Reseach On 3d Object Matching Based On Shpape Feature Combined With Texture

Object matching is one of the hot research areas in computer vision, image processing,it is widely used in various fields,such as daily life, productive and military activities. This paper summarizes the current situation of three-dimensional object expression, some research and deficiencies of three-dimensional object matching. Obect matching based on two dimensional information is not suitable for texture-less object;Object matching based on contour is not suitable for cluttered scenes and Its matching accuracy is low;object matching based on only three-dimensional geometry information cannot reflect the object texture information.This paper studied the object matching technique besed on three-dimensional geometry information combined with the surface texture information,the main work is as follows:Firstly, In order to achieve the full three-dimensional model which is required in three-dimensional object matching. this paper introduces a fast implementation method of reconstruction based on three-dimensional image which Contains the texture information.The Rotation and translation matrix of these depth images from different perspectives can be calculated through the manual registration, the ICP registration and global registration.Vrippack can reconstruct the full three-dimensional image based on the Rotation and translation matrix,the full three-dimensional texture image can be obtained through Texture mapping by TextureStitcher,so as to Implement the three-dimensional texture image reconstruction.Then,In view of the existing three-dimensional object matching method can not solve the problem of free form expression of both shape and texture, this paper presents a method to three-dimensional object matching combined three-dimensional shape and two-dimensional texture feature.Sift feature is extracted from the range image in the scences,and then the range image matches with a series of 2.5d range images which were used for the 3d model reconstruction one by one based on sift algorithm,so tha it can find out the most familiar local range image to the object in the scences.The matching between this local range image and the object is completed through three-dimensional shape feature.It is to initialize the model, in other words,It is to reset the model close to the object in the scenes.A icp algorithm combined with color is used to implemente the matching between the object in the sences and the model which was reset before.in this way the pose of the object in the sences can be calculated accurately.Finally, this paper shows that the design of the SHOT descriptor can naturally be extended to incorporate texture,we change the Weighted coefficient between shape descriptor and color descriptor automatically to get a new descriptor named COLOR SHOT.We use this kind of descriptor to do the matcing work. An overall test have been done on these method and some analysis are also made through comparing with other method.