Classifying simple human activity using synthetic accelerometer data

Identifying specific patterns of human activity is necessary when defining the interactions of people with assistive technologies. In environments where the user is an elderly person, the recognition of patterns becomes useful to analyze the possible responses of the user to certain scenarios where it is necessary to make use of the supporting technology. The participation of the users is necessary to learn from their daily habits. With the rise of ambient intelligence and assistive solutions collect information from both the user and the environment in real time to adapt and be able to learn from it. In this context, the current work proposes to learn activity patterns through training by using agent-based simulations. The objective of the training is to train avatars in 3D scenarios, which recreate those physical activities that the real user cannot perform due to their physical condition. The validation of the proposed approach was in a virtual living laboratory by using 3D simulations as assisted environment scenarios.