Feature extraction and classification of skiing/snowboarding jumps with an integrated head-mounted sensor

With advancing technology in miniature MEMS sensors, wearable devices are becoming increasingly popular, facilitating convenient activity detection. One particular application is in sports performance monitoring. This thesis presents novel real-time jump detection and classification algorithms in skiing and snowboarding using a head-mounted MEMS-based inertial measurement unit (MEMS-IMU), which is integrated with a barometric pressure sensor. The key performance variables of the jump are extracted and evaluated for training and/or entertainment purposes. In contrast to the existing jump detection algorithms based on acceleration signals, the proposed algorithm uses vertical velocity and air time in addition to acceleration in the vertical direction. A support vector machine (SVM) is applied to generate a classification model. The jumps are classified into four different groups – Ollie, Standard, Drops, and Step up jumps. The experimental results show that by incorporating the velocity and air time into the detection algorithm, the sensitivity and specificity increase dramatically to 92% and 93%, respectively. In addition, the proposed classification model achieved 80.5% accuracy.