The Research Of Real-time Mapping And Tracking In Scene Reconstruction

3D scene recovery is a core issue in computer vision, which plays an important role in telemetry, robot navigation, industrial automation, and Archaeology. It makes the equipment have the ability to recognize the scene, so that it can complete some activities without human's help. That earns great application value. And, a reconstruction algorithm with the practice value must be faced with the requirements of real-time reconstruction, good reconstruction result and friendly user interaction. To do this, the researchers have done a lot of research, and put forward a variety of three-dimensional reconstruction algorithm, making the 3D reconstruction no longer a difficult to break. But for the present, designing a robust and efficient real-time reconstruction algorithm is still a challenging task. In this paper, we mainly analyze and study the problem of scene tracking and camera positioning in the process of 3D reconstruction for the fast reconstruction of the whole scenes. The main work is as follows:1, we focus on the SIFT feature and neighborhood template feature. We change these feature descriptors to achieve the requirement of the real-time tracking, the SIFT feature descriptor is simplified, and the ability of heighborhood template feature is improved by lots of effort. And the neighborhood template improved by this paper is demonstrated after the experiment with simplified SIFT.2, this paper realized the tracking method based on the prediction, warping, searching and matching. We use motion model and optical flow to predict the object state. With the prediction, the image is warped by the transform format, and the feature points are matched by the template matching method. The fast tracking method is then realized. In addition, we use the affine transformation template to solve the problem caused by large angle of view, and use the image of pyramid to solve the problem of scale transformation issure.3, this paper integrated the feature and tracking method mentioned above. Using the method of iterative improvement, we position the camera. Using the triangulation method, we calculate the position of the feature points. And then, the scene reconstruction system is completed.