| Target tracking is widely used in real life,and it has important applications.In target tracking area,camshift tracking algorithm is easy to implement and it is robust.So it has been paid more attention by many researchers.Camshift tracking algorithm uses the color histogram to establish a probability map.And based on this probability map,mean shift iteration is used to get closer to the target.Under normal circumstances,the initial target area is not a central symmetric region,and the normal kernel function can not assign weights of the pixels properly.Unreasonable weight function may degrade the weight of target information in some extant,and the weight of some background information may be enhanced,which will affect the stability of the tracking.Besides,when constructing a histogram,the setting of the number of the grid box is a problem.Number is big,then histogram is insensitive to light;number is small,then histogram is much sensitive to light,and pixels in different grid box changes frequently.Therefore distribution of the pixels can not be effectively expressed by the histogram in some extant.(1)To solve the unreasonable assignment of the target information and background information,the kernel function is resized.When the initial selected target region is rectangular,select the minimum side of the original region to generate a central symmetric two-dimensional Gaussian kernel function weight matrix.The weight matrix is interpolated to a weight matrix in the size of the initial target area,and the effective assignment of the pixels in the target area is achieved.(2)To avoid the problem of grid box number setting,the concept of cloud model is introduced.The corresponding concepts are obtained by extracting the cloud model in the area.Calculate the degree of certainty of each pixel belonging to the corresponding concept and ientify the concept of the pixel.Different concepts are assigned different weights so as to enhance the target concept,weaken the background concept,and improve the accuracy and stability of target tracking.Simulations are presented that compared with classical camshift tracking algorithm and significant camshift tracking algorithm.The experimental results show the validity of the proposed method.Especially when the initial target area contains a large amount of background information,the tracking is significant. |
