Surveillance systems in the future will be smart enough to learn your habits and recognize when there are anomalies to them, suggests Camio, a company that makes use of deep learning in security.
Surveillance systems in the future should be smart enough to learn your habits and recognize when there are anomalies to them, suggests Camio, a company that makes use of deep learning in security.
“Longer term, Human Computer Interaction (HCI) is moving beyond keyboards and apps. Video enables ‘ambient computing’ that could coordinate cloud services by observing the video and audio in real-time,” said Carter Maslan, Co-founder & CEO of Camio. “The elderly mom living alone should be able to say ‘help’, wave her hands, or just be absent for longer than expected to trigger a message to her children.”
Camio uses multiple machine learning techniques in different stages of the video analysis pipeline. For example, the first stage eliminates spurious motion detection caused by ‘pixel noise’ from common lighting changes, shimmering leaves, repetitive motion, etc. Even at this first Adaptive Motion Filter stage, there are 10 concurrently running neural networks competing to be more correct than the others based on the event features they've chosen to learn from, Maslan explained.
In its last stage of analysis, Camio uses deep learning to assign event labels like “swimmer” or “sports utility vehicle”.
“Camio is different in that it learns quickly what's important - without making precisely accurate object labeling a prerequisite for that learning,” Maslan said. “Computer vision will likely continue to make mistakes quite often. That's why Camio focuses its ML on highlighting the video events that warrant your attention - even when failing to precisely characterize the event.”
Effective security
The most significant change that Camio brings to security is eliminating the difficulties of manually going through scores of footage to find out important events.
“Camio makes security surveillance simple. Early on, we realized that security video can't be simple without also being smart,” Maslan said. “So, Camio learns to highlight the important video events quickly. It also enables searches like ‘people approaching side alley’ to display the relevant results within two seconds. So rather than unexamined recordings, false alerts, or hours of video to review, Camio highlights important events automatically for any number of users, viewing from any type of device. Camio also makes it simple to share and collaborate across all types of cameras and NVRs, so marshaling a coordinated response to a neighborhood safety concern is easy.”
Target market and factors driving demand
Camio mainly targets homeowners and small businesses, aiming to become a B2B platform for real-time search of the real world.
“Consumers/prosumers at home and work were the best starting point, because their demands for simplicity and low cost imposed important design constraints on the creation of Camio as a platform,” Maslan said. “The major applications are home security, office security, and at-home care. But our very first major customers were dog parents at work keeping an eye on their pups at home.”
In fact, in the short term the company expects such interest from entry-level market to drive the demand. Maslan said that when people receive their first alert like “dog approaching lawn”, they are thrilled.
“All of a sudden, it feels natural to expect the real world to start talking to you!” Maslan said. “The demand is already there once people see a path that's simple and inexpensive.”