction dataset. They followed the instructions from the computer and carried out 20 play interactions (as shown in Fig. 2) in a random order (one session). Each participant finished 10 sessions. We manually label each instance by observing video recordings and pressure distribution changes.

2. Smart-Bedsheet dataset
A Smart-Bedsheet is covered with a 56 × 40 textile sensor matrix on the surface, with a spatial resolution of 3.2cm × 2.3cm, as shown in Fig. 3. The pressure distribution is digitalized by on-chip 12-bits ADCs in the STM32F303ZET6 and transmitted via USB to a laptop at 55 frames/second. Thirteen healthy adults (4 females and 9 males, the weight ranges from 42kg to 110kg, and the height ranges from 155cm to 188cm) were invited. They followed the instructions from the organizer, completed 11 sleeping postures (as shown in Fig. 4) in a random order, and each posture lasted for about 5 seconds. Each participant finished 10 sessions. Although each posture lasted 5 seconds in each session, these frames are extremely similar, so we propose to take a random frame as the instance of this posture.

如需了解这两个数据集的详细信息，请参阅《introduction_to_the_two_datasets.pdf》。
基于压力映射智能纺织品，我们开发了两款应用：智能玩具（Smart-Toy）与智能床单（Smart-Bedsheet），如图1、图3所示。针对智能玩具（Smart-Toy），我们聚焦于日常生活中与该毛绒玩具的常见交互（共20种动作，如图2所示），邀请了10名受试者（5男5女，年龄区间为23~35岁）构建智能玩具数据集，其详细信息将在第2节中介绍。针对智能床单（Smart-Bedsheet），我们利用其识别常见睡眠姿势（共11种，如图4所示），并邀请了13名受试者（4男9女，体重范围42kg~110kg，身高范围155cm~188cm）构建智能床单数据集，其详细信息将在第3节中介绍。
欢迎访问我们的官网：http://pplab.ustc.edu.cn/。若您对这两个数据集有任何疑问，欢迎随时联系我们（邮箱：[e-mail]）。

资助信息：本研究得到国家自然科学基金（项目编号：62072420）以及中央高校基本科研业务费专项资金（项目编号：2150110020）的支持。

1. 智能玩具数据集（Smart-Toy dataset）
本研究构建的智能玩具为一款3D应用。该玩具的大部分表面覆盖有23×16的纺织品传感器矩阵，空间分辨率为1.5cm×1.5cm，如图1b所示。压敏纺织品被缝制在毛绒玩具表皮内侧，经翻折后与表皮一同折叠为长方体结构。压力分布通过dsPIC33F芯片内置的12位模数转换器（ADC）进行数字化处理，并以41帧/秒的速率通过蓝牙无线传输至智能手机。我们邀请了10名健康成年人（5男5女，年龄区间为23~35岁）构建交互数据集。受试者按照电脑提示的指令，以随机顺序完成20种游戏交互动作（如图2所示），单次实验为一个回合。每名受试者完成10组实验。我们通过观察视频录制画面与压力分布变化，对每个样本进行人工标注。

2. 智能床单数据集（Smart-Bedsheet dataset）
智能床单的表面覆盖有56×40的纺织品传感器矩阵，空间分辨率为3.2cm×2.3cm，如图3所示。压力分布通过STM32F303ZET6芯片内置的12位模数转换器（ADC）进行数字化处理，并以55帧/秒的速率通过USB传输至笔记本电脑。我们邀请了13名健康成年人（4男9女，体重范围42kg~110kg，身高范围155cm~188cm）参与实验。受试者按照组织者的指令，以随机顺序完成11种睡眠姿势（如图4所示），每种姿势持续约5秒。每名受试者完成10组实验。尽管每组实验中每种姿势持续5秒，但对应的帧图像极为相似，因此我们提出选取随机一帧作为该姿势的样本。