基于均值漂移算法的复杂背景篮球比赛视频多目标识别OA

While analyzing basketball game videos with complex backgrounds,relying on shallow features of multiple targets will result in lower F1 values in the final recognition results.Therefore,a multi-object recognition method for basketball game videos with complex backgrounds based on mean shift algorithm is proposed.The speeded-up robust features descriptor algorithm is used to achieve interest point registra-tion,to compensate for the moving background of video frame images,and to solve the problem of blurred target information in the continuously moving complex backgrounds of basketball game videos.Adaptive morphological filtering is applied to eliminate interfering noise in the complex background so as to improve the quality of multi-object recognition.A multi-object detection model based on the deep residual network-fast region proposal network-fast region-based convolutional neural network is constructed to mark multiple targets in a single video frame.Innovatively,the mean shift algorithm is introduced to analyze the changes in deep features of multiple targets,enabling more accurate tracking and segmentation of moving targets in the video,and optimizing the multi-object recognition results of basketball game videos.Experimental re-sults show that the target detection loss function of the multi-object detection model established by this method is lower than 0.1,which can be applied to subsequent multi-object recognition work to complete multi-object recognition of basketball game videos with complex backgrounds.All basketball players are successfully captured in each video frame,the F1 value of multi-object recognition results exceeds 0.9,and the maximum target tracking delay time is only 0.2s.This method exhibits superior target recognition performance for videos in complex scenes,and by introducing the adaptive morphological filtering algo-rithm,this method eliminates noise information in the images,and improves image quality,which provides assistance and support for basketball game video analysis.