Implementation of mocap solver algorithm on motion captures to improve accuracy and performance

Abstract

One of the main issues that Motion Capture systems face is to accurately note all markers within the 3D environment. While the accuracy varies according to the system used, there are algorithms that use Deep Learning techniques to place and fix any offset that a Motion Capture may have, according to the training data given. In this Thesis, I will test the MoCap Solver algorithm on the DanceDB Motion Captures in order to fix the misplaced markers. The results show that the animations can be fixed but a better training set for the neural network is required for better results