Research On 3d Object Tracking System For Augmented Reality

3d object tracking system can help augmented reality to effectively process 3d registration of virtual and real objects.The 3d object tracking technology based on monocular vision has attracted much attention in the field of augmented reality due to its advantages such as simple sensor,low power consumption and portability.In the real environment,the object shape and the environment complexity are different,which have a great influence on the quality of tracking.Therefore,this paper proposes an automatic 3D object tracking system solution which can track non-/weak textured objects in real time.This paper introduces the basic knowledge related to pose estimation in detail and discusses many attempts to realize 3D tracking system at home and abroad.The3 d object detection used to initialize the pose and the 3D object tracking used to continuously track the target object are studied deeply.The feasibility of realizing the automatic tracking system is verified from two aspects including precision and realtime performance.3d object tracking technology based on local color histograms.The fan-shaped local region division strategy is adopted to divide the global region into many timeconsistent local regions,which overcomes the shortcoming that the global color histogram does not have strong discriminative feature around the contour.The Kmeans clustering algorithm is used to quantify the local area color space,which overcomes the problem that most color distributions in uniform quantization are 0and reduces much calculation.Define an optimization function based on the GaussNewton method,which improves convergence and speeds up the optimization process.3d object detection technology based on template matching.OpenGL is used to render the object model at different distances from multiple angles based on the sphere to get an appropriate number of template images.Color gradient and local color histograms are used to describe the template image to form a two-modality descriptor,which improves the discriminability of the template.In the detection process,the template component is used as the basis of sliding window matching to enhance its robustness in the case of target occlusion.The corresponding pose of this template can be used as the initial pose of the tracking system,or as the pose of the target object after the tracking failure.Combine the above detection and tracking technologies into a fully automatic three-dimensional object tracking system.For the convenience of users,the tracking system is packaged with Qt tools.To sum up,this paper proposes a fully automatic tracking system solution that can track non-/ weak textured objects in real time,realizing high-precision,real-time3 D object tracking without using GPU acceleration.In the future,this system is very promising to be further migrated to mobile devices,which has a high practical value.