Research On RGB-D Based Robust Object Tracking Algorithm

Object tracking is to automatically track the selected object and output its position and state in the video sequence.At present,the difficulties of object tracking mainly include object extraction in the complex background,self-occlusion,mutual occlusion,shadow processing,data fusion of the multi-camera and real-time tracking.Occlusion and shadow are especially prevalent in the real environment.At the same time,due to appearance change and illumination variation,drift and failure seriously affect robustness,accuracy,and efficiency in tracking process.Color images contain rich color,textures gradient,and spatial information,which is conducive to image processing.However,the color images are disturbed by illumination,occlusion and complex background and other factors,which is difficult to achieve excellent robustness in tracking.In contrast to color channels,Depth images refer to the images that take a distance between a point in a scene and an image collector as a pixel value.Depth images are not affected by direction of light source,emission characteristics of surface and shadow,which solves the problem that color image based object tracking is often affected by illumination,shadow and complex background.In this paper,a multi-candidates fusion based head tracking algorithm particle under the particle-filter framework is proposed to solve the problem that the depth image is not affected by illumination and shadow.In the case of depth information only,we pre-process images to enrich detail information,utilize depth information to estimate the scope of human motion and frame-by-frame update object template.A two-stage head searching strategy and particle filter are applied as the motion model respectively.The early termination criterion for template updating is presented to reliably improve the tracking accuracy.The proposed algorithm achieves the robustness and accuracy in complex backgrounds with illumination,deformation and appearance change.In this paper,we propose a RGB-D based scale adaptive tracking algorithm under the KCF framework.The proposed algorithm combines the RGB histogram of oriented gradient,the color-naming feature and the histogram of oriented gradient of the depth image,controls the classifier's size by analyzing depth information and utilizes tracking judgment to ensure accuracy.We solve the KCF algorithm's failure and drift caused by complex background,scale change,fast movement and occlusion.