Activity recognition in planetary navigation field tests using classification algorithms applied to accelerometer...

Area of Science:

• Biomechanics
• Human-Computer Interaction
• Sensor Technology

Background:

• Accelerometer data offers insights into human activity across various applications.
• Field tests simulated astronaut tasks in reduced gravity to collect sensor data.
• Developing unobtrusive methods to monitor subject well-being is crucial.

Purpose of the Study:

• To apply classification algorithms to identify specific tasks from accelerometer data.
• To explore the potential of environmental sensors for gauging astronaut well-being.
• To classify six distinct leg-movement activities using sensor data.

Main Methods:

• Acquired single-accelerometer data from subjects performing simulated astronaut tasks.
• Utilized feature extraction from accelerometer data corresponding to specific activities.
• Applied and evaluated various classification algorithms, including k-Nearest Neighbors (kNN).

Main Results:

• The k-Nearest Neighbors (kNN) algorithm demonstrated superior performance in activity classification.
• Achieved an overall classification success rate of 90.8% for the identified leg movements.
• Successfully classified six different activities involving leg movement.

Conclusions:

• kNN is an effective algorithm for classifying leg movements from accelerometer data.
• Accelerometer-based activity recognition can contribute to unobtrusive well-being monitoring.
• This approach shows promise for applications in space exploration and terrestrial environments.