Research And Implement Of Semi-Automatic Video Annotation Based On Multi-Target Tracking

With the development of science and technology,a series of researches on deep learning,such as artificial intelligence and AI robot,have entered the period of rapid development.The existing annotation methods consume a lot of manpower and time for data set processing,which slows down the research progress to some extent.Through the improvement of this process,a lot of manpower and time can be saved and the efficiency of deep learning can be improved.It is a very challenging research content to improve the annotation processing process of data sets.In order to verify the performance,this paper carried out annotation test on cvpr2013 video sequence data sets.After analyzing the data sets,it is found that the content in the video is continuous,which means that the same thing will appear continuously in the continuous frames of the video,but there also exists some situations including shielding,semi-shielding,deformation,light and dark changes,etc.Based on such test results,this paper adopts Multi-Target Tracking method to track things in the video.The tracking results obtained in each frame are analyzed and processed,and the results are saved so that the annotation results of the corresponding frames can be obtained.Considering there are many things need disposing in the target specified of the video frame,and Labeling process require improving the work efficiency and reducing annotation time-consuming.Therefore,based on the test of the public algorithm,this paper selects the KCF target tracking algorithm and let it adapt to the application scenarios by optimizing and improving its process.The target tracking algorithm constructs the target model by training the samples through the circular matrix.Combined with the fast Fourier transform,the tracking speed of the algorithm is greatly improved.Meanwhile,the kernel function can be used to realize the integration of the FHOG feature to improve the tracking accuracy of the algorithm.Finally,the same data set is used to compare the improved KCF algorithm with other tracking algorithms,the test results of which shows that the tracking effect is significantlyThe semi-automatic video labeling system realized in this paper has been preliminarily developed and has been tested and applied in the surrounding scientific research laboratories.To some extent,the system improves the speed and accuracy of video content annotation and achieves the purpose of optimizing annotation.