Implementation Of Real-time 3D Scene Reconstruction Techniques On The Telerobot

In recent years, with the development of front fields such as computer vision, artificial intelligence, robotics and virtual reality, study of real-time navigation and three-dimensional scene reconstruction system applied to telerobot is highly sought after by researchers. This paper presents a matching template window of sparse position to calculate similarity measure criteria of feature points. The ’ring’ matching algorithm is presented to track images feature in camera ego-motion pose estimation. For the problems in existing regional dense matching algorithm such as consuming too much and poor accuracy, this paper propose a fast dense matching algorithm based on Bayesian probability. We establish priori distribution functions about parallax values according to triangular meshes generated by robust feature matching points to reduce the time complexity and ambiguity of matching process. The main work of this paper includes the following four areas:1) Using two CMOS cameras of same type, image acquisition card of dual network, steel support and cable to build a binocular stereo vision system platform. To calculate the calibration parameters of binocular camera based on MATLAB calibration toolbox using image sequence of checkerboard calibration plate.2) To detect feature points of two-dimensional image data in real-time using template feature point detection algorithm. To improve template window for the calculation of the similarity measure guideline and present the ’ring’ feature matching algorithm to achieve real-time tracking of image sequence features. To complete real-time estimation of camera motion pose parameters based on the mathematical model and theory of pose estimation.3) This paper propose a fast dense matching algorithm based on the Bayesian probability model and complete dense disparity map of real world images in real time use techniques of GPU acceleration and multi-threaded programming. To compute three-dimensional scene points cloud use calculation method of three-dimensional world coordinate of space object points and camera calibration parameters.4) To confirm the demand analysis and framework of telerobot real time location and three-dimensional scene reconstruction system and achieve the system from programming. We determine the position of robot in real time in the reconstructed three-dimensional scene.