Metric-Scale 3D Scene Reconstruction Using Dual Images from a Calibrated Camera

This paper introduces a technique for 3D scene reconstruction using two images from a calibrated camera, with an emphasis on accurate scale determination and camera motion recovery. The first crucial stage in the procedure is to identify and match feature points between the two photos. The geometric link between the two images is then ascertained by estimating the critical matrix. The motion of the second camera is then calculated to determine its relative position and orientation. By re-detecting and tracking feature points in both images, triangulation is utilised to estimate the 3D locations of matched points, producing a sparse 3D structure of the scene. The recovery of the scene's scale using a known object size—a globe with a radius of 10 cm—is a noteworthy improvement in this technique. By doing this, the reconstruction's arbitrary scale is changed to a metric scale from the real world. The finished 3D model is shown in real-world units, with cameras and points represented in metric dimensions. This method offers a robust foundation for accurate 3D scene reconstruction and has a wide range of applications in computer vision, photogrammetry, and robotics for tasks including mapping, item detection, and environmental modelling. Additionally, it provides flexibility and scalability for a range of real-world uses.