Hello, I'm Ito from our Data & Security Research Laboratory. Today, I would like to introduce an exciting new AI core engine - Multi-Camera Tracking – which is now available from Fujitsu Kozuchi (code name) - Fujitsu AI Platform.
Multi-Camera Tracking is a technology that can track an individual in real-time using several cameras. It is one of Fujitsu Kozuchi’s AI core engines, which enables the fast testing of cutting-edge AI technologies developed by Fujitsu.
With the introduction of Multi-Camera Tracking, we have been able to overcome some major challenges. It was difficult for previous systems to track the same person consistently from camera footage, due to interruptions when tracking them using multiple cameras with different angles. Also, tracking could be interrupted whenever the person temporarily moved behind objects or was overlapped with other people in crowded places.
Benefits of Multi-Camera Tracking and how to use it
The principal value of Multi-Camera Tracking is that it allows us to track the same person, even in crowded and large areas. I will explain two main features.
The first feature is the ability consistently to identify the same person from multiple camera footage. When monitoring large or complex locations, it's necessary to use multiple cameras. However, as I explained earlier, it was difficult to track the same person with multiple cameras before, as each camera’s angle makes the target person look different. Multi-camera tracking identifies the same person based on their appearance, even using different cameras with different angles. This allows us to track the same person using multiple cameras, even in large places like stations or airports, or in complex locations such as building interiors.
The second feature is that tracking continues wherever the target person is located, including when they are temporarily hidden behind objects or people. Previous technology uses a method that estimates the same person based on their direction of movement. With this method, if the person being tracked changes direction when hidden by objects or people, the tracking is interrupted. Multi-Camera Tracking can identify and track the same person even if they are temporarily hidden by objects or people, by utilizing the features of other people in the footage.
Here’s how it works.
- First, you select the person you want to track on the screen and start tracking. （Camera1）
- When the target is detected by cameras from different angles, the color frame and detection time are displayed. （Camera2）
- Tracking continues even when the target person is temporarily hidden behind objects or people. （Camera3）
Features of Multi-Camera Tracking technology
There are two key features in Multi-Camera Tracking technology. The first is a highly accurate person re-identification technology between multiple cameras. By continuously extracting and updating the appearance features of the person being tracked, it can handle the differences in how people look in each camera. This makes it possible to track a person across multiple cameras.
Another feature is the person tracking technology that can handle the situation even when they are hidden behind objects. It compares the features of people in the video over time, and if the level of similarity drops significantly, it temporarily cuts off the person's trajectory. The trajectory is then reconnected based on the comparison with the features of people around the person. This technology achieved the world's highest accuracy in the global benchmark for person tracking, the Multiple Object Tracking Benchmark (MOT-17, 2023.3, MOT-20,2023.10).
Interested in testing the Fujitsu Kozuchi?
This technology is still under development, and we are working to improve it further. In the future, we aim to track people accurately even in very crowded situations or when they wear similar clothes. We are speeding up our research and development to make this technology practical as soon as possible. Our goal is to provide the ultimate peace of mind and safety.
For a demonstration or to test our Multi-Camera Tracking, please contact us here:
