Multi-Activity Step Counting Algorithm Using Deep Learning Foot Flat Detection with an IMU Inside the Sole of a Shoe.

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Step counting devices are effective for athletic training and patient care.
• Previous algorithms achieved ~1% MAPE in simple walking conditions.
• Multi-activity step counting remains a challenge for wearable sensors.

Purpose of the Study:

• To develop and validate a deep learning step counting algorithm for instrumented running shoes.
• To assess algorithm performance in diverse multi-activity conditions.
• To evaluate the feasibility of the algorithm for embedded systems.

Main Methods:

• Instrumenting a running shoe with an inertial measurement unit (IMU).
• Collecting multi-activity data from 21 diverse participants.
• Developing a deep learning algorithm and validating using k-fold cross-validation.

Main Results:

• Step counts highly correlated with gyroscope and accelerometer norms, and vertical acceleration.
• Reducing input data to these three vectors minimally impacted performance.
• Achieved a Mean Absolute Percentage Error (MAPE) of 0.75% in multi-activity trials.

Conclusions:

• The deep learning algorithm demonstrates high accuracy in multi-activity step counting.
• The approach requires low computational resources, suitable for embedded devices.
• Validated the effectiveness of IMU-based step counting in dynamic, real-world scenarios.