Research On Motion Detection And Recognition Technology Based On Golf Video

In the field of sports-aided training such as machine vision,most of the target recognition algorithms have the characteristics of high time complexity,high hardware performance,complicated operation and so on.The auxiliary training equipments often do not have the function of machine recognition,needing manual operation,reducing the training efficiency.In the era of the prevalence of smart phones,if the computer vision technology applied to mobile phone applications,not only can help ordinary people with a lower cost way to assist in training,and will promote the popularization of sports.Therefore,it is very important to design a recognition algorithm running on the smart phone,which can identify and analyze the motion of the video sequences in real time and accurately,and make the training of the sports digitized.In this paper,the algorithm of golf pose recognition based on static image and video sequence is studied.In the aspect of golf image recognition in static image,this paper improves the most time-consuming feature extraction and classification recognition in the algorithm,selects the aggregation channel feature to describe the static golf image in the feature extraction process,and describes the multi-The multi-scale feature approximation method is adopted to improve the speed of feature extraction.In the classification and recognition phase of the algorithm,the improved training time is short and the generalization ability is used as the classifier.In the study of golf pose recognition in video sequences,machine learning techniques are used to identify the behaviors in the video sequence.The ACF detector is used to locate the salient regions in the image.The pose detector is used to scan the video sequences generated by the video sequences.Fractional sequences are used as the eigenvectors of the golf course in the video sequence,and the feature vector is reconstructed by using the threshold sliding window method.Then,the classifier is trained by the linear support vector machine,and the golf pose in the video sequence Real-time classification.The UCF motion dataset and self-created motion dataset are used to test the algorithm,and the analysis and verification are carried out on the PC i7-6700 HQ,8G and Matlab2016 b.The experimental results show that the average gesture recognition time of each frame is2.38 ms,which satisfies the requirements of the practical application.The golf poserecognition method based on the video sequence in the iphone5 s and The recognition speed of more than 30 fps and the recognition accuracy of 97% can be achieved when running in later versions,which further prove the validity of the method in practical application.