Research On Visual Tracking Algorithm Under Complicated Conditions

Target tracking has a wide range of applications in the life and production of modern society,and with the development of intelligence,the application of visual target tracking is expanding.With the deepening of research in this field,some key problems have gradually emerged.For example,changes in the target itself(such as morphological changes)and changes in the external environment(such as changes in illumination)and so on.These complex interferences will affect the effect of target tracking to a great extent,and even lead to tracking failure.For general target tracking applications,the first frame of target location information is needed.This information is mainly given manually,which can not meet the automation requirements.For industrial applications,the types of target objects are usually certain in a specific environment,so we can use depth learning method to detect the target,and then realize industrial automation detection and tracking.In this paper,we have carried out research on specific target detection training using deep learning.Aiming at the target tracking problem under the above complex interference factors,a tracking method based on minimum barrier distance(MBD)and spatio-temporal context(STC)learning model is studied in this paper.This method has strong robustness to image noise and fuzzy characteristics,and can be approximated by Dijkstra-like algorithm to achieve fast calculation.In the implementation process,MBD transform is used to measure the weight of contextual pixels,and the spatio-temporal relationship between the target object and its surrounding area is determined based on Bayesian framework to estimate the most likely location of the target.Aiming at the problem that the update model of the original STC algorithm is prone to collapse,this paper proposes a bounded scale update model.The estimation of target size based on this model reduces the possibility of tracking failure.The above method is implemented on the I5 computer with a speed of about 160 FPS.A large number of sequential tests show that,compared with the original STC algorithm and other mainstream algorithms,the proposed algorithm has comparable or better performance for some challenges,and has better comprehensive performance.