Hybrid Tracking And Registration Method For Augmented Reality In Complex Environment

By the era of mass production is gradually being replaced by the era of mass customization,manufacturing process is facing the new challenge.Manual operation is still important in some cases when manufacturing process is getting more and more complex.But,manual operation is error-prone,hard to be standardization and high proficiency needed.Augmented reality is a visualization and HCI technology which seamlessly combining virtual information with the actual reality to augment the reality which is a better way for cognition and learning.It can provide more effective training and guidance for manual operation.Because the viewpoint of human is unlimited in the operation process,the object being operating in observer perspective can be in complex environment which combines different transformations and situations such as scale change,rotation,perspective distortion,occlusion,out-of-view,motion blur and etc.The background is not unclusttered but clustered by other interferential objects.3D registration,the core technology of augmented reality,is still facing the low precision and tracking stability problem in this kind of complex environment which leads to visual error or bad visual effect that limits the application of augmented reality in different domain especially in the industry.To address the problems above,this paper combines the improved ORB and LK method as hybrid tracking method to improve the stability in complex environment.The main research topics are as following:(1)Improved ORB keypoint method.Study the detection,descriptor and matching method of ORB keypoint and analyze its disadvantage.Propose to use non-linear scale space to replace the Gaussian pyramid to improve the scale invariant and study how to compute and build the non-linear scale space.Study the adaptive non-maximal suppression method to obtain uniform keypoint and reduce redundancy.Compare the original ORB and improved ORB method.(2)Improved Lucas Kanade tracking method.Study and analyze the fundamental and different optimization method of LK and its disadvantage.Study to combine non-linear local descriptor with IC method to address the problem of illumination.Study the forward backward error detection method to evaluate the tracking result and get rid of the wrong keypoint.(3)Hybrid tracking and registration method based on consensus voting.Propose to combine the improved ORB and LK method in parallel.Use the consensus method to fuse the data,vote for each keypoint and cluster the votes.Use the largest cluster as the tracking result and discard other cluster to achieve 3D registration.(4)Dataset experiment and engineering application.Verify the performance of paper method in the object tracking dataset in scale change,rotation,perspective distortion,motion blur,occlusion and out-of-view in complex environment.Verify the tracking performance of paper method in the AR aided wiring system.