Exploring the Feasibility of Classifying Fundamental Locomotor Skills Using an Instrumented Insole and Machine Learning Techniques

• Published in: Digital Human Modeling and Applications in Health, Safety, Ergonomics and Risk Management. Human Body and Motion pp. 113 - 127
• Main Authors: Ajisafe, Toyin, Um, Dugan
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2019
• Series: Lecture Notes in Computer Science
• Subjects: Foot pressure; Locomotor skills; Machine learning; Movement classification; Neural network
• ISBN: 3030222152; 9783030222154
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-22216-1_9

Movement interventions commonly feature Fundamental locomotor skills (FLSs) like hopping. These skills are thought to positively shape physical activity (PA) trajectory in children. However, the extent to which children who are at risk for overweight and obesity deploy these skills during leisure time PA is often unknown. Direct observation methods are cost-prohibitive. Step count from commercial activity trackers fail to capture these movements. This paper explored the feasibility of using an instrumented insole and machine learning algorithms to classify hopping, running, sprinting, and walking. A subject (age: 40 years; mass: 81 kg; height: 1.7 m) walked, hopped, and sprinted while wearing an instrumented insole. The insole features two pressure sensors and a mechanical housing. The mechanical housing held an Arduino/Genuino 101 programmed using Arduino Software Integrated Development Environment (IDE). An artificial neural network (ANN) training and real-time classification software was written in Arduino IDE and downloaded onto the Arduino 101’s non-volatile memory. The ANN used pressure and time derivative data from a dual sensor array and was tested with various statistical parameters. The moving average, standard deviation, min, max, time derivative and acceleration proved most significant for effective training and precision realtime classification. The on-line validation produced mixed accuracy: walking, running, and sprinting were classified with higher than 70% accuracy, but hopping was classified with only 25% accuracy. It is concluded that insole instrumentation with supervised machine learning seem promising to help track FLS deployment. The lower classification accuracy associated with hopping may be due to higher signal variability.