Cross-position activity recognition

This directory contains the cross-position activity recognition datasets used in the following paper. Please consider citing this article if you want to use the datasets.

Jindong Wang, Yiqiang Chen, Lisha Hu, Xiaohui Peng, and Philip S. Yu. **Stratified Transfer Learning for Cross-domain Activity Recognition**. 2018 IEEE International Conference on Pervasive Computing and Communications (PerCom).

These datasets are secondly constructed based on three public datasets:
OPPORTUNITY (opp) [1], PAMAP2 (pamap2) [2], and UCI DSADS (dsads) [3].

------------------------------------------------------

Here are some useful information about this directory. Please feel free to contact [e-mail] for more information.

1. This is NOT the raw data, since I have performed feature extraction and normalized the features into [-1,1]. The code for feature extraction can be found in here: https://github.com/jindongwang/activityrecognition/tree/master/code. Currently, there are 27 features for a single sensor. There are 81 features for a body part. More information can be found in above PerCom-18 paper.

2. There are 4 .mat files corresponding to each dataset: dsads.mat for UCI DSADS, opp_hl.mat and opp_ll.mat for OPPORTUNITY, and pamap.mat for PAMAP2. Note that opp_hl and opp_loco denotes 'high-level' and 'locomotion' activities, respectively.
(1) dsads.mat: 9120 * 408. Columns 1~405 are features, listed in the order of 'Torso', 'Right Arm', 'Left Arm', 'Right Leg', and 'Left Leg'. Each position contains 81 columns of features. Columns 406~408 are labels. Column 406 is the activity sequence indicating the executing of activities (usually not used in experiments). Column 407 is the activity label (1~19). Column 408 denotes the person (1~8).
(2) opp_hl.mat and opp_loco.mat: Same as dsads.mat. But they contain more body parts: 'Back', 'Right Upper Arm', 'Right Lower Arm', 'Left Upper Arm', 'Left Lower Arm', 'Right Shoe (Foot)', and 'Left Shoe (Foot)'. Of course we did not use the data of both shoes in our paper. Column 460 is the activity label (please refer to OPPORTUNITY dataset to see the meaning of those activities). Column 461 is the activity drill (also check the dataset information). Column 462 denotes the person (1~4).
(3) pamap.mat: 7312 * 245. Columns 1~243 are features, listed in the order of 'Wrist', 'Chest', and 'Ankle'. Column 244 is the activity label. Column 245 denotes the person (1~9).

2. There are another 3 datasets with the prefix 'cross_', containing only 4 common classes of each dataset. This is for experimenting the cross-dataset activity recognition (see our PerCom-18 paper). The 4 common classes are lying, standing, walking, and sitting.
(1) cross_dsads.mat: 1920*406. Columns 1~405 are features. Column 406 is labels.
(2) cross_opp.mat: 5022*460. Columns 1~459 are features. Column 460 is labels.
(3) cross_pamap.mat: 3063 * 244. Columns 1~243 are features. Column 244 is labels.

-------- Original references for the 3 datasets:

[1] R. Chavarriaga, H. Sagha, A. Calatroni, S. T. Digumarti, G. Troster, ¨
J. d. R. Millan, and D. Roggen, “The opportunity challenge: A bench- ´
mark database for on-body sensor-based activity recognition,” Pattern
Recognition Letters, vol. 34, no. 15, pp. 2033–2042, 2013.

[2] A. Reiss and D. Stricker, “Introducing a new benchmarked dataset
for activity monitoring,” in Wearable Computers (ISWC), 2012 16th
International Symposium on. IEEE, 2012, pp. 108–109.

[3] B. Barshan and M. C. Yuksek, “Recognizing daily and sports activities ¨
in two open source machine learning environments using body-worn
sensor units,” The Computer Journal, vol. 57, no. 11, pp. 1649–1667,
2014.

本目录包含以下论文中使用的跨位置活动识别数据集。若您希望使用这些数据集，请考虑引用该论文。

王锦东、陈毅强、胡丽沙、彭晓辉和Yu Philip S.。**分层迁移学习在跨域活动识别中的应用**。2018年IEEE国际普适计算与通信会议（PerCom）。

这些数据集是基于以下三个公开数据集构建的：OPPORTUNITY（opp）[1]、PAMAP2（pamap2）[2]和UCI DSADS（dsads）[3]。
------------------------------------------------------
以下是关于此目录的一些有用信息。如需更多信息，请随时联系[e-mail]。

1. 这些数据并非原始数据，因为我已进行了特征提取，并将特征归一化到[-1,1]范围内。特征提取的代码可在以下链接找到：https://github.com/jindongwang/activityrecognition/tree/master/code。目前，单个传感器有27个特征。身体部位有81个特征。更多详细信息请参阅上述PerCom-18论文。

2. 每个数据集对应4个.mat文件：dsads.mat对应UCI DSADS，opp_hl.mat和opp_loco.mat对应OPPORTUNITY，pamap.mat对应PAMAP2。请注意，opp_hl和opp_loco分别表示'高级'和'运动'活动。
(1) dsads.mat：9120 * 408。列1~405为特征，按'躯干'、'右臂'、'左臂'、'右腿'和'左腿'的顺序列出。每个位置包含81列特征。列406~408为标签。列406是活动序列，指示活动的执行（通常在实验中不使用）。列407是活动标签（1~19）。列408表示人物（1~8）。
(2) opp_hl.mat和opp_loco.mat：与dsads.mat相同。但它们包含更多的身体部位：'背部'、'右上臂'、'右下臂'、'左上臂'、'左下臂'、'右鞋（脚）'和'左鞋（脚）'。当然，我们在论文中并未使用双鞋的数据。列460是活动标签（请参阅OPPORTUNITY数据集以了解这些活动的含义）。列461是活动练习（也请查阅数据集信息）。列462表示人物（1~4）。
(3) pamap.mat：7312 * 245。列1~243为特征，按'腕部'、'胸部'和'踝部'的顺序列出。列244是活动标签。列245表示人物（1~9）。

2. 另有3个数据集，以'cross_'为前缀，仅包含每个数据集的4个常见类别。这是为了进行跨数据集活动识别实验（参见我们的PerCom-18论文）。这4个常见类别是躺下、站立、行走和坐着。
(1) cross_dsads.mat：1920*406。列1~405为特征。列406为标签。
(2) cross_opp.mat：5022*460。列1~459为特征。列460为标签。
(3) cross_pamap.mat：3063 * 244。列1~243为特征。列244为标签。
-------- 3个数据集的原始参考文献：
[1] R. Chavarriaga, H. Sagha, A. Calatroni, S. T. Digumarti, G. Troster, J. d. R. Millan, and D. Roggen, “The opportunity challenge: A benchmark database for on-body sensor-based activity recognition,” Pattern Recognition Letters, vol. 34, no. 15, pp. 2033–2042, 2013。
[2] A. Reiss and D. Stricker, “Introducing a new benchmarked dataset for activity monitoring,” in Wearable Computers (ISWC), 2012 16th International Symposium on. IEEE, 2012, pp. 108–109。
[3] B. Barshan and M. C. Yuksek, “Recognizing daily and sports activities in two open source machine learning environments using body-worn sensor units,” The Computer Journal, vol. 57, no. 11, pp. 1649–1667, 2014。