Grant: Idea to Innovation

The commercialization of 3D display technology has increased demand for 3D video content. However, creating 3D video content is time-consuming and expensive. Since most existing content is in 2D format, a number of 2D-to-3D video conversion technologies have been developed. These typically employ several different depth cues that help the human visual system perceive the relative distance of objects within a scene. These techniques, however, overly focus on some depth cues and fail to estimate the depth map accurately for different contexts. The resulting 3D video is of poor quality. We have developed an automated conversion method that much more accurately mimics the human visual system and employs a far greater range of depth cues. A machine-learning algorithm known as Random Forests is used to imitate the human brain's process for fusing such information, producing a very accurate and natural depth map, which is then used to render highly lifelike 3D video. The accuracy of this depth map makes it also possible to generate a number of different 3D viewpoints, allowing a glasses-free viewing experience. In this Idea-to-Innovation project, we propose providing customized prototype 2D-to-3D video conversion solutions for postproduction facilities to convert existing 2D movies to 3D format, which will reduce the high cost of conversion by reducing the manual involvement required. The prototypes developed through this proposal will enable our industrial partners to accurately assess the commercial potential of our conversion technology. With the profile of our technology sufficiently raised through these relationships, we will explore establishing a start-up company to disseminate our technology to a broad range of Canadian and international companies.