The technique of using algorithms to analyze footage from video surveillance cameras in real time began coming into its own five years ago. It's an intriguing adaptation of standard surveillance security. It's still an emerging market, as Jon Cropley, principal analyst at IMS Research, told CSO.
CSO: What trends do we see in video analysis versus five years ago, when the edge-based market was emerging and there was a push to embed algorithms into the processor?
Jon Cropley, IMS Research: There are still a lot of installations more suited to server-based solutions, though we are seeing more edge-based systems. The number of algorithms that can be embedded has increased, and they've been refined to be less processor-intensive.
[Also read PSIM: physical security information management]
Also, processor power has improved. The combination has meant that a number of algorithms that couldn't be embedded before are now a possibility. The embedded market has grown faster than the PC-based market.
What are the biggest markets for video analytics?
The biggest application is still intrusion detection. Systems can be simple, or they can be like the one at a palace in the Middle East located very close to a coastline. Video content analysis is being applied there to scan the coastline for ships that shouldn't be there. You need to deal with the environment, waves and tidal patterns, seagulls.
"China has put cameras at regular intervals along the high-speed train line. Those cameras feature VCA. If they assess an object that is going to damage the train, it is automatically decelerated."
Jon Cropley, IMS Research
Are there new uses that have emerged over the last few years?
An application that is talked about a lot is abandoned-object detection. Those are becoming more realistic. You're seeing video content analysis (VCA) used at airports—when you leave the plane you're supposed to follow a certain path and then leave the airport. You can detect if the person is going the wrong way.
Using VCA in a crowd is quite interesting and there's a lot of work for making it more suitable for crowds. Prisons use it for the opposite, almost, of intrusion detection—they're looking for people trying to break out.
The assumption is that it's for the professional market, but increasingly the DIY consumer-based video solution is being offered. It's mostly for intrusion detection or to see who's at the door. That's an interesting new departure.
In China there is a high-speed rail the government has introduced there. The train moves so fast that the stopping distance can be over five kilometers. That makes it impossible for driver to see an obstacle and stop by himself to avoid the obstacle. So they've put cameras at regular intervals along the train line. Those cameras feature VCA. If they assess an object that is going to damage the train, it is automatically decelerated.
What about false alarms?
If you get a lot of false alarms on that kind of system, the high-speed train can very quickly become a low-speed train. They only look for things that will damage or derail the train.
I talked to a company that installed VCA at an airport and they were really struggling with false alarms due to birds. They needed to combine VCA with a sensor on the fence looking for movement. When an alarm was triggered by one, it needed to be verified by the other. The VCA was used for verification of the alarm from the other sensor.
Is this functionality affordable?
Prices are going down by around five percent a year. I think there's a growing acceptance, but also still reluctance to buy, in part because technology is changing so fast, but also, when VCA was introduced, it was oversold a little bit in one or two cases. Because of that, it's got this kind of reputation that the technology doesn't work as well as it should. There's still a little bit of that in the market.
One thing VCA has struggled with is that most installations aren't monitored live. You have a building, a warehouse, and you're not looking out for live intrusions. In those cases, VCA doesn't offer a benefit to the end user.