RGB-D Visual Tracking Technology Based On Context Information

For a long time,visual tracking technology has been one of the most important research issue in computer vision field,and founds a broad application prospect in military and civilian field.Although the visual tracking has achieved some positive results since its emergence,it still has room for improvement.Microsoft Kinect device proposed in 2010,greatly reduces the acquisition cost of the image depth and accelerates the development of many computer vision technology,including visual tracking.Compared with the traditional tracking algorithm based on RGB data,the visual tracking algorithm based on fusion of depth information is more excellent in tracking accuracy.The traditional object tracking method based on the RGB image sequence is easy to be affected by the factors such as illumination,viewpoint change and mutual occlusion.Therefore,this paper presents an improved spatio-temporal context fusing RGB-D information tracking method by combining the object of spatial and color context with robust illumination and scale features in high dimensions' space.At the same time,facing the problem of significantly reducing of the tracking accuracy,this paper proposes a novel mechanism based on the gradient prediction of object depth in time domain.At last,a prototype system is developed to verify the effectiveness of this work.By utilizing adaptive depth information template,a template fusing RGB-D information tracking model is established,which significantly enhances the complex scene and partial occlusion robust tracking algorithm.At the same time,a novel method to update the depth template is proposed to improve the accuracy of the tracking algorithm.In addition,to develop a prototype system,we propose the occlusion detection and processing mechanism based on the estimation of the depth gradient on time domain.Using the mechanism,we realize the assessment of occlusion of the target.Besides,we will deal with a target with a serious occlusion.These methods improve the adaptability of the algorithm to serious occlusion and cross shade,and further improve the accuracy of the algorithm.The effective combination of the above two methods makes the algorithm more complete,and forms a robust and high precision visual tracking algorithm,which is called improved object tracking method based on RGB-D information.The experimental result shows that the proposed algorithm can effectively improve the tracking accuracy.At the same time,it can maintain relatively stable performance when the target is with serious occlusion,cross occlusion,and even complex background clutter and deformation.However,compared with the traditional algorithm,the proposed algorithm is not perfect due to the lack of adaptability of the model to multi class targets.In the future,we hope to improve the adaptability of the model.