Team-BHP (https://www.team-bhp.com/forum/)

-   Street Experiences (https://www.team-bhp.com/forum/street-experiences/)

-   -   Bangalore gets AI-powered adaptive traffic control system to ease congestion (https://www.team-bhp.com/forum/street-experiences/286284-bangalore-gets-ai-powered-adaptive-traffic-control-system-ease-congestion.html)

The Bengaluru Traffic Police has implemented a new AI-powered adaptive traffic control system in the city. The new Bengaluru Adaptive Traffic Control System (BATCS) is said to revolutionise traffic management and ease congestion.



As part of the BATCS project, the city is upgrading 136 junctions and installing 29 new ones, covering a total of 165 traffic signals. The system is based on an indigenously developed CoSiCoSt ATCS application developed by the Centre for Development of Advanced Computing (C-DAC) and is being implemented with the Arcadis IBI Group.

According to M N Anucheth, Joint Commissioner of Police (Traffic), the old traffic management system lacked real-time adaptability and central monitoring. Thanks to AI technology, BATCS dynamically adjusts signal timings based on real-time traffic densities using inputs from camera sensors.

Source: HT

Link to Team-BHP News

After initial hiccups, I sincerely wish for a great successful outcome for this model that can be implemented across the rest of the nation to curb road rage, over speeding, protect the innocent & improvise road safety. Afterall the borrowed $250M from world bank has to be used effectively.

It would be great if this system can detect presence of ambulance and open signals accordingly.

The timer needs to be enabled instread of displaying VAC/ML.

The pedestrian crossing window is 10s which is too low for wider junctions.

One gripe I have is that some of new signals are extremely bright! It would help to tone it down by one or two steps.

Being a resident of Bangalore, can you comment more on the effectiveness of this system? I am all the more curious about this considering the royal failure of ADAS systems on Indian roads.

To be honest, I have not witnessed much improvement - but I do see potential if calibrated well.

I pass through four of these on my way back from work, and have passed some of these on non-work trips.

At two signals, the traffic seems to flow better. At the other two it is pretty much same as before.

Though the infra has been updated at the junctions, I am not sure if the system has been fully calibrated yet.

These systems work best when a junction has a high traffic road and a low traffic road. When the traffic density is distributed unevenly, then the systems can alter the timings effectively.

If the traffic is distributed evenly, each road will get similar amounts of time.

This will definitely help drivers at night. No more unnecessary waiting for red lights at empty junctions.

Would love to see long term impact and feedback.

All over the world, these sort of systems are operational, without AI.

I would like to understand what AI makes these systems do differently. Because real time analysis of traffic flow is not that difficult at all. It’s been done for several decades. So what does the AI do, that a current version can’t do?

My guess, nothing really, this is just PR, throwing the term AI into the public arena so it appears to be high tech.

Jeroen

Agreed. For example, such kind of systems were operational in the US back in late 1990s too. They had sensors embedded in the road that detect presence of vehicles at traffic lights. Induction loop type sensors are common, but there are many different kind of sensors too.

What makes this different is that it requires no sensors in the roads, and purely based on the images of the road, it detects the number of vehicles waiting at the traffic light. In India, we do not have those sensors embedded in the road, which the other countries had for decades. So in India, we are skipping that traditional method and straight jumping to these camera based systems which detect number of vehicles purely from the images! The adaptive signal timing algorithm would remain the same as used elsewhere, but the method of detecting the number of vehicles waiting is an AI based calculation here from the images, instead of the road-embedded sensors used elsewhere.

In this case, it is not a PR. Calculating number of vehicles waiting at the signal from the images of the road is indeed a genuine AI problem ("Object Detection" from images, to be precise). Although this is now a trivial (or very easily solved) AI problem, given the state of current AI, but still, this is a genuine AI algorithm and not just a PR.

Yes, there are many other so called AI applications where the term AI is thrown into it just for PR or to make it sound high tech as you say, but this one is not that case.

Have a look at the report in The Hindu below. The system is designed to accommodate emergency vehicles too. I agree with your comment about showing the timer, that's a good driver's aid, even if the value is not deterministic in this setup.

Completely agree. Whether the cat is old or young, black or white, it is nice as long as it catches mice. Thus spake a wise man :).

For a country that specializes in digging up roads with sometimes hilarious results, sensors in the roads is clearly beyond our capabilities to manage. Cameras seems to be the right sensor for India, and this system certainly seems more amenable to problem solving than those sensors in the road. From regulating traffic to blocking traffic to a submerged road to 'discouraging' wrong side driving, the applications can be immense.

However, we have to see how the system adapts when some enterprising utility cuts off the power to the cameras!

I am simply thankful that in a city which has given up on urban planning, which has no public master plan, there are still people who have not given up. Who are thinking of prioritizing emergency vehicles, pedestrians, public transport, cyclists (hopefully). Hats off.

Whatever technology is implemented, it becomes useful only when people follow the traffic rules; otherwise, it’s of no use.

Absolutely! A good percentage of the traffic chaos in Namma Bengaluru is caused by our Natural Stupidity (we don’t follow traffic rules) and the authorities now think this can be solved by Artificial Intelligence :D

Wondering why these new traffic lights display “VAC” and what does it stand for or mean?

VAC stands for Vehicle Actuated Control, a system that uses cameras and sensors to adjust traffic signals in real-time based on the number of vehicles.

The signal can be manually controlled (not VAC) and then it shows as MNL.

VAC can lead to scenarios where one road will have green light for extended periods due to that road having more traffic. How smart the AI behind VAC is, only time will tell.

Whatever technology Bangalore uses, Bangalore should have AI dedicated to detect potholes to manholes, suspension killers to craters so that everyone knows when to slow and where to slow and also have automated challan system for illegal parking so that parking woes are out of the question.

If one thing Bangalore needs more than anything is proper signaged roundabouts and exits which would solve half of its woes!

Cheers!
VJ

I strongly suspect AI being used is a hype, at least as of now. Notice that in every signal, there is a timer for pedestrians well synchronised with the vehicle signal. They can, for sure, show the timer for vehicles. But that means they can't project it as an AI system and get their annual bonus :).

Thanks! Let me see if the pedestrian timer shows up if the signal is in VAC mode.

The timer can not be shown in these systems exactly because there is no fixed time when the signal changes.

Basically there are two independent algorithms working together here:

Algorithm 1: This is a method that senses or measures the number of vehicles waiting at a traffic light when it is red, and number of vehicles flowing through it and incoming towards it when it is green. This algorithms is purely an AI algorithm in these camera based system (and once again, this is a real AI and no PR or hype; calculating number of vehicles in real-time, from a video/image is indeed a genuine object detection AI problem and not a hype!). In traditional systems in other countries, this was previously done using sensors embedded in the road, and even those countries are now moving to camera based AI systems.

Algorithm 2: Taking the input from Algorithm 1 mentioned above, it determines when to change the traffic lights from red to green and vice versa. This can be either a complex optimization problem or a simple rule-based model, or a combination of the two. No, this is NOT an AI and nobody claims this to be an AI either. However, although not AI, this is still a dynamic algorithm that does not decide something like "turn this light green in 30 seconds". Instead, it decides "turn this light green NOW", when it senses a certain amount of pile-up at the red light (or in other case, it senses enough vehicles have crossed the green light and says turn the light red NOW).

Since this algorithm 2 takes signal change decisions at any moment, based on real-time conditions at that moment, there is no pre-decided time, nor any advanced notice of a signal change time. For this reason, there can not be the timer shown!

The actual implementations can vary and there is enough room to use different flavors of these implementations (for example, instead of deciding to change the signal NOW, it can add a small delay of a few seconds as a buffer when that decision is made, but that is still not enough to act as a timer), but what I described above is the most generic way these systems work, and why they do not show the signal change timer.