| With the development of virtual reality technology,3D image technology is widely used in video,animation,virtual display and many other fields.However,there are some problems such as high shooting cost and long manufacturing period for traditional 3D images.In order to improve the efficiency of 3D image applications and also reduce the manufacturing costs,lots of3 D image construction methods based on 2D images have appeared.For most tree objects repeatedly appear in virtual 3D scenes,the effect of existing stereo information construction is not very ideal.Most of them unable to fully comply with the structural characteristics of trees and natural laws,some of them also unable to get rich stereo levels.In addition,because of the limit accuracy of detectors,we often lack available stereo tree information of real scenes and also have poor accuracy.So as to improve the utilization of 2D tree images,in consideration of tree features,this article proposed a new approach about construction of 3D tree image based on depth template.This approach firstly constructs reasonable and abundant depth hypothetic information depended on application scenes,then builds 3D tree images by depth information,the tree image objects are composited in 3D backgrounds eventually.This article mainly includes following research aspects:(1)In order to emphasize the individual characteristics of the trees,as well as enhance the stereo layering of depth information,this article proposed tree structure division method based on Lab color model.This method separates the crown area and trunk area of tree object by color difference in Lab color model.Then it achieves tree model matching and crown area resegmentation,these are important prior information for stereo construction.It ensures the match degree between tree structures and stereo information,and also enhances the sense of stereo layering.(2)In order to make 3D information adapt to individual characteristics of trees,this article proposes a depth template based method for tree depth construction.This method establishesvarious types of depth template based on depth gradient hypothesis.It constructs depth maps for typical tree objects with the combination of depth templates and region information.For atypical tree objects,it constructs individual depth information by combination and reconstruction of several depth gradient maps.In this way,needs of different forms of tree objects are all satisfied,and the flexibility of depth template approach is increased.(3)In order to enhance the visual perceived effect of 3D tree image fusion,this article proposed a 3D image construction method based on single tree image and depth information of backgrounds.This method adjusts the tree depth based on fusing position,and it builds tree stereo images by disparity translation based on depth information and single tree image.It’s not only maintains the consistency of 3D image scenes,but also embodies the position relationship between tree and other objects,and the 3D scenes will be more realistic and natural.(4)In order to verify the validity of 3D tree image construction based on tree depth templates,this article designed a 3D tree image construction system.Users can lead the primary processes of construction.This system also has simple visual interface for real-time feedback and modification,it both enhances the flexibility and operability of the system. |
