Structure from Motion

Abstract

With increasing computational capacity, it is becoming easier to simulate much of the human brain activity. Similar to left and right eyes coordination to perceive shape and distance of the surrounding objects, we can use cameras to get the images of an object from different angles and perform a popular Computer Vision technique known as Structure from Motion (SFM) and utilizing this, we can compute a 3D point cloud and visualize the object in 3 Dimensions. Through this article, we have reviewed the latest advancements in SFM techniques, experimented with dataset, attempt to find limitations for low textural variance using Grey Level Co-Occurrence Matrix methods causing missing pixels problem in the reconstructed 3D point cloud and suggest methods to overcome them based on recently discovered techniques. SFM technology can reduce substantial workload while designing virtual reality environment for commercial domain, engineering surveys and medical assistance.