Multioccupancy activity recognition based on deep learning models fusing UWB localization heatmaps and nearby-sensor interaction

Anguita-Molina, Miguel Ángel; Cardoso, Pedro; Rodrigues, Joao; Medina-Quero, Javier; Polo-Rodríguez, Aurora

http://hdl.handle.net/10400.1/27303

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
Multioccupancy_Activity_Recognition_Based_on_Deep_Learning_Models_Fusing_UWB_Localization_Heatmaps_and_Nearby-Sensor_Interaction.pdf		6.5 MB	Adobe PDF	Download

Send Feedback

Authors

Anguita-Molina, Miguel Ángel

Cardoso, Pedro

Rodrigues, Joao

Medina-Quero, Javier

Polo-Rodríguez, Aurora

Abstract(s)

Human activity recognition (HAR) focuses on developing systems and techniques to recognize and categorize human actions automatically based on sensor data. This study combines ultrawideband (UWB) technology and binary sensors to describe and recognize daily activities in real-world environments with multiple occupants, ensuring accurate user localization through noninvasive and privacy-respecting methods. A novel method that combines wearables with UWB technology, which allows the generation of heatmaps for accurate positioning, and binary sensors, which collect nearby interaction with daily activities in naturalistic conditions, is presented. A dataset composed of real-world data collected from three individuals in a real-life environment (house) was compiled. Advanced deep learning models are implemented to effectively fuse spatiotemporal information, leading to an encouraging performance in recognition of daily activities. The promising results suggest that this approach could be viable for large-scale deployments in future smart environments.