Research On Action Evaluation Method Based On Kinect

In the field of machine vision and image processing,the identification of human pose has become an important topic,and has been widely used in human-computer interaction,virtual reality and intelligent monitoring.However,there are still many problems are not well resolved,affecting the machine’s understanding of human behavior.Ordinary cameras can only get two-dimensional images,but the reconstruction of two-dimensional information to three-dimensional information will lose a lot of important data,affecting the accuracy of action recognition.although the researchers designed a lot of image processing algorithms,but still can not avoid the impact of light,texture and other effects.The emergence of the Kinect sensor provides the researcher with a new solution to the problem.It can obtain the depth image with spatial distance information through a pair of infrared cameras,and extract the three-dimensional coordinate information on the basis of the depth image.but Kinect did not give a high-level function of posture recognition,Due to changes in human movements,it is difficult to build a common model to identify.In order to make the action recognition have better effect,this paper uses the Kinect sensor to obtain the three-dimensional data of 20 skeletal coordinates of the human body,and according to the characteristics of human behavior,the relative distance and angle of the human joint are taken as the characteristic parameters.In the recognition of static posture,the sample set is trained by the characteristic parameters,and the posture is identified by using the KNN algorithm.In the evaluation of the motion,after analyzing the time characteristics of the motion feature sequence,the linear regression method is used to train the sample curve.A least squares method is used to fit out an optimal angle curve as a template.Because the curve length is different,the DTW algorithm is used to match the curves of different lengths,and a set of formulas is defined creatively to evaluate the action,the joint angle curve DTW difference is used as the experimental parameters.In this paper,the relative distance and the angle of joint point are taken as the feasibility of action recognition.This paper chooses the four types of tai chi as the static posture recognition object.the experimental results show that the recognition rate can be very high by the method.Based on the analysis of the experimental data,the paper summarizes the distribution rules of DTW difference sample points of 8 angles,defines a set of formulas to evaluate the action,and designs the action evaluation system to demonstrate the rationality of the evaluation method.Since the cardinality and the factor in the evaluation formula are recalculated with the transformation of the action,the complexity of the evaluation method is increased.The next work is to perfect the parameters of the action evaluation formula to make it more efficient and adaptable.