A Device to Develop Depth Map Using Principles of Stereoscopic Vision

Recovery of three-dimensional data from two-dimensional pictures is the primary goal of stereo techniques. The problem of correspondence is fundamentally the problem of retrieving depth (three-dimensional information) from a group of pictures. Knowing the correspondence makes determining the depth a simple geometric problem. In this document, we provide an approach for building a stereo camera as well as an algorithm for building depth maps of any scene using “Two-Dimensional” images that were taken with two different cameras. It is based on stereoscopic vision and intend to mimic human depth perception. It uses an algorithm, based on pre-trained models available in the “torch” library, to determine the pixels corresponding to the same scene. The point correspondence is used to calculate the disparity of each individual pixel and thereby, a depth map is created using this information. This has many advantages over conventional techniques since it is very inexpensive, both in terms of cost and energy. Conventional techniques of LIDAR and RADAR to capture depth images, being highly expensive and bulky, tend to limit their application to only expensive technologies like in military and space exploration. This technology has many applications in forthcoming developing technologies like Augmented reality, Virtual reality, Self-driving cars, etc. It can also be used to create three-dimensional models of real-life objects for the VFX industry.