Summary: | 碩士 === 國立中興大學 === 機械工程學系所 === 104 === In now industry, three dimensional (3D) range image registration is an important technology, such as computer tomography (CT) to rebuild human bone structures. The so-called 3D range image which use 3D scanner to detect the shape and appearance on the object or environment. It is often used for 3D reconstruction. With using the 3D scanner to measure the distance, the range image is also called depth image. However, because of the scanning range is limited at 3D scanner, it’s often need to change the relative position of the scanner and the object or place the object on the turntable. It scans many times to piece together to a complete model of the object. We call the technology of multiple one-sided model integration as the range image registration or alignment.
In this Paper, we use the algorithm of cross correlation in range image registration. Calculating the surface normal vector orientation histogram to achieve translation irrelevant, so that rotation and translation can be calculated separately. We focus on the coarse registration. The algorithm is divided into two steps. The first step is making the orientation histogram of surface normal vector, and calculating the cross correlation of the two range images to obtain the rotating angle. The second step, the range image is transformed to frequency domain and calculates the cross correlation to find the translation. After getting the rotation and translation of both range images, the registration can be finished by rotating and the translating.
As beginning, this paper calculates the surface normal vector of range image and uses orientation histogram to achieve the irrelevant translation of both range images. Then performing the Fourier transform which is defined in SO(3) for defining the orientation histogram in spherical and calculating the cross correlation to obtain the rotating angle. The rotating range image is transformed to frequency domain by using the Fast Fourier Transform (FFT), and calculating the cross correlation to obtain the translation. At last, the coarse registration is completed by translating the range image.
|