Classification of Skiing Techniques on Embedded System

Wearable technology has emerged as a key focus in both academic research and industry applications, especially in sports, where data gathering and analysis are crucial for enhancing athletic performance. This study proposes a 1D Convolutional Neural Network (1D-CNN) designed for real-time classification of skiing techniques using the Arduino Nano 33 BLE Sense, an embedded system that is both cost-effective and compact. The goal is to facilitate immediate feedback for athletes, aiding in performance enhancement and injury prevention. Time-series data from 6-axis accelerometer and gyroscope were collected to represent four skiing techniques: drifting, snowplow-steering, push-off, and tuck. Data was captured at 20 Hz from an amateur skier wearing the Arduino device on the chest. Model performance was assessed on a test set, reaching a good classification accuracy. Inference time and power consumption were measured considering TensorflowLite models deployed on the Arduino Nano 33 BLE Sense, both with and without 8-bit integer quantization for weights, showing promising results. This approach underscores the potential of using compact neural networks on embedded systems for sports analytics, promising further advancements in real-time, field-based athlete monitoring and feedback systems.