Study On The Indoor Three-dimensional Reconstruction Based On Monocular Vision

With the development of the communicate technology, it is the dream of human being that computers robots or others intelligent robots are given visual performance. Machine vision research mainly initiate the human being vision performance, extract information from object things'image or image series, proceed process and understanding, finally carries on the shape and the movement recognition to the objective world three dimensional scenery and the object. Indoor cleaning robot when work needs real-time to examine own condition information and the collection external environment information, controls own condition, the establishment environment map, the autonomous navigation and the localization, evades bonds, carries on the way plan, the sensation work object condition, the in section clean effect and so on. Uses the visual sensor is a very good choice. The visual sensor way because has signal survey merit and so on scope width, gain information integrity, becomes moves the robot a main development direction. In the last few years, as a result of computer image processing ability and technical rapid development as well as massive digital image processing equipment performance price compared to the enhancement, adds the visual system superiority, the robot vision system in moves in the robot the application more and more to be valued people's and to display the very good prospects for development.This article the computer vision theory, the algorithm and image processing fuses, with the monocular camera photography object image, then carries on the analysis and processing to the image, thus carries on the three dimensional reconstruction to the environment, sets a target spatial position. At present uses the image to distinguish the distance the research very to be also few, therefore, sets a target the spatial position based on the picture and applies it to the motion robot vision system is extremely significant, is a computer vision application important direction.1.Difficult point of the subject(1).Camera is the most important part of machine vision. In order to get many kinds of parameter, needs to make camera calibration before using it. Camera calibration is the foundation of three-dimensional reconstruction. The outcoming of three-dimensional reconstruction is determined by the result of calibration directly. How to make camera calibration combine with this topic is the difficulty we need to solve.(2).Binocular vision and multi-view vision can recover three-dimensional information of image by the method of stereo matching. How to get the distance information between the point in space and the cleaning robot by monocular vision system using one picture is the difficulty we need to solve.(3).Because of the cleaning robot is moving all the time, it needs to get useful environment information by vision sensor. Then the real-time of the system is becoming a key problem. Image processing arithmetic needs lots of time and memory. And CCD camera image transmit also could spend a little time. So the real-time of the system is the difficulty we need to solve.2.The research of camera calibrationCamera calibration is absolutely necessary for the space object three-dimensional reconstruction. This paper introduced camera model detailedly and discussed common method of camera calibration. Camera calibration commonly means count camera's inside and outside parameter. Projection matrix should be decomposed to get camera's inside and outside parameter after count it. But the process of counting camera's inside and outside parameter is complex and have biggish error. This paper use monocular vision to make three-dimensional reconstruction. During calibration, it can reconstruct the object through the projective matrix, avoid solving the parameter of the camera inside and outside, decrease calculation complexity. By analyzes, bring forwards a new calibration method, namely the least two multiplication. By design a calibration experiment and analyze, we know, both direction error have the trend to increase along with the height of camera to increase. In one group of the same height, error also have the trend to increase along with the elevation of camera to decrease. In the same image, error of X direction changed little along with the increase of distance, error of Y direction have the trend to increase along with the increase of distance. Bring a method of error compensate. It's biggest error is not excess to 2.2cm. It's biggest relative error is 3.16%. It can fulfil the robot navigation basically.3.Binocular vision three-dimensional reconstructionThis paper use binocular vision method make three-dimensional reconstruct- tion. Analyzed current binocular vision measure technology. Discussed edge detective and linear fitting in image processing. Analyzed some factors that influence this subject's three-dimensional reconstruction by experiment the reconstruction of different indoor environment. It includes the influence of shadow, the influence of useless line information and the influence of fitting arithmetic. The influence of shadow is considerably serious. Discussed corresponding resolve method.4.Image gather and processing systemIntroduced each part of image processing system detailed, completed has decoded the electric circuit design and the on-line parameter adjustment as well as based on the TMS320VC5416 image processing goal board design and the debugging based on the SAA7111chip video frequency; completed read using FPGA takes the effective image data the memory; completed the correspondence between FPGA and the DSP HPI.This experiment is based on linear camera model. But actually linear camera model couldn't describe imaging geometry exactly. So the method of experiment needs to be consummate. The result of experiment indicate that this arithmetic can achieve to get object's space coordinate by processing image. It can fulfil the robot navigation basically.