Research And Implement Of Building Recognition Algorithm In The Mobile Augumented Reality System

With the rapid development of mobile devices in recent years,the augmented reality technology on mobile devices is gaining more attention,however,the limitations of mobile devices can't satisfy the real-time requirements of AR applications,the main challenge is the image feature extraction algorithm such as SIFT and SURF in the process of image matching is robust enough and the calculation is more time consuming,usually choose one of them.Building recognition technology plays an important role in some application fields such as the city of visual navigation based on AR,3D city reconstruction and robot navigation and positioning.This thesis discusses the use of handheld devices to obtain the location information and image acquired in camera to implement building recognition in the outdoor augmented reality scene.This thesis presents an overlapping clustering method based on the location for grouping building location information and it can be used in the process of building recognition.This thesis presents a building recognition algorithm based on KAZE and scalable vocabulary tree of dynamic weights assignment can be applied to the recognition of different building groups.By extending building recognition algorithm based on scalable vocabulary tree,the algorithm is proposed to solve the dynamic growth of the vocabulary tree which is used to solve the increasing number of buildings in the mobile augumented reality system.A multi building object recognition method based on scalable vocabulary tree algorithm and random KD tree algorithm is proposed in this paper.The building recognition algorithm of scalable vocabulary tree based on KAZE and dynamic weights assignment is compared with the common vocabulary tree construction algorithm by using the standard building data set and the actual scene building data set.The multi building object recognition method is compared with the template matching method based on the standard building data set.The vocabulary tree dynamic growth and static vocabulary tree algorithm are compared with experiment of building image recognition rate.Through the test of building recognition in the real scene,we can know that the building recognition algorithm proposed in this thesis is better than ordinary static vocabulary tree building recognition algorithm,it has higher building image recognition rate and it can be applied to the development of mobile augumented reality system based on building recognition.