Trainspotting: Combining Fast Features to Enable Detection on Resource-constrained Sensing Devices
Data Set Description
This data set contains more than 36 hours of continuous raw acceleration data from a sensor node that was deployed at the rails of Deutsche Bahn. Almost 250 trains of different type (such as ICE/IC, regional passenger trains, cargo trains, or locomotives) were rolling by, causing vibration patterns which were recorded by our sensor. The aim of this data set is to evaluate combinations of easy-to-compute features that can be used on a sensor node for on-line classification of these 'train events'.
- The data is recorded with the HedgeHog sensor, logging acceleration in 3 dimensions, sampled at 100 Hertz
- The acceleration samples are time-stamped, and are accompanied with ambient light sensor data
- The trains (the events of interest) are divided into
- fast inter-city passenger trains
- regional passenger trains
- local passenger trains ("city-hopper")
- cargo trains
- unknown: (unscheduled) trains that could not be annotated
The plots below shows the 3D acceleration data (36 hours) and close view of approximately 5 minutes with two events, namely a regional passenger and a cargo train passing by. The X axis displays the time during the day and the Y axis the acceleration values.
Download the data set
The data set (download) contains raw sensor data as well as the annotations stored in the Python numerical library data format (numpy) file. Please make sure to read the README.fst in the archive for details on loading and accessing data.
Citation / first paper
E. Berlin and K. Van Laerhoven, "Trainspotting: Combining Fast Features to Enable Detection on Resource-constrained Sensing Devices". In Proceedings of the Ninth International Conference on Networked Sensing Systems (INSS 2012), Antwerp, Belgium. IEEE Press. 2012. [link]
This data set is opened up to anyone interested in WSN deployments and applications to encourage reproducible results. Please cite our paper if you publish results on this data set, and consider making your own data sets open for anyone to download in a similar fashion. We would also be very interested to hear back from you if you use our data in any way and are happy to answer any questions or address any remarks related to it. You may use this data for scientific, non-commercial purposes, provided that you give credit to the owners when publishing any work based on this data.