AI software accelerates pattern search, facial classification
Representing more than 10 years of development, the software, called Brainchip Studio, uses a spiking neural network. a type of neuromorphic computing that simulates the functionality of the human visual tract. The technology allows the software – which includes advanced facial detection, extraction, and classification algorithms – to work with low-resolution video, and requires only a 24 x 24-pixel image to detect and classify faces.
According to the company, the software was able to detect, extract, and classify in real time more than 500,000 facial images during 3 1/2 hours of video across eight different cameras in a recent field trial. In a separate trial, it processed 36 hours of recorded video in less than two hours while extracting over 150,000 facial images.
The software is designed to address the market for such video analysis, which, says the company, has seen little improvement in capabilities despite the huge growth of video surveillance and exponentially increasing amount of resulting stored video data. “BrainChip Studio’s forensic search capabilities tames this massive amount of video information, making it practical for a single individual to search for exactly what they need across multiple video sources,” says BrainChip’s Senior Vice President of marketing and business development, Robert Beachler.
The software can be trained on a single image in milliseconds, enabling rapid searches of video for patterns not previously known. “Unlike current convolutional neural network technologies that require extensive pre-labelled datasets and expensive cloud-based training and acceleration, BrainChip’s spiking neural network can be implemented in software with traditional CPUs and trained on-premise,” says Brainchip founder and CTO, Peter van der Made. “Therefore, BrainChip Studio can be deployed in the field in highly secure environments that may not have cloud connectivity.”
BrainChip Studio operates on Windows and Linux platforms, and is compatible with all major encoded video formats. It is currently available for trials to select law enforcement and intelligence agencies.
‘Face-in-video’ recognition advances with help from Intel FPGAs
BrainChip CEO gives details of neural network architecture
Always-on face recognition promised by new AI chip
The future of video surveillance: HD, hyperspectral and stereoscopic
AI lip-sync algorithm puts words in someone’s mouth