Depth map generation based on cluster hierarchy and multiple multiresolution camera clusters

A non-transitory computer-readable storage medium comprising instructions for depth map generation using multiple multiresolution camera clusters, the instructions executable by a processor to:

• capture images using the multiple multiresolution camera clusters, wherein the multiple multiresolution camera clusters including a central camera, wherein the central camera having a high-resolution to capture a high-resolution image and remaining cameras in the multiple multiresolution camera clusters having mid and low-resolutions to capture associated mid-resolution and low-resolution images;
• down-scale the high-resolution image and the mid-resolution images to be substantially equal to the captured low-resolution images;
• generate multiple low-resolution depth maps for each cluster using multiple viewpoints of the central camera and the down-scaled images;
• refine the generated multiple low-resolution depth maps;
• generate a low-resolution central camera depth map using the refined multiple low resolution depth maps;
• upscale the low-resolution images to be substantially equal to the mid-resolution images;
• generate multiple mid-resolution depth maps for each cluster using the multiple view points and the up-scaled mid-resolution images;
• generate an initial mid-resolution depth map using the generated multiple mid-resolution depth maps and a down-scaled central image;
• refine iteratively the generated initial mid-resolution depth map;
• upscale central cluster images to high-resolution images;
• refine iteratively a high-resolution depth map using the refined initial mid-resolution depth map and the up-scaled central cluster images;
• and
• generate a final high-resolution depth map using the iteratively refined high-resolution depth map.

An image processing device for depth map generation, comprising:

• multiple multiresolution camera clusters;
• a processor coupled to the multiple multiresolution camera clusters, wherein the multiple multiresolution cameras include a central camera, and wherein the central camera having a high-resolution to capture a high-resolution image and remaining cameras in the multiple multiresolution camera clusters having mid and low-resolutions to capture associated mid-resolution and low-resolution images;
• and
• memory coupled to the processor, and wherein the memory includes a 3D image generation module that is configured to:
• capture images using the multiple multiresolution camera clusters;
• down-scale the resolution of high-resolution image and the mid-resolution images to be substantially equal to the captured low-resolution images;
• generate multiple low-resolution depth maps for each cluster using multiple viewpoints of the central camera and the down-scaled images;
• refine the generated multiple low-resolution depth maps;
• generate a low-resolution central camera depth map using the refined multiple low-resolution depth maps;
• upscale the low-resolution images to be substantially equal to the mid-resolution images;
• generate multiple mid-resolution depth maps for each cluster using the multiple view points and the up-scaled mid-resolution images;
• generate an initial mid-resolution depth map using the generated multiple mid-resolution depth maps and the down-scaled central image;
• refine iteratively the generated initial mid-resolution depth map;
• upscale central cluster images to high-resolution images;
• refine iteratively a high-resolution depth map using the refined initial mid-resolution depth map, the low-resolution central camera depth map, and the up-scaled central cluster images;
• and
• generate a final high-resolution depth map using the iteratively refined high-resolution depth map.