There are several implementations that employ many cameras organized in two- dimensional arrays. One example is Stanford Multi-Camera arrays. Typical applications are related to computing low resolution Depth Maps to obtain higher resolution RGB images, sometimes called Super-resolution images by using Parallax Techniques similarly to human eyes where the differences between the two views of the scene (the “binocular disparity”), allow the brain to calculate the depth for every point on the scene visible by the two eyes. Stereo cameras can partially address this issue but have several limitations. Multiple cameras improve the quality and accuracy of depth measurements, but this is typically a very computationally expensive process. The best results are usually achieved when cameras are positioned on radial lines around a central camera.

Determining final Depth Map could be very expensive, both because of the total number of Parallax computations for many depths, and since the large number of images from different cameras being used simultaneously puts a lot of pressure on efficient memory use.


Cluster Imaging’s goal is to compute the highest quality Depth Map using minimum computations. Camera Clusters concept relates to computational photography systems and methods for creating 3 dimensional (3D) images using a set of digital camera clusters consisting of one high resolution central camera surrounded by a number of specially positioned low resolution cameras. Another aspect of the Cluster Imaging architecture that reduces computations is a hierarchical approach to computing Depth Maps. As a result, the systems can produce high-quality and high-resolution depth maps at 30 frames per second using typical CPUs, without resorting to special purpose computing systems, such as GPU’s.

High resolution Depth Map could be more efficiently produced by using a multi-resolution camera set including central high resolution camera surrounded by several radially positioned camera clusters. Generation of high resolution depth maps could be accomplished in a computationally efficient way by using a hierarchical approach, by initially computing depth maps for clusters having lower resolution and using initially relatively small number of depths. The next step is to refine cluster depth maps and then to compute a high resolution depth map by using high resolution central image, central images of all clusters and clusters depth maps. The depth maps of clusters are a good start for initial depth set for high resolution depth map which could be further refined at the next stage.


NASA Presentation

Cluster Imaging was awarded as a NASA 2019 cycle 1 finalist.

Picture 1

Subset of NASA presentation

Depth Maps open possibility to develop innovative 3D applications