A new type of camera that does not use conventional lenses could allow robots to see in 3D vision.
The system uses new AI-powered image processing algorithms to capture information about a scene in a single exposure.
Lead researcher Weijian Yang from the University of California, Davis explained that the camera was considered “lensless” as it replaced the bulky conventional lenses of a traditional camera with a flexible, lightweight microlens array.
Each microlens in the array is able to observe a scene from different angles, allowing the AI to carry out complex tasks such as building up a 3D picture of an object even if parts of it are obscured.
The camera learns from existing data how to digitally reconstruct 3D scenes and, when combined with the images from the microlens array, this allows it to create a complete 3D image in real-time.
This could potentially be used to give 3D vision to robots, allowing them to navigate in 3D real-world environments or to carry out delicate tasks such as handling fragile objects.
It could also potentially be used in areas such as inspection of industrial parts, gesture recognition, and 3D content used in gaming and other entertainment.
New design uses AI based on artificial neural networks
Other researchers have previously worked on cameras using single layer microlens arrays, but these proved impractical due to slow reconstruction speeds and the levels of calibration required.
The UC Davis researchers said that they considered the microlens array and reconstruction array as a whole system rather than approaching them individually.
The array uses 37 tiny lenses set in a circular layer of flexible polymer with a diameter of 12mm.
This works in conjunction with the algorithm, which is based on an artificial neural network and learns how to map and reconstruct from the images it receives.
It is based on a physical model of image reconstruction, which Yang claims makes the learning process easier and results in better reconstructions.
Once the algorithm completes its learning process, it is able to reconstruct objects at very fast speeds, with no need for extensive calibration.
When testing the camera’s performance, the team found that it was able to perform accurate 3D imaging of objects at different depths and distances.
It also imaged objects partially obscured behind opaque objects – the first time this had been done.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
