The invention provides an anti-
noise 3D
hand motion estimation method applied to idiopathic tremor assessment, the core of the method is a
motion perception hierarchical grouping Mama (Motion-aware Hierarchical Grouping Mama, MHG-Mama) network, and a set of complete
hand tremor automatic assessment
system is integrally formed. According to the method, firstly, high-degree-of-freedom
finger joint movement and stable palm joint movement are separated through
feature extraction, and accurate finger posture
estimation is achieved; the method comprises the following steps of: firstly, acquiring a skeleton sequence, then combining with a layered space-time scanning mechanism in a selective
state space model (SSM) of Mama, capturing global motion features and retaining anatomical constraints, and obtaining a skeleton sequence which is smooth and reasonable in time; and finally, objective indexes are introduced based on the predicted skeleton sequence to quantify motion features, and a classifier is applied to perform accurate objective severity rating. Experiments show that the 3D hand posture
estimation precision is remarkably improved, the motion sequence
noise is reduced, the evaluation variability of clinicians can be reduced, and the method has very high
clinical value.