Smart Traffic Signal Optimization Using Computer Vision and Edge AI Reduces Congestion in Megacities by 35%

Context and Bottlenecks of Urban Transport Infrastructure: Traffic congestion is a serious disease of every developing megacity, causing billions of dollars in damage each year due to fuel waste, reduced labor productivity and huge carbon pollution. This case study delves into solving the traffic problem at an arterial avenue with 15 consecutive intersections (intersections) connecting from the satellite urban area to the city center.

Before the technology project was implemented, the traffic control system here operated entirely based on the old-style Fixed-time Control mechanism. Engineers configure green/red light times based on manual statistical surveys of the past few months. The green light cycle of the main direction is always set to 60 seconds by default, regardless of the actual situation. As a result, at noon or late at night when the traffic on the main axis is deserted, hundreds of vehicles on the secondary axis still have to wait unreasonably at the red light. On the contrary, on rainy afternoons when flooding occurs, the traffic flow on the main road increases 5 times, the 60-second green light cycle is too short, causing thousands of vehicles to congest into a giant parking lot stretching for kilometers. The rigidity and "blindness" of the old system requires a radical surgery: We must give the signal light system "Eyes" to see the real situation and "Brain" to make decisions on its own.

Technical Solution 1: "Eyes" Computer Vision and Edge AI: Instead of tearing up the road surface to install expensive and easy-to-use Inductive loop detectors failure, the project decided to deploy a high-resolution Smart Camera system combined with Artificial Intelligence. However, if all thousands of live 4K video streams (Livestream) are transmitted from intersections to the cloud data center for analysis, the fiber optic network bandwidth will be overloaded, and transmission latency (Latency) will cause data to be returned slowly, distorting the decisions of traffic lights.

To solve the latency problem, the architect team has applied Edge Computing / Edge AI technology. At each intersection control cabinet, a specialized embedded computer for graphics processing (such as NVIDIA Jetson Orin Nano hardware) is installed directly. The cameras will push a short video stream of just 1 meter into this edge computer. Here, the Computer Vision algorithm – specifically the YOLOv8 (You Only Look Once) deep learning model that has been extremely optimized through the TensorRT compiler – will operate at 60 frames per second (FPS). It continuously draws bounding boxes to identify, classify (Cars, Motorcycles, Trucks, Buses, Pedestrians) and accurately count the number of vehicles stopping or moving in each lane. Not a single frame of video was uploaded to the network; The Bien device extracts super light data strings (Data JSON): "North-South direction has 50 cars, average speed 10km/h" and sends it to the command center in less than 10 milliseconds. This Edge AI mechanism not only solves the problem of latency but also absolutely protects the image privacy (Privacy) of pedestrians.

Technical Solution 2: "Brain" Adaptive Signal Control (ATCS) using Reinforcement Learning: At the central control room, data from 15 intersections converge to feed into the Adaptive Traffic Control System (ATCS - Adaptive Traffic Control System). Data scientists do not program IF-THEN style rules for traffic lights, because the traffic environment has millions of random variables. Instead, they deploy Reinforcement Learning - multi-agent Q-Learning algorithm agents.

In the digital space simulation environment (Digital Twin) of the road, AI Agent acts as the light controller. If the AI ​​decides to prolong the green light to reduce the total waiting time of the convoy, it receives a "Reward"; If a wrong decision causes congestion at the next intersection, it gets a "Penalty point". After millions of times of self-learning through simulation, the AI ​​model is put into real combat. By calculating the friction matrix of 15 consecutive intersections, AI predicts: A cluster of 300 vehicles that have just exited intersection No. 1 will need exactly 45 seconds to move to intersection No. 2. The system will secretly count down and automatically change the light of intersection No. 2 to Green just in time for the convoy to arrive. This phenomenon creates a spectacular "Green Wave", allowing vehicles traveling on the main axis to pass through dozens of intersections in a row without having to brake to stop once. At the same time, if it detects that an Ambulance (equipped with a V2I - Vehicle-to-Infrastructure transmitter) is approaching, the system will automatically force all other directions to turn to the red light, clearing an emergency priority flow (Green Corridor) to save people.

Practical Results and Socio-Economic Efficiency: After 6 months of experimental runs and self-tweaking of the algorithm, the Intelligent Traffic system has brought shocking results. dynamic. At the arterial boulevard, the average time spent waiting for a red light during rush hour has decreased by up to 35%, equivalent to saving hundreds of thousands of labor hours per month for people. Average traffic speed increased from 18 km/h to 28 km/h. Reducing the number of times the brakes are applied and restarted by vehicles (especially heavy trucks) has helped directly cut 15% of harmful CO2 emissions released into the residential environment. This case study is the most eloquent proof that: Instead of spending billions of dollars and decades to clear land, expand roads or build overpasses, applying AI core algorithm technology and Edge Computer Vision can revive a dying physical infrastructure system, turning a congested urban area into a true Smart City.

Need a reliable technology partner for your business? See DUDI's dedicated tech team & white label.