Research On Video Target Tracking Algorithm Based On MeanShift

Video target tracking is one of the most important research orientations in the field of computer vision.It has high practical value in military and civil fields,such as imaging guidance,intelligent transportation,intelligent monitoring,human-computer interaction,medical diagnosis and so on.Therefore,this has caused the domestic and foreign scholars to study the target tracking deeply,and scholars have put forward a lot of algorithms about target tracking.The target tracking algorithm based on Mean Shift is a method without parameter estimation.Because the has advantages of simple computation,easy understanding and real-time tracking,it has become a hotspot algorithm in the field of video target tracking.However,there are two serious defects in the algorithm: On the one hand,the algorithm use a single color feature to describe the target,it is easy to lose the target or track t he wrong target when the target is interfered by a similar color target.On the other hand,The kernel function bandwidth of the algorithm is fixed,it can not achieve adaptive tracking of the target's scale and direction when the size and direction of the target change.In order to make up for the defects of the algorithm and improve the tracking accuracy,this paper proposed effective solutions.The main research contents of this paper are as follows:(1).For the defect that the algorithm is easy to be disturbed by color,the colorinsensitive SURF feature and the color feature are used as the target matching feature.When the feature matching process is disturbed by a similar color target,the SURF feature is used to eliminate the influence.The data shows that improved the algorithm by the SURF feature can overcome the influence of similar color targets and improve the tracking accuracy of the algorithm.(2).The SURF feature point has the invariant characteristics of scale and rotation.For the defect that the algorithm can not adaptively track the target of the scale and orientation change,according to the relationship between the SURF feature point 's scale and orientation and the target's actual scale and orientation to achieve the adaptive adjustment of the tracking frame.Experimental results show that the improved algorithm can achieve adaptive tracking of the target's scale and orientation.(3).In order to ensure the real-time tracking of the algorithm,a target position prediction method is proposed by combining the three-frame difference method,the nearest neighbor method,and the orientation parameters of the target.In frame of the video,the location of the target is firstly located using a prediction method;then,using Mean Shift algorithm to iteratively calculate the true position of the target.The experimental results show that the improved algorithm can reduce the number of iterations,reduce computational complexity,reduce the time consumption of the algorithm,and ensure the real-time tracking of the algorithm.