Two new projects target ads at your car
In the UK, and soon in Japan, real-time car recognition tech is guiding ad targeting on digital billboards.
Two new and separate advertising projects are based around the idea that “you are what you drive.”
The new efforts — one in locations around the UK, the other near Tokyo — are identifying individual car types and directing ads to the implied interests of the driver or passengers. They are the newest additions to a growing category that recently added a campaign where Dannon Yogurt ads on digital billboards are displayed according to four levels of traffic speeds.
This week, London-based media owner Ocean launched its first campaign — created by the ad agency Publicis for the new Renault Mégane car — employing its vehicle recognition tech for contextual advertising.
In the campaign, which follows 12 months of testing and development, digital billboards at the exit of a traffic circle near a shopping center in Holland Park in West London show messages directed at cars stopped at a traffic light (See photo on the top of this page).
The images of the fronts of stopped cars in the three lanes of the traffic circle are picked up by three cameras on the left side of the left-most of three screens, with one camera focused on each lane.
Image recognition identifies the make, model and color of the vehicles from their license plates. The license plate is matched anonymously to the vehicle specs through a public database, and the corresponding creative is served in real time.
Any of dozens of text messages are displayed on two of the three billboards. They may both be directed at one vehicle, or at different vehicles. The third billboard shows regular ads.
In a statement, Ocean Outdoor Head of Screen Investment Kevin Henry said the ads can be “for a new model of the same car, or [they] can be leveraged to launch a new product that’s relevant to a particular driven demographic.” The ads could also be for a rival car model.
A key benefit, according to Renault, is that they can — for the first time — speak directly to out-of-home drivers.
Henry told me that the campaign is looking for convertibles, SUVs, hatchbacks and “saloons,” which Americans call sedans. There are currently 30 different text messages, which might start: “Hello, you in the black convertible… ” The campaign is based loosely on the popular car journey game, I Spy, where one passenger describes something only by a clue, such as the color, and the others must locate it.
Henry said that no one else is matching cars to license plates for ad campaigns, and he added that Ocean doesn’t store the license plate info or match it with third-party data about the owners.
The Cloudian/Dentsu project
It does, however, maintain a database of vehicle types at that location. Advertisers are only charged when a message is shown to a targeted vehicle type. Since the campaign has just started, Henry said, there are no results yet. The technology has also been tested on similar setups in Manchester and Newcastle, where campaigns will roll out next month.
Over in Japan, a collaboration between digital storage provider Cloudian and ad agency Dentsu wants to show ads targeted to specific cars passing by a highway. Intel and Quanta Cloud Tech are providing infrastructure support for the project, which has recently completed a proof-of-concept phase.
From fall of last year through this winter, Cloudian developed and trained an artificial intelligence-powered system that recognizes several hundred kinds of different car makes, models and years, including hybrid versus normal gasoline types of the same model.
The idea is that an ad for an eco-friendly product might be shown on a roadside digital LED billboard if a Honda Prius hybrid car whizzes by. If it’s a long-distance truck, the ad might be for a refreshing drink.
The system is intended to identify the vehicle types, and then select and display an appropriate ad, all within the few seconds when the cars are within visual distance.
The ad will be shown for five seconds, and the expectation is that the driver will see it for at least half a second. If no targeted ad is displayed, the regularly scheduled one is shown.
The companies say the vehicle recognition accuracy currently varies by model types, such as an 82-percent accuracy level for a Honda Fit Hybrid 2010 and a 98-percent level for a Honda Fit 2001. Colors don’t matter.
To date, however, the image recognition testing has only been conducted on recorded traffic video (see below) and still images, not from a live video feed of real traffic captured by billboard-mounted cameras, according to emailed answers from Dentsu Senior Manager Ichiro Jinnai and Cloudian Co-founder and President Hiroshi Ohta.
Cloudian CMO Paul Turner told me that the training requires three to five thousand images for each car model, and, once trained, the recognition will be conducted in real time. The assumption is that when an image grab is made from actual traffic, the recognition process will be similar, but that remains to be seen.
In the fall, a field test is planned for one billboard along an expressway in Tokyo, where an all-weather camera with a telephoto lens will be mounted for capturing images of cars 600 to 800 meters away. Vehicle type and traffic speed can be determined within a second. A commercial rollout has not yet been determined.
Cloudian is handling the image recognition, and Dentsu will handle the ads and billboard. When I asked Turner why a digital storage company would be interested in a project like this, he indicated the intent was to demonstrate his company’s ability to turn big data into smart data.
The companies have been vague about which specific ads will be shown for which car types or how a car type is matched with data determining the most appropriate ad. That may be at least partly due to the fact that the field test has not yet been conducted and that, according to Turner, no actual advertisers are yet involved.
It also is not been clear what happens when a dozen cars speed by, and all are targeted vehicles. Are some selected and grouped for a single ad display, such as might be the case if three hybrids were among the dozen? Or does the system wait until traffic has thinned out, when only one or two cars go by?
The companies are also vague about the capture of license data. Turner first told me it wasn’t captured, but then, when I pointed out that it was obviously captured and probably stored, he said it wasn’t being used. The companies indicated that drivers, passengers or other attributes — such as hauling a surfboard on a roof rack — are not currently added to the car profiles.
Neither the Cloudian/Dentsu nor the Ocean projects allow drivers to opt out.