Study On The Identification Of Sports Personnel Number In Video

At present,sports are increasingly attracting people's interest and attention.Monitoring the video to analyze the performance of athletes in the event is the most intuitive way to evaluate athletes.Since sports videos are often photographed with medium and longdistance viewing angles,the poses of the athletes in the video are different,the face is blurred and the occlusion phenomenon occurs,which brings difficulties for the identification of the athletes.This paper studies the deep learning network to realize the detection and recognition of the number on the sports personnel's clothing in the sports scene,so as to achieve the purpose of analyzing the sports behavior of the sports personnel.Therefore,the research based on video efficient number detection and recognition has important theoretical and practical value.Since there is no public data set for such problems at present,this paper firstly prepares the data set of the identification of the athletes by semi-automatic labeling,which facilitates the comparison between the number identification algorithms.Secondly,the precise positioning algorithm of the number area is studied.It is optimized on the Faster R-CNN target detection framework.It optimizes the settings of the anchor box,the training details of the RPN layer,and the loss function,which improves the accuracy of positioning the number area.Finally,Based on the precise location of the number area,the number identification algorithm of the athletes is studied.The advantages and disadvantages of the current number identification methods are analyzed.The number of the athletes is rotated and difficult to be segmented.The number sequence identification network adapted to the affine transformation can effectively identify the number of the sportsman.