A motion recognition technique based on linear matrix representation to improve Parkinson’s disease treatments
Pierre Moreau, David G. Durand, Jérôme Bosche, Michel Lefranc · 2021
The analysis of certain pathological movements of Parkinson’s disease often allows the neurologist to establish a precise diagnosis and to propose an adapted treatment for the patient. This work considers a motion capture device with embedded sensors such as inertial measurement unit without using camera. This paper presents a movement modeling method in a matrix representation by exploiting the Ordinary Least Squares approach. The quality and simplicity of the model allow movement recognition in real time. This method was tested as part of an experiment including 5 movements, each performed by 14 people wearing the motion capture device (Perception Neuron Pro suit). The global accuracy involved by this method is about 92.85%.