Realtime recognition of complex human daily activities using human motion and location data

Area of Science:

• Robotics
• Human-Computer Interaction
• Sensor Networks

Background:

• Daily activity recognition is crucial for robot-assisted living systems.
• Recognizing complex activities involving simultaneous body movements and hand gestures remains a challenge.
• Existing methods often lack efficiency or accuracy in real-world indoor environments.

Purpose of the Study:

• To propose an effective method for recognizing complex human daily activities in indoor environments.
• To develop a wireless power-aware motion sensor node for capturing human motion data.
• To model the spatio-temporal relationships between location, body activity, and hand gestures.

Main Methods:

• Development of a wireless power-aware motion sensor node with an orientation sensor, wireless communication, and power management.
• Attachment of three motion sensor nodes to the thigh, waist, and hand, complemented by an optical motion capture system for location data.
• Implementation of a three-level dynamic Bayesian network (DBN) to model activity constraints, utilizing a Bayesian filter and short-time Viterbi algorithm for efficient estimation.

Main Results:

• The proposed method demonstrated effectiveness and accuracy in recognizing complex daily activities.
• The dynamic Bayesian network successfully modeled the intricate relationships between location, body activity, and hand gestures.
• The use of Bayesian filtering and the Viterbi algorithm reduced computational complexity and memory usage.

Conclusions:

• The developed method provides an accurate and efficient solution for complex daily activity recognition.
• This approach has significant potential for enhancing robot-assisted living systems.
• The findings highlight the efficacy of integrating motion sensor data with advanced probabilistic models for human activity analysis.