Focusing on the problem of human movementidentification in the application of natural human-computer interaction, this paper summarizes the shortcomings of the traditional machine learning model in the identification of body movement.Based on the unique requirements of natural human-computer interaction application, it proposesRandom Increased Hybrid Learning Machine for human body movement identification. Combined with the Error Back Propagation Model, the Increased Extreme Learning Machine and Bidirectional Extreme Learning Machine, the model overcomes the shortcomings of traditional methods. This paper describes in detailthe algorithm theory, model rationality and implementation scheme of the Random Increased Hybrid Learning Machine. Finally, by comparing the experimental results, the paper verifies the Random Increased Hybrid Learning Machine's a better robustness, accuracy and timeliness in identification of human body movement.