Research On Automatic Photo Screening Strategy Of Meeting Room Based On Multidimensional Pattern Recognition

With the gradual development of society and the continuous progress of science and technology,the organization and holding of modern meeting has become more scientific and perfect.As an indispensable part of the meeting,meeting photography can play the role of summary,remembrance and publicity.For the organizers of the meeting,its importance is self-evident.However,photo screening of meeting room often costs massive human and material resources.With the rapid development of computer technology,it is necessary to study how to realize the automatic photo screening strategy of meeting room at a lower cost through computer technology.Aiming at the complex environment of different types of meeting rooms,this paper proposes an automatic photo screening of meeting room based on five-dimensional pattern recognition methods of face,facial expression,human posture,object and text.According to the proposed strategy,the experiments are carried out in the simulated meeting room scene.The main research works include the following aspects.(1)The definition of meeting scene and related abnormal behaviors.According to the number of abnormal behaviors,the meetings are divided into three categories:serious meeting,relatively free meeting and free meeting.The typical meetings in these three categories are studied: lectures,academic seminars and get-together.The abnormal behaviors defined by these meetings gradually decrease.Abnormal behaviors in the get-together includes sleeping and playing with mobile phones.The academic seminar adds eating on the basis of the get-together,while the lecture adds using computers and reading books on the basis of the academic seminar.(2)The design of multi-dimensional pattern recognition algorithm.According to the different objects that need to be photographed in the meeting room,five-dimension pattern recognition algorithms are designed.The objects to be photographed and the pattern recognition algorithms involved are as follows:(1)combined with face,text and human posture recognition to recognize the leader and his body movements during speech,which is the close-up of the leader;(2)combined with facial expression,human posture and object recognition to recognize the expression of other participants when they listen and think carefully as well as the abnormal behaviors in the meeting,which is the close-up of other participants;(3)combined with object recognition and text recognition to identify banners,PPT and other objects that can show the theme of the meeting,and shoot as many participants as possible through face detection,which constitute the overall shooting of the meeting.(3)The design of automatic photo screening strategy in meeting room.Firstly,all the photos are screened for general image quality evaluation,and the current meeting stage and picture type are marked when naming the pictures;secondly,for the banners and PPT,which can represent the theme of the meeting,the text recognition algorithm is used to judge the current meeting type;Thirdly,combined with the five-dimensional pattern recognition algorithms,the objects to be photographed at the current stage is identified in the specific stage of the meeting.Finally,according to the specific screening criteria of meeting pictures,the images are screened to get the final meeting pictures that meet the requirements.The experimental results in the simulated meeting room show the effectiveness of the proposed strategy.