Orthopedics of bone-setting in traditional Chinese medicine (TCM) for the treatments of fractures has such characteristics as little pain, quick healing, and well recovery, which is the first choice for bone injury patients. However, the bone-setting manipulations in TCM are difficult to teach and learn, which hinders their inheritance and development. Inevitably, there will be noise data in the developed simulation and training system of bone-setting manipulations in TCM because of sensors’ qualities and transmission channels, which severely restricts the quantization, analysis, and evaluation of bone-setting manipulations in TCM. Aiming at the complexity of the bone-setting manipulations in TCM, the zeroth order, first order, and second order Kalman filtering (KF) combining velocity and acceleration vectors was proposed; meanwhile the principle of hyperparameter adjustment was determined through experiments. The proposed method is applied on the Australian Sign Language signage data set and the simulation and training system of bone-setting manipulations in TCM. The experiment results show that noise data are filtered and the details of the bone-setting manipulations in TCM are preserved well.