Research On 3D Reconstruction Method For Multi-Source Object Based On SFM

The 3D(three-dimensional)reconstruction technology based on SFM(Structure From Motion)algorithm is very important in computer vision,which takes photos taken from different angles of view as input data and restores 3D model of target by reconstruction algorithm.Because of the advantages of high accuracy,convenient data collection and high practicability,this technique is often used to simulate the real scene in photoelectric virtual reality,to make real models in game animation,and to simulate the tissue structure of human body in medical diagnosis.At present,many applications require not only the reconstruction of the target 3D model,but also the infrared thermal information on the model surface.However,due to the uniqueness of the infrared imaging principle,the infrared image acquired has features such as indistinct details and unclear texture.The 3D model of the scene or target is directly recovered from the infrared image.Therefore,the visible light image and the infrared image need to be registered,and then a 3D model with shape,size,color information and temperature characteristics is reconstructed.This paper fully analyze the development of 3D reconstruction methods,and on the basis of researching the basic principle of SFM algorithm,aiming at the characteristics of visible and infrared images,propose a method of 3D reconstruction of multi-source objects based on SFM algorithm,which realizes the goal of 3D reconstruction of multi-source objects.The main contents of the study are as follows:1)Camera calibration is the prerequisite of 3D reconstruction,and this paper study the imaging model of the camera and the calibration method about it.Firstly,the conversion principle of the correlative coordinate system and camera imaging model needed for 3D reconstruction are introduced.Then this paper introduces several common camera calibration methods,and focuses on checkerboard plane calibration method.Finally,the camera used in this paper is calibrated by using the checkerboard method,and the distortion correction is carried out.2)The registration of visible and infrared images is the focus of 3D reconstruction of multi-source objects,and this paper studies the principle and process of multi-source image registration.First,the features of the visible image and the infrared image are separately pre-processed to enhance the image characteristics.Then the image edge is detected and the SURF(Speeded Up Robust Features)feature point is extracted.And this paper matches the multi-source image with the SURF feature point and register the infrared and visible image by affine transformation.3)This paper mainly study the realization process of 3D reconstruction based on SFM algorithm.Firstly,the principle of SIFT(Scale Invariant Feature Transformation)feature point extraction is introduced,and the image feature is detected by SIFT method.Secondly,the process of SFM reconstruction algorithm is introduced,including constraint of epipolar line,essential matrix solution,projection matrix decomposition and triangulation reconstruction.Combining the reconstruction results of multi-view images.The 3D model of the target is restored by using the visible image.Finally,the infrared image is mapped to the 3D model.The goal of 3D reconstruction of multi-source object is achieved.