Concert Action Based On Gesture Recognition Research And Implementation Of Identification Problems

Human motion recognition covers a very wide range of fields and has extremely wide applications in many important fields.A large number of researchers have conducted in-depth research on human motion recognition,but for many problems that occur in human motion recognition,such as occlusion between people,human motion is not rigid body motion,and there will be scale changes.As well as the problem of subtle movements such as changes in perspective,many difficulties still exist in the task of human motion recognition.Therefore,how to develop efficient human motion recognition technology has become a topic of common concern to researchers.The recognition theme in human motion recognition is human,and human is the main body of the society.Most of the video is human video.Therefore,how to automatically recognize human motion in the video has become an important and meaningful task.The process of human action recognition is generally divided into three parts,including feature extraction,feature processing,and learning algorithm.According to the different types of human motion recognition,there are two types of video and image recognition.Video-based motion recognition is detected by motion sequences,and the method of using image recognition is only by recognizing the motion information in a single image to achieve recognition.In the current stage of research,there are many researches based on video in the action in the image.The action recognition for images,because the static images provide less information,so compared to video-based action recognition,it brings greater difficulties to recognition.In addition,image-based recognition generally has more complicated scenes,including There are many objects,and there may be problems such as occlusion of the human body,so it is difficult to identify.The exciting highlights of the concert require the editor to manually find the motion features,and then manually set the video start and end points for manual editing.This kind of work is very easy to cause the omission of detection.The concert scene is a very complex scene.The scene is complex and diverse,and there are many people in the scene,which makes it difficult to recognize.Therefore,this paper proposes a method of motion detection that fuses skeleton key information and residual blocks.This method combines The residual block extracts the global motion information of the image,can automatically recognize the wonderful motion in the concert,and locate the position of the human body in the image where a certain motion occurs.Combined with the characteristics of joint points of common human actions,three characteristics of joint point angle relationship,limb length ratio and relative position of joint points were selected,and the three characteristics were normalized to classify the actions in the scene,and a kind of active design was designed.The learned concert action recognition model has important practical significance for studying the singer's actions in concert scenes and the dance actions that often appear.Finally,on the pytorch deep learning framework platform,I tested the concert data set I collected,analyzed the results of action recognition,the causes of misdetection and missed detection,and proposed improved methods.Experiments show that the method of fusion multi-feature recognition action based on bone points proposed in this paper achieves an accuracy rate of 92.5%on the data set.Therefore,the method of fusing bone and residual block proposed in this paper has practical feasibility and important significance.