Development of pilot system of artificial vision for the acquisition of a point cloud using 3D vision technologies

Abstract

One of the goals of artificial vision is to permit that a computer analyzed a real scene, as if a person does. To achieve this purpose it is necessary to create a 3D model of that scene using a reconstruction. 3D reconstruction is the process by which the shape and the appearance of a three dimensional object or scene from a volume by analyzing the digital information provided by different types of sensors is recovered. The sensors may be passive (not interact with the object, such as different types of cameras based on the light of the visible spectrum) or active (the interaction comes from an object in response reflected waveform that is captured by the device). On the other hand, 3D reconstruction has several applications, such as robot navigation, allowing it to know in what part of the scene is located and being able to plan its movements without needing human help. It is also useful for determining quantities such as distances, areas or volumes, which may be applicable for quality controls as it can verify the processes and areas of objects that are being manufactured. Another application is the digitization of historical monuments and museums to create virtual tours, which users can access from the Internet. Besides, 3D reconstruction are given in the area of biomedical engineering. Anatomical reconstruction from medical images such as MRI structures has become an important tool in medical diagnosis and therapy planning and surgical procedures. Apart from the above applications, 3D reconstruction has many more applications in different areas. These are some of the many uses of three-dimensional reconstruction and for this reason there is a need to develop this project. The purpose of this project is to do an algorithm that, based on images, obtain a points cloud of an object. To achieve this aim, in first place; the different techniques developed about 3D reconstruction were studied in order to know the different possibilities. Some of these techniques such as telemetry laser, stereo vision, flight time or structured light which obtain models that are very accurate or not, but with the disadvantage of using expensive equipment in some cases. In second place, perform camera calibration using a calibration method. Finally, get the point cloud object to rebuild. This work shows that an algorithm can be done to reconstruct an object in three-dimensions, leaving for future developments the optimization for all kinds of objects. In addition it is an important basis for future developments, as many different techniques for image analysis were studied and compared.