An in-depth understanding of hand gestures during accordion playing is crucial to the skillful mastery of accordion playing. This paper focuses on constructing a three-dimensional model of hand posture when playing accordion using algorithms, and utilizing motion capture technology to capture and classify hand posture when playing accordion. The coordinates of the two-bit joint points of the hand are estimated from the image by a neural network, and the hand mask is obtained using the threshold segmentation method. The predefined hand template mesh is rigidly deformed using a linear hybrid mask to fit the 2D joint points in the view, and finally, the personalized neutral hand model is constructed. According to the pose information solution method, the hand pose information during accordion playing is calculated, and the position information of the hand is calculated by combining it with the Earth reference coordinate system. Based on the forward kinematics equation of the human hand, the hand joint motion projection algorithm suitable for hand posture capture is designed, and classifiers such as support vector machine are integrated to realize the motion capture and classification of hand posture. The experimental results show that the three-dimensional model construction of hand posture and the motion capture and classification algorithm adopted in this paper can effectively realize the modeling and classification of accordion playing hand posture and provide a strong technical support for intuitive and clear mastery of accordion playing skills.