tags:
- time series
- time series classification
- monster
- HAR
license: other
pretty_name: Skoda
size_categories:
- 10K<n<100K
Part of MONSTER: https://arxiv.org/abs/2502.15122.
| STEW | |
|---|---|
| Category | HAR |
| Num. Examples | 14,177 |
| Num. Channels | 60 |
| Length | 100 |
| Sampling Freq. | 98 Hz |
| Num. Classes | 11 |
| License | Other |
| Citations | [1] |
Skoda captures 10 specific manipulative gestures performed in a car maintenance scenario [1]. Its purpose is to investigate different aspects related to the gestures, such as fault resilience, performance scalability with the number of sensors, and power performance management. The dataset comprises 10 classes of manipulative gestures, which were recorded using 2 x 10 USB 3D acceleration sensors positioned on the left and right upper and lower arm. The sensors have a high sample rate of approximately 98 Hz, ensuring precise capturing of the movements.
In terms of activities, the dataset includes 10 distinct manipulative gestures commonly performed during car maintenance. The data was collected from a single subject, with each gesture being recorded 70 times. In total, the dataset offers around 3 hours of recording time, enabling thorough analysis of the gestures in various scenarios. The processed dataset consists of 14,117 time series each of length 100 (i.e., representing approximately one second of data per time series at 98 Hz).
[1] Piero Zappi, Daniel Roggen, Elisabetta Farella, Gerhard Tröster, Luca Benini. (2012). Network-level power-performance trade-off in wearable activity recognition: A dynamic sensor selection approach. ACM Transactions on Embedded Computing Systems (TECS), 11(3):1–30.