MotionMiners Missplacement Dataset

The MotionMiners Miss-placement Dataset 𝑀𝑃1 is composed of recordings of seven subjects carrying out different activities in the intralogistics, using a sensor set-up of On-Body Devices (OBDs) for industrial applications. Here, the position and orientation of the OBD change with respect to the recording-and-usage guidelines. The OBDs are labeled with respect to their expected location on the human body, namely, 𝑂𝐵𝐷𝑅 , 𝑂𝐵𝐷𝐿 and 𝑂𝐵𝐷𝑇 on the right arm, left arm, and frontal torso. Tab. 1 (see manuscript) presents the different miss-placement classes of the dataset. This dataset considers the miss-placement as a classification problem; however, differently, the 𝑀𝑃 dataset considers rotations miss-placements—commonly appear on deployment from practitioners experience.

The 𝑀𝑃 dataset contains recordings of seven subjects performing six activities: Standing, Walking, Handling Centred, Handling Upwards, Handling Downwards, and an additional Synchronisation. Each subject carried out each activity under the case of up to 15 different miss-placement situations (soon updating to 20 different miss-placement situations), including a correct set-up of the devices. The 𝑀𝑃 dataset is divided in two subsets, 𝑀𝑃_A and 𝑀𝑃_B. 

Each recording of a subject contains:

• raw data of Acc, Gyr, and Mag in 3D for a certain number of samples, making a matrix of size [Samples times 27]
• annotated data of Acc, Gyr, and Mag in 3D for a certain number of samples, making a matrix of size [Samples, Act class, [27 channels]]
  • for MP_B, it includes the synchronized recording of the correct sensor set-up, so the matrix becomes  [Samples, class, [27 channels of the miss-placed setup], [27 channels of the correct set up]]
• the miss-placement annotations [Samples, Miss-placement class]
• the activity annotations [Samples, activity class, [19 semantic attributes]]
  • the semantic attributes are given following the following paper: "LARa: Creating a Dataset for Human Activity Recognition in Logistics Using Semantic Attributes", Sensors 2020, DOI: 10.3390/s20154083.

If you use this dataset for research, please cite the following paper: "Miss-placement Prediction of Multiple On-body Devices for Human Activity Recognition", Sensors 2020, DOI: 10.1145/3615834.3615838.

For any questions about the dataset, please contact Fernando Moya Rueda at fernando.moya@motionminers.com.