Applying artificial intelligence to solve the problem: Detecting and monitoring traffic violations

Artificial Intelligence (AI) for short can be considered a great achievement of modern science, when it brings great and meaningful benefits to all areas of life.

Currently, there are many solution providers for traffic monitoring using image processing techniques with effective data processing capabilities, in which AI Camera is a trend that has been receiving attention by many people. compact and integrated processing directly on the camera and send data to the center. Another approach to the problem of detecting and monitoring violations on the road is the solution of using Edge devices located at the camera point to directly process data at that camera and then send it to the center.

These methods have the advantage that the system is simpler because the data from the camera is processed directly and then sent to the center, where the center only plays the role of management and storage. However, besides the benefits brought, these solutions do not meet the requirements of processing features with speed, accuracy and quality of input video, scalable flexibility,…

Traffic violation monitoring system of HTSC is built based on a centralized processing structure of data transmitted from the camera system through the VMS system at the server. This system offers the following advantages:

• At intersections, it is only necessary to use a maximum of 2 cameras for 3 lanes to handle all violations and save vehicle information (while the above solutions usually have to install one camera in each lane, and each camera in the traffic zone). usually only handle one violation…).
• The system allows for a wide range of cameras to be used (up to 4K video streams are supported) and can be added easily without much impact on the system.
• Another advantage is that centralized processing at the server allows zoning, troubleshooting and centralized maintenance in the data center.

The highlight of the system compared to other solutions is that artificial intelligence (AI – Aritifical Intelligence) techniques are incorporated into the video processing application. First, an automatic vehicle detection model will recognize common types of vehicles to be monitored such as cars, buses, trucks, motorbikes, etc. Then, detected objects are introduced. into object-specific recognition models and object-tracking models, from which common vehicle characteristics such as vehicle make, color, and license plate can be recorded. These features support the process of finding vehicle types according to future tracking requirements as well as penalizing violating vehicles.

In addition, vehicle tracking combined with traffic rules will help detect common vehicle violations such as crossing red lights, stopping/parking in the wrong lane, crossing the median line, going in reverse. dimension, speed violation, etc. Centralized processing at the server allows the system to simultaneously process large number of objects with high performance. In addition, the system has the ability to automatically update data from the videos collected on the server and retrain the model with new data, thereby automatically upgrading the model with high accuracy.

With the above architecture, the system allows flexible management of vehicle characteristics, can add or remove features to be managed, especially the system integrates a module to clarify license plates based on technical skills. Deep learning algorithm learns variable representations of characters that can automatically convert and recognize characters in number plates that are distorted due to the video input quality. The system also allows the flexibility to choose to monitor vehicles and types of violations at each time and traffic condition to suit the working practice of the functional forces.

In addition, the database of the stored vehicles effectively supports the vehicle tracing activities at the request of the investigating forces when there is a request to search and trace the vehicles with the Common features such as license plate number, color, car brand, travel time, etc.

With these outstanding advantages, the traffic violation monitoring system of HTI Investment and Technology Joint Stock Company has excellently won the Sao Khue Prize 2021 in the group of “New technology solutions” – this is the award. The oldest and most prestigious award of Vietnam’s IT industry.

Keeping up with the Government’s policy and trend of digital transformation of data, centralized database management systems like traffic violation monitoring systems when put into synchronous deployment promise to create Data interoperability effectively supports management, public administrative services, etc. in a transparent and fair manner, towards building e-government.