CN115294762A - An adaptive traffic control method based on TinyML - Google Patents

Abstract

The invention provides an adaptive traffic control method based on TinyML, which belongs to the field of machine learning, TinyML, and edge computing. Through sensors and MCU templates, the speed and length of vehicles are statistically calculated, and the duration required to pass a green light is calculated. Change the duration of traffic lights to ease traffic stress.

Description

An Adaptive Traffic Control Method Based on TinyML

technical field

The present invention relates to the fields of machine learning, TinyML and edge computing, and in particular to a TinyML-based adaptive traffic control method.

Background technique

Transportation is an important foundation and support for economic and social development. With the development of cities, the number of traffic vehicles in each city has gradually increased. In order to solve traffic congestion, the traffic management department has issued various regulations, but the results are still unsatisfactory.

Contents of the invention

In order to solve the above problems, the present invention provides a TinyML-based adaptive traffic control method.

Technical scheme of the present invention is:

An adaptive traffic control method based on TinyML, through the sensor and MCU template, statistically calculate the speed and length of the vehicle, and calculate the duration required to pass the green light. Change the duration of traffic lights to ease traffic stress.

further,

Including (1) arranging piezoelectric sensor belts on the road, calculating the speed and length of the vehicle, and the duration of the vehicle leaving the signal light (green light). (2) Establish a random forest model, and use the model to train the data collected above. (3) Select the MCU template and deploy the model on the template. (4) Calculate the duration of the vehicle passing through the signal light (green light) by arranging piezoelectric sensor belts at the intersection, and adjust the duration of the green light reasonably according to this time. This invention provides an optimization method and device for adjusting the time of traffic lights, which breaks through the limitations of traditional traffic communication strategies, optimizes the traffic strategy at traffic intersections, relieves the traffic pressure at traffic intersections to a certain extent, and improves traffic efficiency. .

further,

The steps include: data collection, random forest model training, MCU selection and deployment, deployment of piezoelectric sensors and templates at intersections, calculation of the duration of green lights for vehicles passing in the north-south direction, and calculation of the duration of green lights for vehicles passing in the east-west direction.

in,

For data collection, 4 piezoelectric sensor strips are arranged on a road to collect vehicle information and the duration of the vehicle leaving the signal light (green light).

Build a random forest model and use the model to train the data collected above.

Through independent design or selection of existing equipment, the basic requirement is to deploy the model on the template based on the template of the open source RISC-V architecture.

Two piezoelectric sensor belts are arranged on the four sides of the realized intersection, and the information of the vehicles entering the intersection is calculated statistically, and the time to pass the green light is calculated through the matching template.

First count the vehicles accumulated during the red lights in the north and south, calculate two times, and take the maximum value of the two as the duration of the green lights in the north and south directions.

Then calculate the vehicles accumulated in the east-west direction during this period, calculate two times, take the maximum value of the two as the duration of the green light in the east-west direction, and repeat the above steps.

Description of drawings

Fig. 1 is a schematic diagram of the workflow of the present invention;

2 is a schematic diagram of the layout of the piezoelectric sensor;

Fig. 3 is a schematic diagram of the time stamp when the vehicle passes through the piezoelectric induction belt.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the protection of the present invention. scope.

The present invention proposes an adaptive traffic control method based on TinyML. Firstly, two piezoelectric sensors 1 and 2 with an interval of D meters must be installed in the middle of the road at a certain distance from traffic lights to detect the density of vehicles. When a vehicle passes the piezoelectric sensor, a small voltage pulse is generated. These voltage pulses are delivered to an AI-enabled microcontroller. By processing the collected data, the vehicle length (measured as the length between the two axles of the vehicle passing the same piezoelectric sensor) is calculated to identify the vehicle type and grouped according to the speed at which it passed. Then install two piezoelectric sensors 3 and 4 with an interval of D meters not far from the traffic signal light to detect the time when the vehicle leaves the signal light (green light) (Figure 2). The type and number of vehicles are used as the input of the pre-trained random forest regression model, and the time when the vehicle passes through the signal light (green light) is used as the output (label). Build a model and deploy it on an open source template.

The input part of the model is the vehicle type and quantity, where the vehicle type mainly records the vehicle speed and vehicle length. The time T ₁ is counted when the first axle of the vehicle passes the first strip, followed by the time stamp T ₂ when the first axle passes the second strip, and finally when the second axle of the vehicle passes the first strip time stamp T ₃ . Then the formula for calculating the speed of the vehicle is

The length of the vehicle is

s=v(T ₃ -T ₁ ) (2)

When the vehicle has a third axle, the calculation formula is similar to the above, just add the lengths of the two parts.

