Research On Moving Target Tracking Based On Kernel Correlation Filter Algorithm Based On Embedded Platform

Target tracking is an important part of computer vision research and application.With the rapid development of technologies such as image processing and artificial intelligence,target tracking research and application have also been greatly improved,and it has been widely used in various fields such as transportation,medical treatment and military.Attention and application.The early target tracking algorithm mainly aims to realize the recognition and tracking of moving targets by establishing a target model.In recent years,based on the correlation filtering algorithm has attracted people's general attention,and achieved more significant results,and has been widely studied and applied.On this basis,the kernel correlation filtering algorithm has been further improved,and its tracking accuracy and operation speed have been greatly improved,with good robustness and practicability.In this paper,the main methods of moving target tracking algorithm are analyzed firstly,especially the related filtering methods are deeply studied.The basic characteristics and tracking effects of the kernel correlation algorithm(KCF)are analyzed and discussed.On this basis,based on the tracking performance of the kernel correlation filtering algorithm,some performance analysis and experimental comparisons are made on the basic parameters,such as regularization parameters(lambda),feature extraction parameters(cell_size),and model update parameters(interp_factor).Get an initial understanding of the main features of the algorithm and the scope of the applicable objects.Aiming at the problems existing in the algorithm,an improved method is proposed to dynamically update the basic parameters of the tracking model,the accuracy is improved,the speed is accelerated,and good results are obtained.For the occlusion problem,the algorithm has a good effect and solves the tracking failure phenomenon in the original algorithm.For the scene with similar target and background,the algorithm also has better performance and can overcome the misjudgment caused by similar background.Based on the improved kernel correlation filtering algorithm,this paper transplants it into the embedded platform.In the Linux environment,the OpenCV and Qt working environments are built,and the model library function of the improved algorithm is established to realize the task of tracking the moving targets.The experimental results show that in the embedded environment,the improved algorithm has better tracking ability for moving targets,faster calculation speed,better robustness and good results.