[ad_1]
Researchers in computer vision have demonstrated that they can use light sources and special sensors to see around angles or through thick filters, allowing them to reconstruct invisible forms of objects.
Researchers at Carnegie Mellon University, the University of Toronto and University College London said the technique allowed them to reconstruct images in great detail, including the relief of George Washington's profile on an American neighborhood.
Ioannis Gkioulekas, an assistant professor at the Carnegie Mellon Institute of Robotics, said that it was the first time that researchers were able to calculate shapes on the millimetric and micrometric scale of any kind. curved objects, thus providing a new important component to a wider series of non-curved objects. -Sight (NLOS) imaging techniques being developed by computer vision researchers.
"It's exciting to see the quality of hidden object reconstructions getting closer to the images we usually see for objects in the line of sight," said Srinivasa Narasimhan, a professor at the University of California. Institute of Robotics. "Until now, we can only achieve this level of detail for relatively small areas, but this capability will complement other NLOS techniques."
This work was supported by the REVEAL program of the Agency for Advanced Defense Research Projects, which develops NLOS capabilities. The research will be presented today at the 2019 Computer Vision and Pattern Recognition Conference (CVPR2019) in Long Beach, California, where she received the Best Paper Award.
"This document is making significant progress in rebuilding off the line of sight – in essence, the ability to see in the corners," says the price quote. "It's both a beautiful paper and a source of inspiration, and continues to push the boundaries of what is possible in computer vision."
Most of what people see – and what the cameras detect – comes from the light that is reflected on an object and bounces directly back to the eye or the lens. But light is also reflected on objects in other directions, bouncing off walls and objects. A small part of this scattered light can reach the eye or the lens, but it is evacuated by more direct and more powerful light sources. NLOS techniques attempt to extract scattered light information – of natural origin or otherwise – and produce images of scenes, objects or parts of objects that are not otherwise visible.
"Other NLOS researchers have already introduced NLOS imaging systems capable of understanding room-sized scenes, or even extracting information by using only one-size-fits-all. natural light, "said Gkioulekas. "We are doing something complementary to these approaches: allowing NLOS systems to capture specific details over a small area."
In this case, researchers used a high-speed laser to bounce light from a wall to illuminate a hidden object. By knowing the light pulses triggered by the laser, the researchers were able to calculate the reflection time of the light on the object, bounce off the wall when it returned and reach a sensor.
"This flight time technique is similar to that of the lidars often used by autonomous cars to create a 3D map of the car's surroundings," said Shumian Xin, Ph.D., a robotics student.
Previous attempts to use these flight time calculations to reconstruct an image of the object depended on the luminosity of the reflections thereof. But in this study, Gkioulekas said researchers had developed a new method based solely on object geometry, which allowed them to create an algorithm to measure its curvature.
The researchers used an imaging system that is actually a lidar capable of detecting simple light particles to test the technique on objects such as a plastic jar, a glass bowl, a bowl in plastic and a ball bearing. They also combined this technique with an imaging method called optical coherence tomography to reconstruct images of American neighborhoods.
In addition to seeing the corners, the technique has proven effective in piercing diffusing filters, such as thick paper.
Until now, the technique has only been demonstrated over short distances – one meter at most. But researchers speculate that their technique, based on geometric measurements of objects, could be combined with other complementary approaches to enhance NLOS imaging. It could also be used in other applications, such as seismic imaging and acoustic and ultrasonic imaging.
In addition to Narasimhan, Gkioulekas and Xin, the research team included Aswin Sankaranarayanan, an assistant professor in the Department of Electrical and Computer Engineering at CMU; Sotiris Nousias, PhD student in Medical Physics and Bioengineering at University College London; and Kiriakos N. Kutulakos, professor of computer science at the University of Toronto.
The researchers are part of a larger team of researchers from Stanford University, the University of Wisconsin in Madison, the University of Zaragosa, the Politecnico di Milano and the French Research Institute. German St. Louis, which is currently developing a series of NLOS imaging techniques.
In addition to DARPA, the National Science Foundation, the Office of Naval Research, and the Natural Sciences and Engineering Research Council of Canada supported this research.
Source link
