Visual Tracking Based on Kernel Principal Component Projection
Dai Jia-sh · Jisuanji fangzhen · 2013
Visual tracking is the research focus in computer vision,with widespread applications. In order to adapt the irregular changes of object apparent in visual scene,we proposed a visual tracking algorithm based on kernel principal component projection. In the initial stages of tracking,the particle filter algorithm was used to get the object apparent image. And based on the joint histograms of color and oriented gradients,the kernel function was defined and the kernel matrix was calculated. Then kernel principal component analysis was used to obtain the projection martix of image samples in kernel sapce. Finally,the best estimation of the object state was obtained through bayesian filter and simulation. The experimental results show that the proposed algorithm is of robustness in real video surveillance scene and occlusion environment,meanwhile is adaptive to object scale changes.