Sensor-Based Skill Assessment for Health and Fitness Applications

Exercising can contribute to a healthier and fuller life, however, correct and regular execution of the chosen exercises is of great importance. In the current project I investigated the possibilities offered by modern smartphones to enhance the effectiveness of training without a trained supervisor, through automated skill assessment. I implemented a realtime capable algorithm for the assessment of balancing board exercises, and integrated it into an Android application. I concentrated onto accelerometer and magnetometer sensors as means for measuring the orientation of the smartphone, which in turn was used for the analysis of motion patterns. Using a database of recorded exercises I compared the automated scores with scores given by a human expert. The computed score had a linear cross correlation coefficient of above 0.51 with the expert’s scores for dynamic and more than 0.76 for static exercises. On a 0 to 100 scale, the absolute difference of human and automatic scores was smaller than 10 points in more than 77% of the dynamic exercises and 89% of the static exercises.