According to the speed of the above vehicles and the length of the vehicles, the types of vehicles can be roughly divided into 7 categories, namely bicycles, electric vehicles (motorcycles), cars, buses, buses (3 axles), trucks (3 axles), heavy trucks (4, 5 axles).

The specific steps of the algorithm are (using Python as the programming language)

1) Collect data sets, each piece of data is a timestamp list of 2 stripes.

2) Create 4 lists, namely 2 axles, 3 axles, and 4 and 5 axles.

3) Take out the time stamp list of the first band and the first item of the time stamp of the second band, and use formula 1 to calculate the speed v of the vehicle.

4) Take out the second item and the first item of the time stamp list of the first band, calculate the length s of the vehicle according to the formula (2), check whether there is a third item, if not, save the result as a tuple and add it to 2 axles list, skip to step 2).

5) If there is a third item, then according to the formula (2), calculate the length s1 of the vehicle again, add s and s1 as the new length s, check whether there is the fourth item, if not, save the result as Add the tuple to the list of 3 axles, skip to step 2).

6) If yes, repeat step 4), and then check whether the fifth item exists, if not, save the result as a tuple and add it to the list of axles 4 and 5, and skip to step 2). If step 4) is repeated, save the result as a tuple and add it to the list of axles 4 and 5, and skip to step 2).

7) The loop ends.

8) Add the category information of each vehicle according to the length and speed of the vehicle.

9) Calculate the time for the vehicle to pass through the intersection according to the above algorithm.

10) Enter the pre-training, use the random forest regressor (RFR) to take the number of vehicles in each category as input, and the green light duration required for the vehicle to pass through the intersection as the label. When predicting, enter the accumulated vehicles recorded during the red light period, and predict the green light duration required to pass the intersection.

11) Through independent design or model selection in existing equipment, the basic requirement is to deploy the model on the template based on the open source RISC-V architecture. The input of the board is the accumulated vehicles during the red light, and the output is passed to Predicted duration of traffic lights.

12) According to the four directions of the intersection, first predict the vehicles accumulated during the red lights in the north and south, calculate two times, take the maximum value of the two as the duration of the green lights in the north and south directions, and then calculate the accumulated vehicles in the east and west directions during this period For the vehicle, two times are calculated, and the maximum value of the two is taken as the duration of the green light in the east-west direction, and the above steps are repeated.

The above descriptions are only preferred embodiments of the present invention, and are only used to illustrate the technical solution of the present invention, and are not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims (8)

1. A self-adaptive traffic control method based on TinyML is characterized in that,

and counting and calculating the speed and the length of the vehicle through a sensor and an MCU template, calculating the duration time required by the green light, and changing the duration time of the traffic light.

2. The method of claim 1,

comprises (1) arranging piezoelectric sensor strips on a road, calculating the speed and length of a vehicle, and the duration of the vehicle leaving a signal light (green light); (2) Establishing a random forest model, and training the acquired data by using the model; (3) selecting an MCU template, and deploying the model on the template; (4) The duration of the vehicle passing through a signal lamp (green lamp) is calculated by arranging the piezoelectric sensor strip at the crossroad, and the duration of the green lamp is reasonably adjusted according to the duration.

3. The method of claim 2,

the method comprises the following steps:

1) Collecting data;

2) Training a random forest model;

3) MCU model selection and deployment;

4) Deploying a piezoelectric sensor and a template at the crossroad;

5) Calculating the duration time of the green light passed by the vehicle in the north-south direction;

6) The duration of the green light for the east-west vehicle to pass is calculated.

4. The method of claim 3,

data collection, placement of 4 strips of piezoelectric sensors on a road, collection of vehicle information, and duration of time the vehicle left the signal light (green light).

5. The method of claim 4,

and establishing a random forest model, and training the acquired data by using the model.

6. The method of claim 5,

and (4) selecting a template based on an open source RISC-V framework, and deploying the model on the template.

7. The method of claim 6,

two piezoelectric sensor belts are respectively arranged on four sides of the realized intersection, the information of vehicles entering the intersection is calculated through statistics, and the time of passing through a green light is calculated through a matched template.

8. The method of claim 7,

firstly, counting vehicles accumulated during red light periods of north and south, calculating two times, and taking the maximum value of the two times as the duration time of green light in the north-south direction;

and calculating the vehicles accumulated in the east-west direction in the period of time, calculating two times, taking the maximum value of the two times as the duration of the green light in the east-west direction, and repeating the steps.

Images