Synopsys EV6x vision processors portend the demise of traffic lights
As you doubtless recall, a couple of weeks ago I had my mind boggled at the Embedded Vision Summit (see Day 1 and Day 2). Since that time, it seems that every day I'm presented with a new mixture of embedded vision-related devices, IP cores, technologies, and tools. Also, I find myself spending a lot of time thinking about the implications of embedded vision and how it's going to affect our lives.
In the case of embedded vision IP cores, for example, the guys and gals at Synopsys have just announced their next-generation DesignWare EV6x family. In addition to an optimized convolutional neural networks (CNN) engine, the EV61, EV62, and EV64 vision processors contain one, two, or four cores, respectively, where each core comprises a high-performance 32-bit scalar processor and an accompanying 512-bit vector DSP.
The EV61, EV62 and EV64 processors are designed for use in applications this require HD resolutions, such as advanced driver assist systems (ADAS), video surveillance, drones, augmented reality, virtual reality, and simultaneous localization and mapping (SLAM).
Actually, it's the whole embedded vision meets automotive space that I've been particularly pondering over the past few days. Just this morning, in fact, my chum Rick Curl emailed me to say:
Hi Max, I was sitting in my car this morning at a seemingly interminable red traffic signal with no other vehicles in sight when I began to wonder... how different would traffic signals be if the technology being developed for self-driving cars was applied to them?
We’ve apparently already got sensors that can detect the presence of cars, the speed and trajectory of the cars, and the presence of pedestrians. They should also be smart enough to realize that, given a vehicle's current rate of deceleration, it will likely run a red traffic signal so the traffic from other directions could be alerted.
As far as I know, the present "state of the art" traffic control systems can only tell if there are cars waiting at an intersection and cannot detect approaching traffic. I presume these new sensors could also detect the presence of emergency vehicles and give them priority. Is anyone developing something like this?
When you come to think about it, this opens up a Pandora's Box of possibilities. Most of the people I've talked to about this space recently have focused on embedded vision in the cars (and trucks) themselves. It won’t be long before vehicles are equipped with at least 10 cameras. In addition to the outward-pointing devices used to detect impending obstacles and things in the driver's blind spots, there will also be inward-pointing cameras mounted in the rearview mirror assembly observing the driver to see if he or she is tired or distracted, either of which situations may prompt the vehicle to modify its alerts and/or actions.
As Rick alludes to, however, it could be really advantageous to have traffic signals with embedded vision capabilities. On a personal note, there's a set of traffic lights on my way home that are driving (no pun intended) me insane. They force me sit there on red for an unconscionable amount of time when no one is coming from the other direction. It would be a really great first step if these lights could see how many cars (if any) are approaching from the various directions and then change appropriately to achieve the highest possible traffic flow and the lowest possible driver frustration level.
Following on from Rick's query, since drivers sometimes leave their braking until they get close to a junction, I'm not sure if the signals would have the time to respond in a meaningful way to someone running a red light, but I do think it would be possible for them to detect the flashing lights and other characteristics of emergency vehicles and then set their signals appropriately so as to clear any vehicles from the junction, prevent additional vehicles from entering the junction, and give the emergency vehicles a big green "thumbs up" as they sail through.
If you really take all this to its logical conclusion, however, the future may be bleak for traffic lights, which may well be destined to go the way of the dodo, but first...
The world's first traffic signal of note was manually-operated, gas-lit, and short-lived. Installed in London, England, in December 1868, this Heath-Robinson design combined three semaphore arms with red and green gas lamps for night-time use. Sad to relate, it exploded less than a month later, severely injuring its policeman operator. It wasn't until 44 years later that the first electric traffic light was developed by another policeman, Lester Wire, in Salt Lake City, Utah, who also used red and green lights to indicate "Stop" and "Go," respectively.
Since that time, the little rascals have bred like rabbits and sprung up everywhere like mushrooms (I never metaphor I didn’t like). I daren't even hazard a guess as to how many traffic signals there are in the USA or around the world (I do know that there's one too many on my drive home).
But we digress... There's a lot of talk about autonomous cars these days. Many industry experts believe that the mass production of self-driving cars will be in full swing circa the early 2020s, and driverless cars will be deployed around the world by the 2030s. These little scamps (the cars, not the industry experts) will employ a variety of technologies to perform their magic, including embedded vision.
The point is that, if you have fully autonomous vehicles, all of which can see the road ahead and communicate with each other, then this may well sound the death knell for traffic lights. When you think about it, there would be no reason for cars to waste time slowing down, stopping, and queuing up at junctions; instead, employing only relatively slight variations in speed, they could crisscross past each other in a manner similar to this video of a synchronized Japanese walking team (watch especially the 45 and 55 second marks).
I don’t know about you, but the thought of all this gives me the willies (I'm referring to self-driving cars going through traffic light-less junctions, not synchronized Japanese walking teams). Being driven to work could end up being like a prolonged game of chicken. Fortunately, I have the answer -- pass me my Virtual Reality Headset.
What say you? Are you looking forward to an autonomous automotive future with dread or delight? Will you crouch hunched in the car screaming inside your head (much like myself when my wife is driving); will you relax comfortably in your chair waving merrily at other passengers as you pass them by; or will you join me in an alternative reality where we try to distract ourselves from our impending doom?