Research On Human Posture Recognition Based On KinectV2

Human body gesture recognition is one of the hot topics in computer vision research,and gesture recognition algorithms provide technical support for the development of humancomputer interaction,virtual reality and other fields.The time-varying nature of human motion and the complex variability of motion scenes lead to unsatisfactory real-time,accuracy and robustness of human body gesture recognition.Therefore,real-time and accurate capture and recognition of human posture has attracted a lot of attention in the field of vision.Aiming at the problems in the feature extraction and recognition of human posture,this paper studies the construction of human posture model and posture recognition based on the KinectV2 platform.The main contents of this article are as follows:First of all,this paper obtains the three-dimensional information of the human body's marked joint points based on the KinectV2 skeleton tracking technology,and calibrates the obtained joint point information;then the angle and distance feature construction of the used joint points is carried out,and then the feature information of the posture model is converted.The pose feature representation efficiently utilizes bone information and effectively reduces data redundancy,so that the pose feature representation has the advantages of high combinability and low computational complexity.Then,aiming at the defects of undefined posture interference in static human posture recognition,a model network matching algorithm based on small sample learning of human posture joint angle and distance characteristics is proposed.First,the sample data set is trained and classified,and the model from one data end to the other is trained;then the small sample learning network matching classifier is used to calculate the similarity between the trained data set and the test data set,and the parameter conversion is performed for weighting.And calculation;Finally,the template type of the sample with the highest degree of fit is selected as the recognition result,thereby solving the interference of the undefined posture.The experimental results show that the human body gesture recognition based on this algorithm has the advantages of high accuracy and strong real-time performance.Finally,in view of the problems of key frame gesture recognition in dynamic gesture sequences,a multi-resolution feature gesture model recognition method based on the fusion of Gabor wavelet and Curvelet double transformation is proposed.First,the fuzzy logic model is used to select the key frame poses in the dynamic pose sequence;second,the Gabor wavelet transform is used to process the texture features of the obtained key frame poses to obtain feature vectors in different directions and scales,and then the Curvelet transform is used to calculate the features in a specific direction The information obtains the feature vector;then the two feature vectors are merged to obtain the final feature representation of the key frame pose,and a large number of pose sequences are processed in this way to construct a multi-resolution feature pose model;finally,the constructed pose model is matched and recognized by a model matching algorithm.The key frame posture is selected through the fuzzy logic model to avoid the interference of redundant frame sets.Gabor wavelet transform and Curvelet transform can effectively extract the feature information of the key frame posture and improve the recognition accuracy.Experimental results show that the multi-resolution feature pose model constructed by this method has high recognition accuracy.