Research On Basketball Player Target Detection And Tracking Based On Games Video

Moving target detection and tracking is an important research direction in the field of computer vision.With the continuous and rapid development of computer technology,mobile networks cover all regions of the country,the performance of modern electronic equipment is continuously enhanced,and people are receiving more and more information in the form of multimedia.Among them,basketball game videos are deeply Loved by the general public.The research object of this subject is a basketball game video,which clearly displays the player’s movement status in the video,and realizes the target detection and tracking for the designated player.The combination of Faster-RCNN and improved L-K optical flow algorithm for target detection and tracking is more robust than traditional target detection and tracking in a realistic and complex playing field.For the target detection problem of basketball players,this thesis analyzes and compares the three target detection algorithms RCNN(Region-convolutional neural network,abbreviation:RCNN),Faster-RCNN and YOLO(You only look once,abbreviation:YOLO)of the convolutional neural network in detail.Differences in strategy,etc.Use the key frames of multiple basketball game videos to make a player data set,use this data set to train and detect in the more robust Faster-RCNN algorithm,and provide accurate target initial positions for the follow-up operation of player targets information.In terms of moving target tracking,based on the pyramid LK optical flow algorithm,this thesis expands the Kalman filter and combines the two to solve the problem of target loss caused by the rapid movement of the basketball player’s target and the occlusion when the player moves The problem.Because the ordinary optical flow algorithm cannot track fast moving targets,the pyramid L-K optical flow algorithm continuously downgrades the original image to reduce the resolution to finally meet the range of L-K optical flow calculation.When the occlusion is serious,the Kalman filter is used to predict the occlusion of the target to solve the problem of inaccurate tracking of the player due to occlusion or the loss of the tracking target.And the Kalman filter is simple to calculate and has a small amount of data,which will not affect the overall tracking efficiency.Finally,through dozens of fragments of several different basketball game videos,the experimental results show that the combination of Faster-RCNN and improved LK optical flow algorithm has higher target detection and tracking accuracy,and the player moves quickly and the tracking process also achieves good results when occlusion occurs.