Machine learning in surveillance
The video surveillance industry has enjoyed fast growth and revolutionary new technology over the last decade.
Not that long ago, analogue cameras dominated the industry and recorders were rarely accessible over the Internet. Today the majority of cameras installed are IP and it would be almost unheard of to install a professional surveillance system that the customer could not view remotely from anywhere with an Internet connection. IP surveillance systems have opened a world of possibilities to manufacturers and integrators.
As with any industry experiencing above average growth and innovation, the surveillance industry has attracted the eye of outsiders who want to capitalize on the opportunity. Video analytics and machine learning in particular have spawned dozens of new start-ups and have even drawn the attention of large corporations who want to invest their time and money in the video surveillance space.
Technology like machine learning is evolving at astounding rates, thanks to companies like Google, Amazon, Apple and more who continue to incorporate it into every platform or product they have. While these massive companies focus on applying machine learning to their existing platforms, smaller start-ups are looking for untouched markets where machine learning is less mature and the competition is not as fierce. The video surveillance industry offers just such an opportunity.
Machine learning is the ability for computers to learn something that was not specifically programmed into them. For example, anytime a website or streaming platform recommends similar products to you, this is a function of machine learning where the platform is making recommendations to you based on what it has learned about your online behaviour as well as the behaviour of other users with similar interests to you.
For video surveillance, machine learning is most applicable in the area of video analytics, and this is where we are seeing existing companies as well as many start-ups focusing their energy. The value and possibilities of applying machine learning to video surveillance footage are endless. The most obvious benefits would be the ability for cameras to learn what they are looking at and then react to anomalies or produce detailed reports on the type of activity they observe. For instance, a camera in the mall may learn what areas visitors typically congregate in and could alert security if a crowd develops in an area that is usually quiet.
Data acquired through machine learning in video surveillance could ultimately be more valuable for marketing and operational use than for security, but with all this potential, many new entrants to the space do not understand all the nuances of the industry. For example, tech start-ups might expect to charge high fees for their services, but pricing in the surveillance market is extremely competitive. Where other markets might easily justify service fees in the thousands of dollars per month range, the surveillance industry is closer to the alarm monitoring space where end users expect to pay a few dollars a month per device.
Another challenge industry outsiders often overlook is that surveillance systems are most often installed by technicians with fairly basic IT skills. This makes it difficult for technicians to learn or understand how to configure highly complex programs like those required for machine learning. Lastly, processing in the Cloud is the easiest way for start-ups to scale without investing in hardware, but for video surveillance it is not feasible to constantly stream high frame rate and high resolution footage to the Cloud. For start-ups to succeed in the surveillance industry, they need to understand the requirements to perform processing on the edge through devices that are not overly complex to install.
Machine learning is quite obviously the next big innovation opportunity for the industry. For manufacturers who already compete in the space, their biggest challenge may be proving they are different. For start-ups or otherwise new entrants to the space, the learning curve may be the most difficult part. Understanding how integrators operate and what end users expect will be key to the acceptance of their new innovations. For both new entrants and existing manufacturers, machine learning offers an exciting opportunity to get away from commoditized hardware and instead focus on something that truly has endless possibilities.
Colin Bodbyl is the chief technology officer for UCIT Online.
Central Station Monitoring Symposium
October 17, 2017
Security Canada Central 2017
October 18-19, 2017
October 27, 2017
November 22, 2017
Focus On Health Care Security
December 6, 2017