A smoother joint trajectory estimation method based on motion capture marker data

The estimation of Center of Rotation (CoR) is one of crucial stages of automatic recovering of articulated skeleton from motion capture marker data. In computer character animation and human-computer interaction, the joint trajectory has to be smooth and the accuracy of joint position, in contrast, is not of great importance. In this paper, we present a new method for the joint position estimation by nonlinear optimization. Using geometric algebra provides a good initial point that will reduce the iterative times effectively and guaranties the solution converged to the global minimum. Testing results with real long sequences of motion capture data show that our method only needs a few times of iteration (3 times) and the trajectory is quite smooth without any other filtering. Besides, the method can be worked in real time and provides the joint position for each frame independently, so it can largely benefit from parallelization on multiple CPUs.