EgaitSegmentationValidation2014 - Original Labels#
Further Links:
A challenge for stride segmentation from continuous IMU data focused on straight strides.
General Information#
- Dataset
The Egait Segmentation Validation 2014 dataset [1] is used for this challenge (usage example, download). It contains 4x10m gait tests from 30 participants (10 controls, 10 Parkinson’s disease, 10 geriatric patients) and 15 simulated @home tests from 5 participants per cohort (5 controls, 5 Parkinson’s disease, 5 geriatric patients ).
- Sensor System
All participants wore Shimmer2R (102.4 Hz) IMUs laterally on both shoes.
- Reference System
All trials where recorded with a video camera to support labeling of the strides in the raw data. The camera data was used to remove turns and stair strides from the annotations. Another set of annotations exists, that contains all strides including turns and stairs. All annotations where performed following the stride definition by [1]. However, this challenge used the original annotations that excluded turns and stairs. For an equivalent challenge using the new annotations, see
egait_segmentation_validation_2014.
Implementation Recommendations#
This challenge version uses the original annotations that excluded turns and stairs. In result, the validation favors algorithms that are very specific to straight strides. Algorithms that naturally have a higher sensitivity and also detect turns and stairs will be penalized.
Overall the data only contains gait and no other movements. This should make this a relatively easy dataset for gait segmentation.
References#
Barth, Jens, Cäcilia Oberndorfer, Cristian Pasluosta, Samuel Schülein, Heiko Gassner, Samuel Reinfelder, Patrick Kugler, et al. “Stride Segmentation during Free Walk Movements Using Multi-Dimensional Subsequence Dynamic Time Warping on Inertial Sensor Data.” Sensors (Switzerland) 15, no. 3 (March 17, 2015): 6419-40. https://doi.org/10.3390/s150306419.