Research On The Key Technologies Of Automatic Broadcasting For Panoramic Football Videos

Football is known as "the world’s first sport" with the largest audience in the world.Moreover,the broadcasting video is the leading way to spread football matches.Generally speaking,the broadcasting of a football match is produced by multiple cameras,staff,and several professional broadcasters arranged around the venue.This kind of broadcasting method brings people a high-quality viewing experience,However,at the same time,the required broadcasting cost is very high,which not only requires a lot of labor and material resources but also has high requirements for the professional quality and ability of the broadcasting personnel.However,small and medium-sized matches such as non-professional leagues,campus football matches,etc.,have high frequency,small scale,and limited expenses,they obviously cannot afford the high costs of professional broadcasting.Therefore,based on the panoramic football videos,this work conducts research on the key technologies of the automatic broadcasting method for panoramic football videos,aiming to achieve automatic broadcasting for football matches without human participation.By exploring innovative ways of broadcasting small and medium-sized football matches such as non-professional football and campus football,the cost of broadcasting a football match has greatly reduced,enabling more football matches to be displayed to the audience in the form of videos.This paper researches on the key technologies of the automatic broadcasting method for panoramic football videos,and the following research results are obtained:(1)PRDet is proposed to detect the regions of interest in panoramic images with motion information to solve the problem that existing target detection methods cannot effectively detect the regions of interest in panoramic football images.PRDet is a dualinput branch region detection network.It simultaneously extracts image and motion features and constructs new feature representations through feature fusion.Finally,It can locate the audience’s attention quickly and effectively when watching a football match in a panoramic football video,which greatly reduces the calculation and time overhead of subsequent algorithms.(2)A target detection model based on high-resolution feature reconstruction named as HRDet is proposed.The proposal of HRDet focuses on solving the problem that existing target detection methods are not good for the detection of small targets such as football in panoramic images.It continuously reconstructs high-resolution features in the process of feature extraction,and simultaneously performs convolution operations on feature maps of different scales in parallel to obtain high-resolution features with strong representation capabilities,and performs target detection on features of multiple scales.In this particular field of panoramic football matches,we have proved the superiority of HRDet over other target detection networks through extensive experiments.(3)Combined with the above two results,a multi-clue fusion automatic football video broadcasting method was proposed.It adopts the modular design idea,extracts and integrates multiple clues from panoramic football video,including visual information,semantic information,temporal information,and football domain knowledge of the image,and automatically generates the broadcasting video from panoramic football video in realtime.Moreover,a matching automatic broadcasting system of a panoramic football video is designed and implemented.The effectiveness of the proposed method is verified by a combination of subjective and objective evaluation method under a large number of experimental videos and realistic scenes.