Samenvatting

Kinematical joint analysis during road cycling is a new trend. Reviewing the methods for 2D and 3D lower limb kinematical research, the sensor to segment alignment appeared to be the most challenging part due to unknown orientation with respect to the anatomical joints. A methodology that is suitable for future cycling kinematical research under match conditions, needs to be easy to use and fast. A method suggested by Seel et al. (2012) was found that uses a Newton/Gauss optimization protocol for sensor to segment alignment with arbitrary calibration motions. In this thesis only the methodology for the 2D hinge joint is examined, implemented and tested.
For the 2D joint kinematics, only the IMU’s gyroscope data was used. The accelerometer data was used for synchronization. Magnetometers were excluded due to the chance magnetic field disturbances. In total, nine measurement were performed using two IMU’s. Seven tests were used to test the alignment method using a self-made hinge joint with two segments. This joint-model was also used in the eighth test, which was filmed to test the accuracy of the angular rotation of a mechanical hinge joint. In the ninth and final test, angular joint rotations of a human knee were measured during treadmill cycling and compared to a video reference.
In all tests the Newton/Gauss optimization functioned accordingly and succeeded to find the joint axis direction. For the first six tests, an average of 15 iterations were needed for finding direction the joint axis. The angular rotation test of the mechanical joint showed a root mean squared error (RMSE) of 1.69o and the human knee from the final test a RMSE of 4.3o, which are slightly higher than the results from Seel. The only function from the method that did not reach the hypothesis was the automatic function to determine if the z-axes of the sensors point in the same global direction (sign of the joint axis).
The method shows promising results regarding the estimation of the joint axis direction and the measured angles compared to the video reference. The results may be improved by adding a Kalman filter to remove any drift and white noise and using an optitrack system to set as golden standard instead of 2D video. The function to determine the sign of the joint axis direction should be reviewed in future works.