An Engineering Challenge: Video Analytics vs. Photoelectric Beams

I love stories like this. I was talking the other day with a fellow named Thomas Keller, who is a VP for Teecom Design Group, which does engineering and design work for security, telecommunications and audiovisual systems. He was describing some of the pretty extreme systems they have set up for securing residences–concentric rings of protection that encompass photoelectric beams, buried cable detection systems, surveillance cameras concealed in birdhouses and behind fake ivy so the VIPs don’t feel like they’re living in a prison camp, the whole nine yards of perimeter protection. It got really interesting when I asked Keller about video analytics, which I reported on last summer. (See “Look Smart: Video content analysis is getting better all the time, but it’s still new enough that buyers should proceed with eyes wide open.”)

Come to find out that Keller is what you might call a late adopter of video analytic technologies. Why? It has nothing to do with expense or complexity. About three years ago, his group did a “shoot-out test” where they took a single surveillance camera on a person’s property, ran the video stream through a splitter, and allowed seven different video analytics companies to process the data. All Teecom wanted to know was whether a person had run into the field of view of the camera. They didn’t want to know about clouds overhead, or rain drops, or a golden retriever. Just a real live person. For comparison’s sake, they also put in a photoelectric beam that would be triggered when someone or something crossed the same target area. Then they let the whole thing run for three weeks.

You already know who won, right? “The beam was the best performer,” Keller said. It wasn’t perfect–sometimes a branch had blown into view or something else triggered a false alarm. But, he said, “When there was an alarm, we could clearly look at the video from that trigger and identify what happened. We had qualitative information that showed us exactly why that alarm occurred.”

With video analytics, on the other hand, not only were there false positives, Keller’s group oftentimes couldn’t even tell what the systems were detecting. “We had thousands of alarms where you couldn’t in the recorded view see what triggered the alarm. Those little red boxes were popping up frequently with nothing in there.” Was it light reflecting off the tree? Who knew?

Fast forward three years, and Keller is just starting his first installation involving video analytics. Teecom hasn’t redone the “shoot-out,” but he’s convinced that the technology has come far enough along that it’s worth testing in the real world. One key development: the system they’ve selected analyzes the raw, uncompressed video right on the camera–in other words, “on the edge,” which we reported last September was where things were headed. It’s nice to hear that people are taking a chance on an over-hyped technology that really, by all accounts, has come a long way in the last few years.-Sarah Scalet