Reconstructing human movements from inertial measurement units

1. Write some code to read in one of those CSV files and pull out the required quaternions for each IMU and also each model body segment. 2. Write some code to implement quaternion regression (using Rivest & Chang’s algorithm or another of your choice). 3. Choose some U and V, e.g. IMU1 and ModelThorax. 4. Fit regression coefficients (e.g. A and B in Vhat = AUB). 5. Compare the residual error [e.g. E = (Vhat*)V] where Vhat* is the conjugate of Vhat. Here V is the true orientation and Vhat is the estimated rotation from the regression model. The residual at each timestamp is a quaternion, but you can convert it to an amount in degrees by converting to the angle-axis representation and keeping the angle part. Then you can average over all time to get a mean error. 6. Check the accuracy on a testing set. For example you might train your regression model on trails 1-3 for a participant but then test on trials 4-6. Alternatively you might train on some people and test on other people. 7. Report which sensor (which choice of U) gives the best accuracy for each model segment (each choice of V). [login to view URL] This link has the data and that is the procedure for now to follow. The picture is my project title and outline. If you can understand fine otherwise we can have broad meeting so that i can explain