February 06, 2011

How Kraft’s Face-Scanning Tech Will Tell You What You Like to Eat

Face scanning kiosk recommends recipes and meals for you. The most unique selling point for the Meal Planning Solution is its integration of anonymous video analytics software, which detects age and gender to determine which recipes a person may be interested in viewing.

How Kraft's Face-Scanning Tech Will Tell You What You Like to Eat

After a harried day in the office, meal planning may be the last thing on your mind. It might be evident on your face, though.

That’s the premise behind an interactive technology Kraft and Intel recently introduced called the “Meal Planning Solution.” The kiosk-like display, which is likely to show up in at least one retail location this year, is meant to help weary shoppers find new recipes during last-minute grocery trips.

Forget filling out a profile of favorite recipes and ingredients, though — this bad boy scans your face to figure out what you may be interested in cooking. For even more customized results, a user can choose to pull in purchase history data through her grocery store loyalty card, her shopping list on Kraft’s iFood Assistant mobile app, or her recipe-browsing history on kraftrecipes.com.

To get a better look at the research and technology behind the solution center, we spoke with Don King, Kraft’s VP of retail experience, Chris O’Malley, Intel’s director of retail marketing for the embedded and communications group, and Jose Avalos, Intel’s director of retail and digital signage for the embedded and communications group. Read on for an in-depth look at how the solution center works and what Kraft and Intel aim to accomplish with it.

What’s the Point?

The idea of having a machine scan your face and then scoop up all the products it tells you to buy sounds weird, to say the least. So, when we heard about this solution center, we were a bit skeptical of its usefulness.

It turns out there are some solid applications. Intel and Kraft were collaborating over much of 2010 to build the display, and in doing so, they’ve come to solid ground on what problems it’s trying to solve for four groups of individuals:

Retailers: “In terms of managing their revenue on the top line, for retailers, it’s all about basket size and foot traffic, or returned trips to that particular store,” says King. “That lends itself to having a retailer provide a service that nobody else offers.”
Consumers: “Everybody’s got shopping lists, but they tend to be disorganized with a sticky note here or a scrap of paper there,” King noted. “Approximately 70% of the time, somebody will go into a store to find items for dinner that night, and they haven’t completely decided what they’re going to make. So, they’re buying the items on the fly and then going home to make them. Lastly, the average cook isn’t, frankly, that accomplished — they have a mental Rolodex of seven to 10 things they know how to make well enough that it’s not stressful for them and that will be good enough that the family will eat them and enjoy them. The question, then, is what can we do to help with that meal planning process?”
Kraft: “Our objectives are to increase shareholder value of our products by selling more of them,” explains King. “We have an outstanding resource with our Kraft’s Kitchens Group that develops recipes for our products, alone and in combinations with our others brands and others. We’d love to get those recipes into people’s hands as they need them.”
Intel: “At Intel, we think the technology — anonymous video analytics along with the immersive digital experience in-store, has huge potential,” says O’Malley. “It’s one thing for Intel to say that, but to actually find key brands and retailers in the industry who also think that this ability to personalize an experience to the end-customer is important, is more compelling.”
With those goals in mind, Kraft and Intel moved forward with creating the solution center.

Behind the Video Scanning Technology

The most unique selling point for the Meal Planning Solution is its integration of anonymous video analytics software, which detects age and gender to determine which recipes a person may be interested in viewing. “The technology is approximately 86% accurate in determining gender,” Avalos explain. “For age, it depends on the bracket. For young adults and adults, the accuracy is approximately 70%. For children and seniors, the accuracy is approximately 80%.”

Avalos got technical for us about exactly how the video scanning technology works:

“Sensors send real-time video feeds to Intel’s anonymous video analytics (AVA) software, Intel AIM Suite, which scans a square NxN pixel region across the video frame to detect whether it contains an arrangement of pixels that resembles the general pattern of a human face. This entails looking for coarse intensity information such as eye regions being darker than the rest of the face, the nose region being brighter than the rest of the face, the two dark eye regions being roughly on the same horizontal line, etc. For all candidate frames that pass the above criteria, the frame is considered to contain a face, and impression data can be gathered.

“Intel AIM Suite face detection algorithms have statistically learned the pattern of a human face by being trained on an audience database of thousands of pictures of human faces,” he continues. “The demographics of a face can be determined using a similar process to how a face is found, where the patterns being looked at correspond to male or female faces, or certain age brackets (children, young adults, adults, seniors). The algorithms have learned which face features have the strongest weighting for each gender and age bracket. The combination of various facial features such as eye positioning, nose shape, cheek bones, and overall facial structure are among the variables that are taken into consideration during this process.

“All sensor video data is processed and destroyed in real-time. No images or uniquely identifiable information are stored.”

Avalos also noted that the sensor also collects data on how many people use the solution center and the amount of time a person spends at it.


How Kraft's Face-Scanning Tech Will Tell You What You Like to Eat

Posted by keefner at 09:05 PM