In order to realize the accurate and efficient recognition of human motion gesture feature points, an infrared feature extraction and recognition method of human motion gesture based on similarity measurement is proposed. The model constraint method and CNN recognition training method are used to collect the key feature points of human motion. The human body model is constructed by optical markers, and the feature points are classified under the constraints of kinematics and inverse dynamics. A similarity measure regression model is established to determine the corresponding relationship of feature points, and the infrared recognition of human motion posture is realized by combining feature fusion and spatial clustering. The experimental results show that the average reconstruction integrity of the method is 98.03%, the average recognition time is 1.305 s, and the recognition success rate is as high as 97.25%, which shows that the method has the advantages of high efficiency and accuracy, and can effectively improve the application performance of human motion posture recognition, and provide reference for research and application in related fields.