Flexible sheet cameras promise revolutionary views and information
Wed 13 Apr 2016
Researchers at Columbia University have developed functional prototypes for sheet-based cameras which can wrap around objects and provide extraordinarily comprehensive and potentially revolutionary fields-of-view.
In a paper headed for the IEEE International Conference on Computational Photography (ICCP) and entitled Towards Flexible Sheet Cameras: Deformable Lens Arrays with Intrinsic Optical Adaptation [PDF], Daniel C. Sims, Yonghao Yue and Shree Nayar put forward fascinating possibilities for imaging devices which can be deformed as easily as a sheet of paper. One example of a potential application is a 360-degree camera wrapped around a self-driving car and providing it with a complete visual reference for the external world, without blind spots; a security camera wrapped around a lamp-post; and a consumer camera with sensors on one side and a monitor on the other.
Many of the unusual issues which have had to be addressed in the research involve obtaining clear and unambiguous imaging in situations where sensors will inevitably be ‘sparse’ in some sections and dense in others, due to the potentially irregular application areas:
The individual lenses involved in a sheet-camera array are in themselves deformable, changing shape and focal length according to how the entire camera is disposed in itself, or on an object. The researchers were able to solve the problems of aliasing and blurring by developing the principle of ‘passive optical adaptation’, whereby each individual camera/sensor is capable of taking on far more work and a wider individual field of view in cases where it has to cover more external imagery than its neighbours.
Many permutations of the sheet camera concept seem likely to produce information which is valuable for data-analysis, yet practically impossible for a human operator to view in any meaningful analogous way. The example of the SDV ‘omni-view’ wraparound camera could potentially be turned into a viewable VR stream, but since (unlike the cylindrical security camera) the base geometry the camera is applied to is so erratic, only a machine-based system which is not limited by binocular vision will be able to ‘see’ the entire picture.
The team tested the concept of the sheet camera by fabricating a 33×33 lens array using liquid silicon rubber:
‘Fabrication of the deformable lens array. (a) An aluminum mold for a 33×33 lens array was machined. A liquid silicone rubber solution was degassed to remove air bubbles and poured into the mold. (b) The solution was cured at 80 for 5 hours to solidify and the lens array was then peeled off. The array was glued to a flexible plastic sheet with a 2D aperture grid. A diffuser sheet is attached to bottom of the aperture sheet to form 33×33 images of the scene.’
The researchers intend to develop a higher-resolution version of the apparatus, and to experiment with printing organic sensors directly onto the base material constituting the camera. It will also undertake further work to develop a stress-sensor system capable of understanding and compensating for being mounted on extremely varying topography, whilst still delivering a comprehensible and data-versatile imaging stream.
