La Trobe University researchers develop AI to track cycling performance via smartphone video

Researchers at La Trobe University’s Holsworth Biomedical Research Centre are developing artificial intelligence models designed to estimate cycling performance data using only smartphone video. The technology aims to provide cyclists with sophisticated biomechanical information, such as the forces applied to pedals, which typically requires expensive specialized hardware like instrumented pedals and heart-rate sensors.

The system utilizes deep learning to establish a connection between a rider's visible movements and the actual forces exerted during pedal strokes. By training the model on synchronized laboratory recordings—comparing video of a rider in motion with real-time force measurements—the AI learns to predict these forces from visual data alone.

Preliminary research involving researcher Rodrigo Bini has shown promise, using recurrent neural networks to predict three-dimensional pedal forces and power at lower-limb joints. Previous studies reported correlations between measured and predicted variables ranging from 0.79 to 0.96.