Computer Vision News - June 2018

Computer Vision News Application 36 Application: Altia Systems imager and its own ISP (Image Signal Processing). Their IP is a streaming architecture running in an FPGA , which they call the PanaCast Vision Processor . In total, there are nine processors in the system, and one of the chips is the core IP. It also has two microphones and an audio DSP (Digital Signal Processor) . The three adjacent cameras overlap by about 20 percent. Then they run in real time a software algorithm, a dynamic scene stitching algorithm, to do pixel accurate stitching. Computationally, this is a very hard problem. Many try to do this in the cloud or as a post-production step afterwards, but the team at Altia Systems figured out how to do it in real time at the camera level inside the device. It’s extremely fast in order to collaborate, and the incremental latency for the stitching is 5 milliseconds . Which is exceptionally fast. When asked about the computer vision techniques that they use to create this impressive technology, Natarajan explains: “ We are using something like multi-view geometry to figure out how to stitch these three videos simultaneously. ” Today, new AI technologies do not require viewing the face from the front to recognize a person. It can still detect the person from any angle, even if the person turns around. If someone else enters the room, the camera understands and expands the view to include that person. As a result, Altia Systems enhances the experience by expanding the distance in which these algorithms work and improving the accuracy in the identification of people. “Altia Systems figured out how to do it in real time at the camera level inside the device”