|Knowledge about all aspects of 3D displays and their application.
|There are many different ways to create a stereoscopic image pair (or for multiview displays a multiscopic image set). Using physical or computer graphics cameras to create images of 3D scenes is common, but it is also possible to create stereoscopic images using almost any imaging technique. Camera techniques are summarised below, other approaches are described in the pages on the left.
Good quality stereoscopic image creation for synthetic images (computer graphics, games and visualization) and natural images (photography, TV, video and cinematography) is a skilled process. Although good results can be produced by trial and error it is much better to use one of the software tools available to calculate camera setup and control the imaging process; those we use at DVL are listed on the resources page. A key thing to bear in mind is that the camera setup for taking a picture of a scene is not usually the same as the eyes looking at that scene. There are four principle approaches to the setup of stereoscopic cameras.
Toed-in ("verging") cameras
In this approach the cameras are rotated so the lens axes verge at a point. Objects in the scene at the same depth as this point will appear close to the plane of the image in the final stereoscopic presentation, other objects will appear in-front or behind this point. This approach is often used for natural imaging, i.e. photography, video and film, but it introduces geometric errors which reduce quality and need to be removed in post-production.
Ideally all stereoscopic images for human viewing should be captured using a parallel camera arrangement. In this approach cameras are setup to have parallel optical axes and image cropping is used to adjust the final depth effect so that it is centered on the viewing plane. The cropping is easy to implement in computer graphics packages, it can be done using image editing software for photographic images and very rarely it has been implemented in the camera hardware. Images captured with parallel cameras do not have the geometric errors inherent in the toed-in approach.
Automated stereoscopic camera (ASC) methods
Recently several research groups have considered the problem of stereo imaging as a 3D mapping problem. That is how to automatically calculate the camera settings to map 3D depth from the scene being imaged onto a comfortable range of 3D depth on the stereoscopic display being used. ASC depth control approaches have been successful for synthetic and natural imagery and we now use them routinely at DVL to give us precise control over the composition of the depth effect in our images.
Multi-rig camera methods
These are an extension of ASC methods that use different stereoscopic settings in different regions of the scene, or for different objects in the scene. For example stereo depth can be highlighted in the foreground where an object of interest is, such as the main character, and reduced in the background where there is less of immediate visual interest. Multi-rig methods are often used in 3D movies where compositional flexibility is important. It requires advanced camera software and/or post-production techniques to achieve this level of control. We used our own multi-rig techniques and software tools in the production of the 3D movie "Cosmic Cookery".