A computer vision model is constructed to recognize the expressions and movements of musical performers so as to analyze the changes in the forms and techniques of musical performances under the constructivist theory. The model needs to perform pre-processing, such as grayscale conversion, thresholding, background subtraction, noise reduction, and so on, on the related video before recognizing expressions and actions. The performer’s movements are recognized based on TC-HGCN, while the expressions are recognized based on CSACNN. The average accuracy of TC-HGCN in recognizing dance movements in musicals reaches 87.95%. The facial expression recognition method based on CSACNN can overcome the shortcomings of commonly used expression recognition models. The use of computer vision modeling analyzes the constructivist theory on the form and technique of the next musical performance, which enhances the performance effect and has more infectious and artistic charm.