Please enable JavaScript in your browser.

fltech - 富士通研究所の技術ブログ


Introducing Traffic Image Surveillance

Hello, I'm Zhang Nan, from Fujitsu Research & Development Center Co, Ltd. Today, I would like to introduce an exciting new AI core engine –Traffic Image Surveillance –that is now available from Fujitsu Kozuchi.

This blog is part of a series introducing the AI core engines of Fujitsu Kozuchi. At the end of the article, you can find a list of all the previous articles!

Traffic Image Surveillance is a technology that uses road footage to monitor traffic volumes and detect traffic events such as reversing and disasters in real-time. This technology significantly enhances the efficiency of traffic control and management. It is one of Fujitsu Kozuchi’s AI core engines, which enables the fast testing of cutting-edge AI technologies developed by Fujitsu.

On the road, various entities such as cars, motorcycles, bicycles, and people are on the move. Furthermore, the road environment is constantly changing due to factors like weather changes including rain and snow, snow accumulation, falling objects, road damage, and landslides. For safe traffic control and management, it is necessary accurately to detect vehicles, motorcycles, people, and other elements 24 hours a day, in the continually changing road environment where various objects and people are constantly passing by. In addition, it's not only necessary to detect objects and people for monitoring the real-time traffic situation, but also to recognize scenes such as weather conditions. To overcome these challenges, we have developed Traffic Image Surveillance.

Benefits of Traffic Image Surveillance and how to use it

The principal value of Traffic Image Surveillance is to use road footage to monitor traffic volumes and detect traffic events such as reversing and disasters in real-time. This technology significantly enhances the efficiency of traffic control and management.

Currently, Traffic Image Surveillance has 27 functions. I will introduce the various elements that Traffic Image Surveillance can detect in five different categories.

① Traffic counting
Traffic Image Surveillance can count the number of vehicles, people, and bicycles. It can also display their movement paths on the screen. Vehicles, people, and bicycles are indicated by color-coded squares, and their paths are shown with yellow lines. Additionally, by setting multiple detection areas within the footage, it is possible to count the traffic separately for both the uphill and downhill lanes.

② Traffic event detection
It detects various events on the road. Specifically, it includes falling object detection, traffic jam detection, lane-change detection, low-speed detection, high-speed detection, short-distance detection, reversing detection, and parking detection.

③ Road safety (Detection of abnormalities on the road)
It is also capable of road flood detection, road snow detection, landslide detection, and road damage detection. For example, with road snow detection, it can color-code and display snow-covered areas and show the percentage of the area covered by snow.

④ Environment sensing
It senses road conditions such as weather and visibility (the distance at which objects can be clearly seen with the naked eye). In weather classification, the weather state such as "rainy" is displayed on the screen. In visibility measurement, an index called the VI value that evaluates visibility is displayed. Additionally smoke detection, image defogging, and low-light enhancement are also possible.

⑤ Abnormal camera detection
The final category is abnormal camera detection. If there is an issue with the fixed camera, it can detect any abnormality and correct it. For example, camera angle change detection can immediately detect slight changes in the camera's angle of view. If the direction of the fixed cameras is shifted due to a shock or unexpected movement, or if there is even a slight change in the shooting range, the accuracy of the image recognition AI decreases. With this function, it is possible to prevent a decrease in accuracy by immediately detecting changes in the camera's angle of view. In addition to camera angle change detection, it offers functions such as scene-change detection, video anti-jitter, shift correction, and abnormal camera detection support stable operation.

Features of Traffic Image Surveillance technology

Traffic Image Surveillance engine has base models for segmentation, detection, and classification that are specialized for traffic scenes and generated from actual image data. Furthermore, with high-speed models and algorithms, it achieves real-time processing. This engine's technology has the following features.

  • High-precision (average >95%) and high-efficiency. AI models based on the know-how accumulated in the development of traffic surveillance services and technologies such as traffic counting and event recognition, etc. For example, we have designed novel algorithms for nighttime vehicle detection and tracking, achieving traffic counting accuracy of over 90% for such nighttime hard case.
  • Respond to various environmental changes. We design functions of environment sensing (visibility measurement, low-light enhancement, image defogging, etc.), enabling the system automatically to perceive environment changes and improve the accuracy of other traffic detection functions through image enhancement.
  • Uses large-scale models, for example SAM, for automatic data generation. To address the issue of insufficient data for disaster detection, we utilized large models to extract landslide and road damage samples accurately, automatically generate training data for various scenarios, which improves the robustness of the function.

Looking ahead, we will continue exploring how to utilize large models, including multimodal large models and large vision models, to optimize our existing system for continuous accuracy improvement.

Interested in testing the Fujitsu Kozuchi?

Our technology offers a wide range of features to explore. With just videos or images of traffic monitoring scenarios, you can experiment with any feature that takes your interest. Additionally, we provide a wealth of examples to help you understand each feature through hands-on experience. We will also develop more features to enhance road safety further with our system in the future. For a demonstration or to test our Traffic Image Surveillance, please contact us here:

In addition to Traffic Image Surveillance, we also introduce other AI core engines of Fujitsu Kozuchi on our TECH BLOG.