Comparative Analysis Of Volleyball Serve Action Based On Human Pose Estimation

National sports teams are paying more and more attention to combining the latest com-puter technology with traditional sports in order to improve the level of training and analysis of athletes.At present,in sports training for volleyball,it is still mainly dependent on the personal experience of the coach,the training cost is high,and the quality is difficult to maintain stable.In this regard,some domestic sports units have introduced professional software such as DataVol-ley to assist training,but they need to manually enter complex motion record descriptions? there are also research institutions that rely on human sensors to obtain motion parameters,but it will cause inconvenience to athletes’ activities.Serving is one of the basic and important technical movements of volleyball,and its standard movement is of great significance to the stable per-formance of the scene.Therefore,this article focuses on the method of evaluating the quality of volleyball serve with a single perspective.Based on the improved human pose estimation technology,this paper constructs a pose-based action evaluation index,and compares and analyzes the volleyball serve action based on this,and forms an automated method to compare and evaluate the quality of volleyball serve action,and focuses on the limbs in the single-view video Occlusion and motion blur issues.In this paper,the improved method of video human pose estimation based on stacked hourglass network can predict the inter-frame offset through multi-scale image features and inter-frame timing information,thereby optimizing the effect of pose estimation.Compared with the tradi-tional method based on optical flow,the method in this paper requires less parameter space and can more effectively improve the efficiency and accuracy of attitude estimation.In the realization of the method,the application of the method in this paper requires four steps: data collection,pose estimation and preprocessing,construction of evaluation indicators,and action comparison analysis.First,the data acquisition step is responsible for converting the original video image collected by the monocular camera into an OpenCV format image.After the pose estimation of the video image,pose preprocessing combines the shooting angle and other setting parameters to construct a coordinate system centered on the neck,and scale it.Then,the constructed coordinate system is used to calculate the angle index,speed index and other action evaluation indicators,so as to be used for the later action quality evaluation.Finally,the bidirectional dynamic warping algorithm is used to align the input action with the standard action,and the action evaluation index is compared and analyzed for the aligned action sequence,and the final action quality comparison analysis result is obtained.In this paper,the proposed method is applied to the actual shooting video,which can effec-tively locate the athlete’s pose without disturbing the athletes,and evaluate the quality of their movements,which can provide valuable analytical information support for coaches’ guidance.