1 of 7

AATLaSAuto-Adaptive Traffic Light System

Team Valley:

Do Duc Anh

Tang Minh Anh

Nguyen Minh Duc

Phan Nhat Hoang

Pham Nguyen Hung

Nguyen Ngoc Minh Nghia

Team Valley

2022

www.9slide.vn

2 of 7

Problem Statement

Consider the problem: We have a busy crossroad with 4 traffic light clusters. We have traffic camera(s) that can give us information about the directions each car is travelling.

What is the optimal green-light time for each travelling direction at the crossroad?

Team Valley

2022

www.9slide.vn

3 of 7

Our Approach

1. Infer the traffic flow for each of the 12 possible travelling directions at the crossroad

We utilizes the data from the cameras to perform

2. From the traffic flow, calculate the optimal green-light time for one of the two sides of the crossroad.

www.9slide.vn

4 of 7

Our Approach

First step

Second step

Performed with a deep neural network model for object detection and tracking

Performed with a deep reinforcement learning model to calculate the green-light time

We refered to the paper "Zero-VIRUS: Zero-shot VehIcle Route Understanding System for Intelligent Transportation"[1], and used transfer learning to re-trained the model with our custom dataset.

We first created SUMO simulation based on real-life traffic data. For the model, we built a Deep Q-Network. We then train the model in that simulated environment. The model was trained in 10 epochs with 200 steps.

www.9slide.vn

5 of 7

Our results

Lost function of the controlling model (total waiting time at crossroad)

Team Valley

2022

www.9slide.vn

6 of 7

Outstanding features

Consider all movement flows

Utilise ubiquitous data

Real-Time model potential

01

03

02

Represent reality more accurately

Easily accessible data from traffic cameras

Flexible with ongoing situations

Team Valley

2022

www.9slide.vn

7 of 7

Advantages

Reduce costs

Comparing to the conventional methods, this system reduces the cost of building infrastructure, implementing 5G connection, and reduce the time of transmitting data.

Extendable

Since AATLAS can receive as many junctions as possible, it can be well implemented at national level.

Asynchornized system

We can create a network of AATLaS working simultaneously

www.9slide.vn