Detailed information on the occurrence and duration of human activities is crucial to enhance the efficiency of manual processes. Thus, methods of sensor-based human activity recognition (HAR) gain relevance. Training a classifier for this task demands a large amount of data, as human movements are highly variable and diverse, in particular in the diverse environments of industrial labor.
This presentation will therefore discuss the issue of dataset creation for HAR. It is crucial to gather data in such a way that a classifier may generalize among industrial scenarios, deviating physical characteristics of the humans, the sensor placement and configuration, etc. to allow for transfer learning. Additionally, experiences from the practical application of HAR methods in the industry will be discussed.
|Keywords||Human Activity Recognition, Dataset Creation, Logistics, Transfer Learning|