Research On Behavior Detection And Target Tracking In The Classroom Recording And Broadcasting System

Due to the huge differences of educational conditions in our country,educational resources in urban areas are usually better than those in rural areas.A classrom recording and broadcasting system can give a better educational chance to rual schools.Many universities and companies have begun to develop classroom recording and broadcasting systems.A classroom recording and broadcasting system mainly uses image processing techniques such as target detection,target tracking and behavior analysis to realize automatic recording of classroom teaching by cameras without manual intervention.In this thesis,we conduct research on behavior detection and target tracking in the classroom recording and broadcasting system.The main work and achievements are as follows:(1)This thesis proposes a student behavior detection method based on motion estimation.It is mainly used to detect stand-up and sit-down behaviors of students.Firstly,the system collects YUV data frames through the camera,converting YUV data frames into grayscale frames,and applies median filtering on grayscale images,then perform interframe difference and morphological processing on adjacent frames.Secondly,the system extracts the bounding rectangle of a moving object,carries out motion estimation in this area,and calculates the horizontal displacement vector and the vertical displacement vector of each macroblock in the moving block and then calculating the displacement vector of the macroblock base on the horizontal displacement vector and the vertical displacement vector of the current motion contour.Finally,the system calculates the angle of the image block,and determines the student behavior of according to the value of the motion angle of the image block.The proposed method considers the real-time requirement of video processing in a classroom recording and broadcasting system,and reduces the amount of calculation,so that it can detect student behaviors more efficiently and quickly.(2)This thesis proposes a long-term target tracking algorithm based on Kernel Correlation Filtering(KCF).It mainly applies to the target tracking system.Firstly,the algorithm uses samples to train the tracker,so as to estimate the position of the target in the current frame and realize the normal target tracking process.However,when the moving object is occluded or seriously deformed,the tracker will bring too much background information,resulting in target tracking failure.Therefore,target re-detection mechanism is introduced to solve the problem of target frame drift.Firstly,the algorithm needs to use the kernel correlation filter tracker to track the target and calculate the Peak-to-Sidelobe Ratio(PSR)of the tracking target.Then,by comparing the peak response strength with the empirical threshold size,it can judge whether the tracking object is lost.When the PSR is less than the empirical threshold,the algorithm determines that the target tracking fails,then triggers the re-detection mechanism.In this mechanism,the motion estimation algorithm is used to get the rough position of the target in the current frame near that of the target in previous frame,and then the tracker is used to get the precise position again in the neighourhood of rough position to track the target again.This system considers the real-time requirement of video processing,and has strong adaptability to target tracking of teachers and students.