/// Embedded Analytics Can Save Time, Money and Lives
Imagine a pair of glasses for the visually impaired that combine video cameras in the lenses with a dedicated microprocessor and a speech-synthesis component that whispers in the user’s ears as he or she walks down the street: “Your destination is 75 feet ahead. Keep to the right to avoid sidewalk repairs. Your friend Chaz is approaching on the left.”
Welcome to the world of embedded analytics.
The business analytics market is booming, and an IBM ad even promises a day when the police arrive at an anticipated crime scene ahead of the prospective criminal. But analytics is increasingly taking place at the semiconductor level as well — within digital signal processor chips, or DSPs, to be precise — and that could fundamentally change the way we all interact with the world.
DSPs are specialized microprocessors optimized for speed. The first programmable DSPs from Texas Instruments appeared exactly 30 years ago and soon became central to the cellphone revolution. Because these chips can mediate our interactions with the world in very nearly real time, they make it possible for voices to be digitized, transmitted, received and then converted back into sound waves fast enough for natural conversation.
DSPs have also made digital imaging ubiquitous, enabling smartphones to shoot high-definition video and to stream everything from a blockbuster movie to yet another instance of a cat misbehaving.
The DSPs’ first two successes hinged on fast, efficient compression of data. But the latest high-performance, low-power DSPs are producing a wave of embedded analytics systems that boast an additional strength: smarts. And engineers are suggesting new potential applications for this intelligence almost daily.
Consider the enduring problem of drunk driving, for example.
An official with Mothers Against Drunk Driving recently asked me whether technology could help reduce the thousands of drunk-driving fatalities that still occur in the U.S. each year. I mentioned this to John H. L. Hansen, who’s head of electrical engineering at the University of Texas at Dallas and part of a study that uses specially equipped vehicles to monitor driver behavior. He told me that drunk drivers are easy to spot by subtle changes in the way they use the pedals and the steering wheel, a pattern of behavior that evinces guilt as surely as a failed Breathalyzer test.
Here’s where the analytics comes in. Embedded analytics involves gathering data from sensors, processing it in real time, using algorithms to make conclusions and then initiating action. In this case, the DSP would monitor usage of the pedals and the wheel, comparing it to the profile of a sober driver. When it determined a significant discrepancy, it would take action, perhaps easing the car off the road and shutting it down (or even calling 911), saving up to 10,000 lives a year in the United States and as many as 250,000 worldwide.
Let’s consider some of the other ways DSP-based analytics could improve our lives.
Facial recognition systems could both eliminate the need for passwords and thwart cybercrooks’ ability to breach your security. Simply look at a screen in your hotel room in Mumbai and you’ve got access to everything available to you through the cloud: your bank balance, a video link to your spouse in Barcelona, you name it.
Imagine communicating perfectly naturally with all the devices around you using touch, voice and even gesture. Analytics can make today’s remote controls as dated as rotary dial phones.
Imagine robots that can interact with us with near-human naturalness as DSP-based analytics enables them to perceive all our subtle cues of gesture, tone and facial expression. (Robotics is particularly interesting because this involves ultimately giving a robot the same ability we have to interact intelligently in real time with the world. That could be huge.)
On a grander scale, you can combine DSP analytics with radar, and suddenly you’ve automated the monitoring of all the flights going in and out of a major airport. Such a system could track the size, speed and location of each aircraft and make real-time traffic management recommendations.
With computational imaging you can integrate the perspectives of multiple cameras to produce 3-D results. Or to create higher-resolution images than possible with only one of those cameras. Computational imaging is particularly promising for security and for automated monitoring of industrial processes. It also enables you to combine the input from cameras on all four sides of a vehicle into one bird’s-eye perspective.
Your arms are full of groceries and it’s just starting to rain. But your front door recognizes you, and it graciously unlocks and opens as you approach.
Imagine a camera that can instantly analyze an image it has taken of a blood sample for signs of disease, a capability that’s equally useful in a busy first-world ER and a remote third-world village.
Your son loses his backpack, so you ask your phone, “Where’s Ethan’s backpack?” Locating it via a GPS tag in the label, your phone immediately replies, “In his locker at school.” Problem solved.
And then there’s the vast realm of health monitoring. If heart disease runs in your family, then DSP analytics and body sensors could continuously monitor you for the earliest symptoms. Your cardiologist then becomes your second opinion.
Which brings us to a fundamental question: Do humans really always need to be at the nexus of information flow?
The algorithms that drive DSP analytics can oversee utilities usage better than the most attentive homeowner, so why not take ourselves largely out of the loop and have the DSPs make seasonal temperature adjustments, start the dishwasher late at night when electricity costs are lowest and alert us about an unexpected spike in energy use? That helps free us up for other things — whether it’s spending more time with the family, finally reading “War and Peace” or just watching more cat videos.