Page 107 - Security Today, March 2018
P. 107
Fear not, dear reader, for deep-learning analytics are poised to offer a level of accuracy and reliability in object and behavior classification that delivers on the lofty claims made in the past. The ability of deep learning algorithms to view a scene intuitively, as a human viewer would, means that detection accuracy increases dramatically, while false alarm rates fall.
What’s most interesting and intriguing about this process is that there is not a single program allowing these analytics to be performed. Instead, programmers are creating learning algorithms and exposing the systems to terabytes of data. In essence, the system is being trained, and then figuring out for itself how to recognize the desired objects, faces, or behaviors.
While this concept isn’t new, what is new is the ability to harness the vast computational power, and the enormous storehouses of data are essential to making analytics work well. Deep learning algorithms improve accuracy to as high as 99.9 percent, according to IHS Markit. It can also reduce time spent looking for relevant video footage by processing and analyzing larger volumes of video data in shorter time frames than conventional technology.
Surveillance industry leader and innovator Hikvision has been working with deep learning analytics for the past two years to ensure the final products they bring to market perform at a much higher level than people are accustomed to. Given analytics’ relatively poor reputation, Hikvision is taking the time to understand and refine the technology to ensure that they come out of the chute with video analytics products the industry can rely on.
“Hikvision is leading the way with deep learning products that provide users with the most reliable and accurate video content analysis,” said Doug Gray, Hikvision’s senior product manager. “Hikvision’s new DeepinView security camera series and DeepinMind NVR series of products apply deep learning algorithms to deliver some of the most accurate video analytics in the market today.”
Mr. Gray points out that verticals with critical infrastructure type of projects, such as power plants, airports, and high-risk facilities – as well as key sectors in multiple verticals – will be the first to realize the benefits and get onboard with these higher performing analytics products. He further believes that verticals such as healthcare, education, and retail will follow as it meets better price points.
He is clear in pointing out that this trend is simply at a starting point. He adds, “The point of deep learning technology is that it will be constantly improving as it learns. The product itself – using the same technology, the same algorithms, and the same processors – it is going to get better. The possibilities are very exciting.”
Hikvision is developing many deep learning vertical market applications that help with people counting, crowd control, entry/exit management, traffic and vehicle management, license plate recognition and more. Accurate human detection, face detection and recognition, and behavioral analysis are all improvements created by Hikvision’s application of deep learning technology.
Mr. Gray concludes, “All of these applications will benefit from the accuracy of the algorithm and the analytics, offering enhancements to even human detection. We will be able to accurately filter characteristics of those humans being detected: clothing, gender, age, accessories, bag, purses, glasses, you name it. Continually enhancing, better performing, and always learning.”
ScanSource Networking and Security Focus SPRING 2018 | 2