Research Of3D Reconstruction Based On Visual Navigation Of Robot

3D reconstruction is one of the main contents of computer vision, which aims at obtaining3D information through the corresponding2D information, that is to say, the technology need to research the relationship between3D coordinate and its corresponding2D coordinate to have a quantitative analysis of the object. On account of its advantages, this technology has been widely used in mobile robot navigation systems,3D measurement, industrial automation systems, military simulation and medical imaging and so on.In this paper, we use the pioneer robot to capture the video information of the object by its built-in CCD camera to achieve the3D reconstruction. We do the reacherch of3D reconstruction mainly in image preprocessing、scene segmentation and3D reconstruction of the small and large object. This paper will be divided into the following sections:Image preprocessing. It is necessary to perform the image preprocessing before the image processing, including scene segmentation、3D reconstruction and so on. The technique of removing gaussian noise from images is studied in this section. For the phenomenon that the method of filter based on partial differential equations(PDE) has countless iterations and undefined effect, the standard quantity of noise and the thought of genetic algorithm are introduced in our method. The experiment result from the method in the paper have been proved well by the measurement of filtering effect named peak signal to noise ratio(PSNR).Scene segmentation. A new method of recognition based on ROLD and Otsu has been presented in this section, which recognize sky firstly according to the different distribution of ROLD statistics, then distinguish between the trees and roads through the Otsu. The experiment results show that the method in this section has an advantage in the effect of recognition.3D reconstruction. In this section, the method based on Bundler and PMVS is introduced to have the3D reconstruction of the small and large object, which takes the video frames as the input of the Bundler to obtain the sparse cloud structure, and then uses PMVS to translate this sparse cloud structure into the3D dense structure.In consideration of the large capacity of video information, having a interval selection of all frames before reconstruction is necessary. The final reconstruction result express that the method satisfies the basic visual effects. Furthermore, the time of reconstruction is relatively less due to the interval selection, and so, calculation time has been further optimized.