The design and reconstruction of 3Dmodels has now become a significant part of computer graphics and of computervision applications and also in areas such as architecture, visual and specialeffects in entertainment, augmented and virtual environments, games,engineering and education. 3D reconstruction can be achieved by two approachescalled as active and passive. 3D Max and Blender are the examples of thetraditional modeling system that uses 3D meshes for the construction of 3Dmodels. However, reconstruction from these 3D meshes was challenging andtherefore, resulted in need for a better approach.
A major progress has beenseen in the last decade to resolve this issue. Photogrammetry, Structure frommotion(SFM) and Image-Based Modelling(IBM) are the widely used modern approaches for building 3Dmodels. There are few alternative technologies such as laser scanner, lidar, structuredlight, camera calibration, etc. used for the same purpose. 3D laser scanner isaccurate but quite expensive and hence not adapted easily.
Though structurefrom motion and photogrammetry fail to reach high performance, but it is aninexpensive approach and is more accessible when compared with laser and lidar.Structure from motion(SFM) approachcomes from the roots of photogrammetry and computer vision. Photogrammetry isthe technology of extracting information from images. It measures and processesfeatures such as length, depth, angular position, etc in images(photographs).
This technology has been successfully employed in a wide range of industries.The main application of photogrammetry3 is to generate 3D models out of theimages taken from an object, and also used for many different purposes. Processingcomplicated shapes of small objects, uneven surfaces, reflective surfaces andobjects, extracting features of transparent object surfaces are the challengesthat need to be improved.SFM allows to achieve 3D structure ofobject from images as well as the camera motion which is done by bundleadjustment. SFM takes extracted features from 2D images that include pixelpositions, edges and corners of objects, lines and curves across the cornersand edges, angular positions of object with respect to 3D coordinates, lightintensity and other measurements such as length, depth as input. These imagesare taken at different angles from a single camera by either moving the cameraposition or by moving the object.