A Survey on Point Cloud Generation for 3D Scene Reconstruction

With the intention of providing a general idea of the process and computational problems to recover a 3D scene from a point cloud, this paper presents a state-of-the-art review on point cloud generation from images for 3D scene reconstruction, including applications with unmanned aerial vehicles. This review is oriented to higher education beginners, both in computer vision and learning algorithms, which are looking for the comprehension of the general literature in order to propose a technological and educational innovation project. The review is focused on the seven stages related with the point cloud generation from images, which are picture extraction, picture matching, camera motion estimation, sparse 3D reconstruction, model parameters correction, absolute scale recovery, and dense 3D reconstruction. As the first five stages are known as Structure from Motion (SfM), the papers were presented in three groups: i) One dealing with one, some or the whole stages of SfM. ii) Another centered on absolute scale recovery. iii) One more related to dSfM (dense SfM) and dense 3D reconstruction. Then, a discussion on the problems faced in reported literature is provided, identifying complex computational problems in each group, such as overstock for images, lack of depth, lack of detail, high processing time, and big size of the point cloud. Lastly, a conclusion regarding the benefits and limitations of the reviewed contributions is given.