Research On Moving Multiple Object Tracking Algorithm Based On Video Images

At present,multi-target tracking is a popular research area of machine learning.It is of significant importance in the field of military and civil application.Especially,due to the complexity and diversity of the military mission,multi-target strike has become an imperative research field,which further promote the development of multi-target tracking technology.Taking pedestrians in video images as research objects,this paper mainly focuses on the study of multi-target tracking.The main content of this paper consists of the following three parts:(1)The current development status of multi-target tracking will be briefly introduced,and the existing main tracking algorithms are summarized.Two evaluation systems: CLEAR MOT and MOT Challenge 2015 are illustrated in detail,which are widely used in the world,including the evaluation index of the system and the benchmark test data sets.(2)The generalized maximum multi clique problem algorithm based on batch processing is deeply studied,and an improved algorithm is proposed for the deficiency of the algorithm.On the basis of the objective function of the original algorithm,the real node item was further increased,and the detection confidence was taken as the weight value of the real node item.After that,the corresponding constraint of real node in objective function was designed.Then,the completeness and solvability of problem were guaranteed.Finally,the improvement of tracking accuracy of improved algorithm was achieved.(3)Markov decision process algorithm based on online learning was studied comprehensively.Since the original algorithm is not capable of predicting the target causing some missing scenarios during target tracking,an improved algorithm of Markov decision process was provided in this paper.Adding virtual tracking state in the original Markov decision process is able to solve the missing to a certain extent.