security technology inside high-end server based analytics is expected to remain. Lars Wilson thinks that video analytics at the edge will grow faster than server-based analytics. “It is cheaper to do the analysis in the cameras and you might not add analytics to all your cameras. We support both solutions, but we can see it is has taken off stronger at the edge”, he says. Providing business intelligence Video analytics applications can basically be split into three categories: detecting someone or something while it is happening, for example, search for information after an incident has taken place and finally using video analytics in order to generate business intelligence data. Lars Wilson believes that video analytics for business intelligence will grow faster with applications like people counting and conversion rates and Anders Karlsson agrees: “Business intelligence applications will probably grow faster and become bigger than the other two, perhaps not this year or next year, but soon. However, the other two will grow too, because the technology will become cheaper and personnel guarding more expensive.” – Will deep learning bring a revolution? By Henrik Söderlund deep learning and artificial intelligence are the new buzzwords within video analytics. “A lot of challenges that seemed unsolvable are being unlocked because of artificial intelligence and using deep learning with GPU processing”, says Avigilon’s Senior Director of Global Marketing, Willem Ryan. The market analysis company IHS Markit has listed deep learning as one of the major video surveillance trends in 2017. IHS describes deep learning as the fastest-growing field in artificial intelligence. Deep learning is a training process where machines can improve the accuracy of pattern analysis or classification automatically over time by receiving massive amounts of data. The technology learns what the typical activity is in a scene and alerts the operator when non-typical activities occur. A critical mass Avigilon is one of many video surveillance manufacturers that has great belief in deep learning and recently showcased its deep learning video analytics at ISC West in Las Vegas. Willem Ryan, Senior Director of Global Marketing at Avigilon, claims that the industry has reached a critical point, or a critical mass. “We continue to grow the industry providing solutions that enhance performance, better surveillance, more storage, more cameras, high resolution, and this has created an imbalance between the desire and need for more video information and the lack of human attention to actually focus and view that data that are continuously collected and stored in our system”, he says. The emergence of GPUs (graphics processing units), which provide the deep learning infrastructure to cameras and recorders, has made deep learning possible. Willem Ryan claims that the analytics becomes part of the system without the user having to set up or calibrate it. He says: ”GPUs allow us to create even more powerful analytics that are changing the way that we interact with the video surveillance system. Some of the new advancements that are now being leveraged through GPU capabilities and artificial intelligence, are bringing a new wave of analytics to the industry.” Believes in combined analytics Milestone System’s Regional Sales Manager for the Nordic markets, Lars Wilson, also believes deep learning and artificial intelligence will improve video analytics, at least when the technology becomes a bit more mature. He says: “Self learning and artificial intelligence are misused concepts, but I believe in these kinds of solutions. You want to achieve reliable results and that the system does not only react on the information that you are interested in.” Lars Wilson believes that video analytic systems will develop greater reliability and become more intelligent. He also thinks that analytics will be combined to a greater extent than it is today. “If you look at video analytics in retail, you should combine heat mapping – which gives you movement patterns of how customers move in a store under a certain time frame – with for example people counting.” More cloud services Anders Karlsson, Product Marketing Manager for Bosch in Northern Europe, sees two main trends within video analytics: deep learning and more additional services in the cloud. “Deep learning is perhaps not a big trend yet. But the systems will become more self-learning and a lot of manufacturers are working on this right now. The second trend is to split the analysis: let the cameras and products generate metadata, and then supplemental analysis is conducted Processors must be better Willem Ryan believes deep learning will provide a more efficient user experience and allow the user to search for critical information even in live video, while video analytics runs in the background. He says: “We are going to get more advanced object recognition and detection so you will be able to detect people in crowded areas and detect faces really well. You will also be able to detect vehicles and distinguish between different types of vehicles.” Anders Karlsson says face recognition in real time is still quite hard because it requires very big databases. He also stresses that ”Deep learning is perhaps not a big trend yet. But the systems will become more self-learning and a lot of manufacturers are working on this right now.” via cloud services rather than through a standardised server”, he says. Anders Karlsson is not sure whether deep learning will revolutionise the physical security industry, but he is convinced it will provide a lot of additional services in the future. “Deep learning does not have any limits. Things that users need to tell the system 10-12 times – this is right, this is right, this is wrong – before the system understands, will not be needed in the future because it will be self-learning”, he says. most facial recognition is done afterwards. Except for generating too many false alarms, video analytics requires quite a lot of processing power. Anders Karlsson believes it is one of the biggest challenges for video analytics. He says: “The better processors, the more we can do. It also means that you can increase the resolution of the analysis algorithm, how many pixels it can use. And then, you can perform a lot better and do a lot more video analysis since there is more information.” dete kto r in te r n at i on al • 29