Picture yourself behind the wheel of a minivan. As you go whizzing down the highway at 55 mph, you approach a billboard. It changes its display ad, now showing an image for the next Pixar feature film. The kids in the back seat squeal approvingly.
The technology to do that has already been proven. Cloudian, a California-based smart storage outfit, has teamed up with Japanese ad firm Dentsu Inc and several tech companies to develop a smart billboard system that will change ads to suit the driver of the next car coming down the highway.
“We proved the concept with them (Dentsu), “ said Paul Turner, CMO at Cloudian. “Now that we can do this, how can we make this effective?”
The system, based on the Cloudian HyperStore OS, starts with a video camera, mounted near an LED billboard, aimed at the highway. As a vehicle is “seen” by the camera, its digitized image is sent to a database, where it is compared against a library of similar images for that car type. All this took about six to eight months to engineer, Turner said.
Artificial Intelligence will recognize the car type, and then call up an ad that should pitch well to the demographic associated with that car type. All this should take about one second.
“We proved that the prototype can recognize cars based on learned images,” Turner said. Teaching the system to absorb the images is another story. About 3,000 to 5,000 images of a single car type is uploaded into the system. The built-in AI then is taught how to associate that image with the car type, regardless of the car’s color or what angle it is viewed from.
Loading several thousand pictures of every car type is a good example of Big Data. Cloudian’s solution involves turning that into “smart data”. That can involve classifying the data, attaching tags to images and cataloging the meta-data for the AI to sort through and eventually recognize by comparison against the visual image seen on the highway.
“We’re teaching the algorithm get smart at recognizing different kinds of cars,” being able to distinguish different makes and models, even differences in the model year, Turner explained. The system will then note the percentage likelihood it has made its identification correctly.
“Building the infrastructure took up the bulk of the time,” Turner said. “It was a matter of weeks, not months, to train the algorithm” Turner stressed that this system was built around visual recognition of car types. The system is not reading license plates, then combing through motor vehicle registrations to make a match.
Right now, the technology stands at the “proof of concept” stage, Turner noted. He expects Dentsu to try certain ads to reach certain demographics to see what response rates they get. Someone behind the wheel of a Ford F-150 pickup truck is likely to have different preferences than the driver of a Fiat 500 or an Audi A-8.
Starting this week, Dentsu and Cloudian will begin another round of testing tofurther gauge HyperStore’s capabilities. Come July, testing will be done with an LED billboard plugged into the system. Results are expected in August. The launch date for the “smart billboard system” has not been set yet, Turner added.
It is most likely to debut in Japan, where the culture is more accepting of the fusion between advertising and technology. So what kind of person drives a Nissan Note and what can we sell them?
