Research And Implementation Of 3D Reconstruction Based On Computer Vision

Computer vision is called the eye of automation, which is widely applied in fields like national economy, scientific research as well national defense project. Three-dimensional modeling based on computer vision simulates the way human's eye does to deal with scenery, which could be used to measure the three-dimensional information of scenery under many circumstances, whose function is irreplaceable by other three-dimensional re-construction method. Research on stereo vision has great merit with respect to vision biologic and engineering application. This method is very efficient, as an important branch of development of three-dimensional modeling.Stereo vision works the same way as human vision's process of sense. It observes scenery from two or more observing spots to attain sensed images under different vision angles, then through the calculation of place difference of image pixels based on triangular surveying principle the three-dimensional information could be acquired. The basic process for three-dimensional reconstruction research is chosen as image capture, camera calibration, feature detection, stereo matching as well as three-dimensional information reconstruction. The process is based on the survey and reference of mature three-dimensional reconstruction process nationwide and worldwide, aiming at the particular experiment environment. Stereo vision uses the scenario of any angle rotate of target object on rotating platform with fixed camera, and the object's three dimentional reconstruction is done using matlab.This thesis starts from the theory about computer vision, and later introduces some common image pretreatment method with their shortcoming and strength analysis, after which corners detection methods is introduced in great detail based on uniform B-spline with shape parameter. Our thesis delves into featured points matching algorithm and a matching algorithm is proposed based on density degree where featured points reside. It is proved that algorithm precision and calculation efficiency are improved through validity analysis. The chapter of three-dimensional reconstruction introduces the three-dimensional reconstruction chapter introduces the experiment environment and procedure, and the experiments concerning pictures with different pixels are done as well. The results show that, using our proposed reconstruction procedure, error goes down with larger pixels, and the precision is higher, the visual effect is better. However, the calculating time is exponentially to the increase of pixels. At last, conclusion is drawn and future work is proposed.