Although technology writers are just as distracted by shiny things as you are, we are trained to ask lots of business questions. Ordinarily, a journalist trolling a trade show has a stock set of questions to ask vendors: "Is this shipping? How much is it? Who's the target user?"
It's a relief, sometimes, to focus on a technology and its possibilities – not necessarily on its business case. And based on research projects I saw in early March, plenty of cool research is underway to make any self-respecting geek squeal with delight.
The CeBIT conference held annually in Hannover Germany is beyond huge. Where most trade shows would be pleased to fill up one or two conference halls, CeBIT has dozens. One entire building is given over to security products, for example, and another to Internet tools (such as e-commerce and content management). My favorite, though, was Hall 9: Research and Development. Most of what was found inside were booths from universities and other organizations demonstrating their work-in-progress – even if the software was still in the "Hmm, what shall we do with this?" stage.
Herein, I share with you my favorites among the research projects I saw. (Forgive the iffy non-staged photo quality; that's what you get from snapshots at a trade show.)
Facing the future
Two facial recognition projects caught my eye. Both made me think, "Please use this power for good and not for evil."
The Karlsruhe Institute of Technology used clever geek bait to attract me to its booth: video clips from The Big Bang Theory. Formally, the project "focuses on the exploration of new methods in computer vision to enable detection and analysis of faces and people in both images and video."
Or more informally: The project demonstrates the software's ability to recognize an actor's face, and then mash up that data with other resources. For instance, touching the screen can bring up the actor's record on imdb.com (destroying any argument with your spouse that begins, "Really, that's the same actor who was in an episode of Firefly!"), or enabling better video search results (such as, "Show me all the scenes with both Sheldon and Penny").
As with all these projects, it isn't quite "cooked" yet. The researchers explained that the software is currently about 80% accurate, and it takes an hour or two to process each TV episode. And that's with it paying attention only to the main characters; if you spot a walk-on character in the background of a bar scene, you can't look up whether that really is the same actor you saw in another show. (I guess that means I have to continue arguing with my spouse.)
Still: This research has possibilities – far beyond TV (though I know people who'd pay good money for a more-accurate version, right this second). Anything that lets us improve video search is a godsend for semantic indexing of multimedia data, to begin with. And as the research material suggests, identifying users lets software automatically select tailored, individual settings (such as recommending TV programs). Plus, they add, "Computer vision allows [us] to glance into the minds of people. For example, estimation of the viewing direction allows [us] to find which objects a person is interested in and hence attract his/her attention."
I'm in a good mood. The software says so.
The southern German university isn't the only one working on facial recognition. Fraunhofer, Europe's largest application-oriented research organization, showed off several research projects including browsing the libraries of the Bavarian State Library and using smartphones for key access. But the one that attracted my attention was the facial recognition system that incorporates input from "Felix the robot."
Felix is used mainly to mimic facial expressions (sad, happy, surprise, anger), making it easier for humans to recognize what "happy" looks like so they can incorporate it into software solutions. When you smile at him, he smiles back; robots and avatars could assess a user's mood and respond appropriately (which is more than I can say for skill of the retail clerk I just encountered).
The Fraunhofer software used a camera trained on the audience to demonstrate – in an effectively crude way – how it recognizes facial details including gender, estimated age, and mood. It's not perfect either; on Monday morning, it claimed I was a teenager, but by Wednesday afternoon, with trade-show fatigue, it identified me as an adult (damn it).
The researcher I spoke with suggested several possible commercial applications. Advertising applications might identifying a human approaching a store window or train platform and pop up relevant ads (perfume versus men's underwear). It can measure the time of visual attention and fixation, too, so an advertising company can know if it's failing to attract interest. One organization is experimenting with an automobile application that recognizes driver stress or distraction. Imagine if the driver age can be identified, and have the car go into a "safe mode" if it's a teenager. For security, the software can serve as short-term anonymous memory, since a face that appears multiple times (such as a stalker? I didn't ask) can be recognized and recollected.
Austria's Guger Technologies demonstrated an odd-looking gizmo that looked like a bathing cap connected to an EKG. This is more than a "What could we do with it?" presentation, however: It's meant to do very specific brain scans and measure recovery progress after someone has had a stroke.
Related technology using a brain-neural computer interface (BNCI) lets someone with ALS or other physical disability "paint" on a screen by looking at it. There's a personal EEG-based spelling system, too, which aims to be a complete patient-ready system for spelling, communication, and control. It's designed to be installed and operated by the patient's caregivers at home (and I think may be available for rent or purchase at this point, though that wasn't clear). The user sequentially selects characters from a keyboard-like matrix on the screen by paying attention to the target for several seconds. Most patients can get it up-and-running in 10 minutes, I'm told.
The main goal of the Back Home project is to help end users who are affected by motor impairment due to acquired brain injury or disease to achieve tasks which would be difficult without depending on a caregiver. The interface connects patients' voluntary physiological responses, including the brain's electrophysiological signals, with applications to facilitate the activities of daily living. Since so much of what we do can be accomplished via the Web, nowadays, any interface that removes typing or physical interaction is a godsend.
A little less Awww, but no less impressive is a project from the Distributed Artificial Intelligence Laboratory called "Clever LOGging of Activities" (CLOG). Its visible display was a bicycle tied into a Google Earth map with a standard smartphone. Exercise bikes tied to Google maps aren't exactly new (you can hack your own), but this one makes it harder to pedal uphill when the map data suggests it (Alps, anyone?) and tracks calories used.
At least that's the demo. CLOG is a mobile application that tracks a human's activity throughout the day, whether everyday movements or dedicated training sessions. The software gathers data from both internal smartphone sensors and wireless sensors, making it easier to track and optimize health efforts, define goals, and, say the researchers, to view your activity history in dynamic heatmaps, charts, or live widgets. CLOG can connect to the Common Health Model (CHM) backend for data synchronization with sensor data in other services.
Shopping without dropping
Not every research project comes from a university. Among the cooler projects was one from IBM Research called Augmented Shopping (video).
If you're a label-reader or otherwise do product comparisons before buying, you know how time consuming it is to look at every box on the shelf: Does this gluten-free cereal contain rice? Is this item vegan? Which of these options has the lowest carbs? Which is cheapest, when measured by the ounce? The IBM Augmented Shopping App, developed by IBM Research in Haifa Israel, takes a photo of a store shelf, and superimposes filters on the images to sort the products by relevant characteristics, such as price or ingredients.
You don't need to point a QR-code reader at a specific box; the app looks at the image of the whole product display and does its database lookup (in the cloud, naturally) automatically. Hold up your smartphone, filter the results by ingredient ("no peanuts, please"), and snag the chosen box off the shelf.
This research isn't too far away from being deployed in the real world. IBM's Amnon Ribak, who developed the app, explained that a U.K.-based grocery chain is setting up to test a customized version of the app in a few stores. That's a rather brilliant way to enhance a retail loyalty program, given the wealth of mobile apps that help shoppers find merchandise cheaper at nearby stores or online.
For all the time we spend grumbling about technology, and the death of innovation, and other downers, it's good to be reminded that smart developers are still using computers to improve the quality of life.