Thermal Customer Counts Get Cold Shoulder From The Limited's CIO

Many retail stores rely on customer-counting systems triggered by body heat despite the fact that (as proven by the case of roasted chickens being confused with infants) the technology has its shortcomings. Up until this summer, The Limited, a 240-store clothing chain with outlets in malls and shopping centers, was one of them, said CIO Roger Coville.

Coville said the chain suspected the thermal-imaging system wasn't too hot when it came to accuracy and, convinced by a pilot project that compared the system's capabilities for truthfulness with those of new equipment that relies instead on analysis of video data, The Limited decided to leave thermal-imaging out in the cold. Both Coville and an executive at the company that supplied The Limited with the new system said the thermal system proved to be more inaccurate.

"It's my understanding [that thermal imaging] is the standard out there," Coville said. "There's a pretty wide installation of that equipment, and I'm sure it's getting better all the time." But in his testing, video-based counting was 20 percent more accurate than its thermal predecessor. Oh, the CIO added, it's also a lot cheaper, which makes it a hard combo to ignore.

"When we first deployed this in pilot stores, our counts were dramatically different" than the thermal-imaging system's counts, said J.D. Story, a marketing exec with Digiop Technologies, which worked on the project. "I got a call saying the counts were way off."

Because the system was video based, Story had an easy way to find the truth. "I said 'Look, we've got the video. We can view an hour of foot traffic in six minutes, so let's just count the people.'" He said the new system's counts turned out to be accurate at least 95 percent of the time, while the old thermal system was accurate only 60 percent of the time.

The new counting system is combining the more accurate customer counts with data from the stores' POS systems to gain better knowledge of individual store performance, said Ed Troha, managing director of global marketing for another vendor in the project, ObjectVideo. "With that kind of data, [The Limited] would be able to make more intelligent decisions about its operations, including staffing and conversion rates, the kinds of things that enable them to squeeze the most out of every store location."

The new system provides standard video surveillance in conjunction with video analytics installations. It costs about $4,000 per store and about $5 per store per month in ongoing service fees, said The Limited's Coville.

He said the monthly fee is about six times less than the fee the chain was paying to the thermal-imaging company. "That's the biggest thing in terms of cost, month after month after month," he said. "Beyond that, what we were excited about is the fact that we could use same system to do loss-prevention work."

Story of Digiop Technologies said thermal systems sometimes have difficulty differentiating between individuals and groups of people entering a store. They may falter when it comes to counting the number of people in those groups or determining if they are entering or leaving, he added. The system is not based on facial recognition technology, something ObjectVideo's Troha said "is a database approach that is not very accurate."

Instead, it uses what he described as "pixel-based video analytics" to study the video streams being captured by the cameras. "It can determine reliably if a group of pixels is a human or not," Troha said. "Once it identifies it as a human, then it will count it."

The system sends all information to The Limited's corporate headquarters in Ohio for aggregation and analysis, but the data is also kept on-hand at each store for conversion-rate calculation, Coville said. In addition, it has the capability of doing more than mere body counts.

Although The Limited opted to not initially use the other capabilities, Coville and Story said the retailer can activate functions that will measure how long people stay at in-store displays, provide data on store occupancy rates at given periods of time and tell employees in the stockroom, via alerts over headsets, that they are needed at the POS when lines at the register get too long.

"It can do shrink reduction and analytics," Story said. "It can say there was a refund given with no customer present at the register or that there was a manager override but no manager there. It uses the same cameras to catch slips and falls and to count people, and it can help tie employee bonuses to their conversion rates."