Injuries are a sportsman’s worst nightmare: They withhold players from playing, they weaken teams and they force clubs to buy a bench-full of substitutes. On top of that, injuries can cause an awful lot of pain. Regulating match and training intensity can drastically decrease the
...
Injuries are a sportsman’s worst nightmare: They withhold players from playing, they weaken teams and they force clubs to buy a bench-full of substitutes. On top of that, injuries can cause an awful lot of pain. Regulating match and training intensity can drastically decrease the risk of getting injured. Furthermore, it can optimally prepare players for matchday. To enable player load regulation, accurate measurements of player positions, velocities and especially accelerations are required. To measure these quantities, JOHAN Sports develops a sports player motion tracking system. The device that is used for motion tracking contains a 9-DoF MEMS Inertial Measurement Unit (IMU) and a GPS receiver. These low-cost MEMS sensors are combined via sensor fusion. The challenge in this filtering problem lies in the limited and low-quality sensor measurements combined with the high-dynamic player motion consisting of rapid orientation changes and sensed impacts in every step. Being able to overcome these challenges paves the way for injury prevention, saving sports clubs, teams and players a lot of misery. To estimate player motion from the measurements by the tracking device, four sensor fusion algorithms are developed. For estimating rotational motion, the quaternion-based Unscented Kalman filter as described by Kraft and the Madgwick filter are devised. For estimating translational motion, a traditional linear Kalman filter and an Unscented Kalman filter are designed. These filters are combined to solve the sensor fusion problem. On simulated data, it is shown that the Madgwick filter outperforms Kraft’s quaternion-based unscented Kalman filter in both estimation accuracy and computational load. In estimating translational motion, the simulations show that the UKF and the linear Kalman filter achieve similar estimation accuracy. Subsequently the filters are tested on real data in different experiments. It is shown that, due to the on-chip filtering operation of the GPS sensor, position estimates do not benefit from sensor fusion. Furthermore, it is shown that the filters in their current state are unable to accurately estimate player accelerations from the sensors in the JOHAN Sports tracking device.