Something for Akida 2.0 to get its teeth into, perhaps.
https://www.intellisenseinc.com/innovation-lab/augmented-intelligence/real-time-image-enhancement/
Real-Time Image Enhancement
No longer the stuff of spy movies, Intellisense Systems is developing deep learning-based, super-resolution algorithms that can enhance and clarify images in almost real time. Learn more about this innovation and how it is making military, rescue, and other hazardous operations safer.
“Zoom in and enhance.” These words have become ubiquitous in nearly every spy movie and TV show. Typically, a group of people will gather in a shadowy room surrounded by dozens of screens showing surveillance footage. One of the monitors will zoom in the pixelated visage of a passerby. But real-time image enhancement turns the once blurry freeze-frame into a crystal-clear picture, often revealing the face of a brave hero, or a nefarious villain.
Many viewers may roll their eyes at this cliché, but this technology can greatly bolster the safety and efficiency of intelligence or combat operations. Real-time image enhancement can not only help differentiate friend from foe, but it can also identify key items in people’s hands, as well as clarifying marks on vehicles or structures. Getting a clearer picture of people or the environmental improves the U.S. intelligence and ensures that targets are acquired while civilians remain safe.
With this goal in mind, the United States military solicited work from the Department of Defense’s Small Business Innovative Research (SBIR) program to develop a means of advancing real-time image enhancement technology. After successfully proposing a solution to this requirement,
Intellisense Systems developed and demonstrated the ability to enhance images in low-light and nighttime conditions based on the novel use of a convolutional neural network (CNN). A CNN consists of layers and algorithms that mimic biological neurons, and it requires relatively little pre-processing compared to other image classification algorithms. This method of processing is ideal for image enhancement and video analysis; thus, the Intellisense team implemented into embedded hardware to help servicemembers in both recognition classification and laser pointing.
The machine-learning specialists at Intellisense devised this real-time image enhancement system to detect military-relevant targets in both still images and live video. Using a tablet, the software could automatically identify key elements in either a photo or video and present bounding boxes, icons, or color highlighting to key points of interest. The operator can then select an area of the image or video stream for enhancement. From here, the CNN begins its layer processing, increasing the image’s resolution and improving its contrast, acuity, and stability.
It can eliminate motion blur or make out certain items that were previously undetectable, like a weapon in an enemy combatant’s hand or text on a mobile phone.
To further relieve the effort required by personnel, the CNN employs unsupervised learning and processing methodology.
This means the network uses datasets as examples. As a result, it processes data without a specific answer or outcome to identify. Instead, the system can automatically determine the structure of the data (in this case, a still image or live video) without human input. This enables the CNN to identify patterns based on the datasets and autonomously enhance pictures and video, taking some of the burden off the system’s human users.
This system’s processing can be completed via a tablet to bolster the ease of use and mobility. It is compatible with the next generation of the U.S. Armed Forces’ handheld targeting system, which can be packaged into compact housing and mounted into a variety of locations. This innovation enables remote viewing and control via radio network. Additionally, the open-architecture approach makes this solution compatible with various software implementations. This enables real-time image enhancement algorithms to be developed independently of the armed forces’ new handheld targeting system.
With this solution, service members can gather intelligence and acquire targets with greater success and efficiency, all while keeping civilians and non-combatants out of harm’s way. And with funding from the SBIR, this solution can be commercialized to serve in other applications, like search-and-rescue missions and emergency response. Over the next few years, Intellisense will continue to train and refine these CNN algorithms so that real-time image enhancement can greatly improve decision-making, reduce users’ workload in detecting key information, and most importantly, save lives.
TT
https://www.intellisenseinc.com/
) says:
