🕦 🍜 👛 Fighting traffic jams in a small town for a small budget: results of 6 months of the project 👛 🦀 🌜

The good old Soviet traffic light has two modes: it works and does not work. Its first optimization is to add night and day modes to the relay. The second is the same: add morning, evening and afternoon, differing in different delays of the lamp switching timer. And then everything. Next, we need sensors and external information flows, or even a connected network.

The simplest example of what can be done with a trivial induction loop on the road or an infrared sensor is not to switch the traffic light to the direction where there is no one right now. This is very convenient in the scheme "a large main road through the city and many secondary roads."

But we went a little further: in the city of Novomoskovsk (120 thousand inhabitants) we put cameras at traffic lights, changed all the controllers and connected it all into one network. The city has a small budget, so the rules are heuristic so far without any kind of space like data mining and machine learning, there are not so many traffic lights (because even putting 21 cameras is already expensive), but we were able to achieve very specific results.

The speed at the intersections with our “smart traffic lights” and at regular intersections nearby has increased. We learned how to prioritize the flow of cars in the morning to a large plant, count and process transit wagons, and even waved at the GLONASS ambulance sensors to remove possible traffic jams in front of them.

Motion Control Platform Interface.

How do you deal with traffic jams?

There are three basic approaches to what can be done with transport collapses:

Expanding roads and building new infrastructure is the most expensive way.
: – , — 25–35 % .
: ( — ).

I want to immediately draw attention to the fact that “smart traffic lights” connected to a network of coordinated traffic solve not only the tasks “so that there are fewer traffic jams”, but also more specific ones. For example, in our example of Novomoskovsk, one of the important tasks is for everyone to reach the plant in the morning on time. We can create conditions when optimization leads to the passage of this particular stream. And similarly, in experiments with prioritization of the ambulance, it is obvious that the average flow rate will decrease, but a car with an urgent patient (and everything next to it, since we operate with clusters about the size of a quarter) will be eight to nine minutes faster.

What problems can be in a small city?

Transport collapses are peculiar not only to megacities, but also to small cities. Many of our cities were rebuilt and redesigned according to the principles of the 60s. The road infrastructure was simply not ready for the flows of the 2010s. Of course, the smaller the city, the less usually you are stuck in traffic, but if you can win 10-15% of the travel time (this is a practical example) - then why not? This will allow us not to expand the road where it is still physically impossible to expand, and this will allow us to get a lot of any good.

In our case, the city is between M4 and M5, and collapses occur both inside it and on the road. Departure has sufficient bandwidth, but accidents occur there. Inside the city, traffic and rush hours create problems.

Of course, you can put more guards (who actually just make the very decisions that the “smart traffic light” could make), but this is not the trend of the ministry and it is simply not economically feasible. And here we find ourselves in an amazing situation when a small city with a small budget can solve part of the problem. In a metropolis, ASUDD needs an extensive network of managed objects, and the problems there are more likely from total overload than from non-optimal flow control. But in cities with up to 200-300 thousand inhabitants, the main roads are very easily automated. And very cheap. What played a decisive role, of course. Separately, I note that at first the administration was very worried that a new specialist would be needed to manage and maintain the ASUDD: this is an unplanned "extra" employee. However,automating decision support to the maximum by integrating into the city’s intellectual platform, we just had to train an existing employee.

. , , ?

The most important thing is that each traffic light object has a controller and may have some kind of sensors, most often in our case a camera. Thanks to specialized software, camcorders recognize vehicle registration plates and determine the density and average flow rate. The collected data is transmitted to the controller, which makes decisions to reduce or extend the duration of the green traffic light. Further statistics accumulated over a five-minute interval are transmitted to a central point for processing. After analyzing the statistical data, the system automatically makes a decision on the inclusion of the necessary coordination programs at traffic lights. So the automatic mode allows you to effectively control traffic flows, improving the quality of the "green waves".At the same time, it remains possible to manually control traffic lights for the priority passage of emergency vehicles and special services ("green street").

You can use different types of sensors: an optical cable in the roadway, induction loops, laser sensors for the presence of vehicles (similar to speed measuring instruments), different flow sensors (similar to very simplified semi-blind cameras with cut-off interfaces) and so on. But we used cameras of sufficient resolution to automatically recognize license plates, and we plan to determine the type of vehicle (passenger, bus, freight, incomprehensible).

Due to the limited budget, we have identified those intersections (the regulation cluster at the intersection is called the “traffic light object”) that have the greatest impact on traffic. They turned out to be about 30%, mainly just on the road through the city, where transit traffic passes. Then we arranged our devices in such a way as to provide the greatest influence on the flow, because cameras can not be stuck everywhere, and according to two intersections, restore the state of the “blind” intersection between them.

As a result, the minimum set of equipment for the first part of the project was determined. Here's what they used, for example:

Crossing Str. Labor reserves - st. Kuibyshev (address: 15a, Kuibyshev st.)

No. p / p	Name of equipment	Quantity
1	Camcorder (Type 1)	3
2	Camcorder (Type 2)	1
3	Lens	1
4	Control cabinet (type 2)	1
5	Switch (Type 1)	1
6	PoE Extender	1
7	Lightning protection	8
8	Floodlight	1
9	Power Supply (Type 1)	1
10	Power Supply (Type 2)	1
eleven	Optical Transceiver (Type 1)	1
12	Optical Transceiver (Type 2)	1
thirteen	Installation kit	1

Examples of specific models: controller , video camera , zoom telephoto lens , another camera (type 2).

What does software ASUDD do?

Through the cameras we see the presence of transport, measure the flow rate, determine the congestion of each strip. Based on this, you can make decisions on how to switch traffic lights.

Each traffic light has a failsafe program on the controller for autonomous operation, but in our implementation, the main control goes through the central node - the traffic coordination server. That is, all data flows into one model, and then on the basis of them a decision is made how and what to regulate.

Somewhere here I could say the words about the big data being collected, about the neural networks and other things, but, I recall, the budget was strictly limited. Therefore, we used the built-in neural networks of road service employees. They sat with statistics, carefully tried different regulatory regimes on paper, and then - in practice, and came to certain conclusions about what and how to do. The result was a set of 24 heuristics for different situations.

In fact, the first two months were calculated (or intuitively selected) and confirmed the optimal duration of the phases of traffic lights in each situation: in the fog in the morning, in the rain on the weekend, in good weather on Monday evening and so on. We found the main phases for traffic lights without detectors, which change four times a day.

You can see the statistics.

What other benefits does such a system give?

. ( ) « () ». : , . , , , , , . , , , , — ( , ).
. , , . , - . , 35 , 42 . . , .
«» . , , - , , - ( ), . , - . , , . , , . .
. , , « » . , , , . — ( ), .
. , . , , . , , , . « », , .
. , « - , 20 ». , , -, , . -, , 60 % .
Counted the workers of the vehicle. For statistics on the organization of public transport (such as bus routes or railway stations), it is very important to know how many people in the city go to work outside its borders. The same number recognition along with the rule “leaves four to five times a week and returns after 8-14 hours” made it possible to understand what was in the details and with the directions.

All the system data is added up, and someday it will be possible to process it with more complex methods. Now, for example, the work shift traffic within the city was estimated by people both for optimization and for estimating the required number and frequency of buses, but it will still be possible to do this constantly automatically. Plus, the data is useful for planning the expansion of roads in the city and in the region, when needed.

Experiments

, — « ». , : , , «» — . , , «» : , , , . 15 — , . , , . : , «» , .
, .
. , . , () , , , - - ( - 800 ). . «», , , . «, !» . : , , , , — .

12 , . 30 % , - , (, ).
4 /. .
, .
.
. — , . .
.

Here's the press release . If you want to tell one of the traffic experts in your city about this, then please just drop the link to Habr and immediately - to my mail VBabiy@technoserv.com, you can send questions about the system, about budgets and deadlines.