According to different data processing methods, there are three architectures of information fusion system: distributed, centralized and hybrid.

1) Distributed: First, the original data obtained by independent sensors are processed locally, and then the results are sent to the information fusion center for intelligent optimization and combination to obtain the final results. Distributed has low demand for communication bandwidth, fast calculation speed, good reliability and continuity, but the tracking accuracy is far less than that of centralized one. Distributed fusion structure can be divided into distributed fusion structure with feedback and distributed fusion structure without feedback.

2) Centralization: Centralization sends the raw data obtained by each sensor directly to the central processor for fusion processing, which can realize real-time fusion. Its data processing accuracy is high and its algorithm is flexible, but its disadvantages are high requirements for the processor, low reliability and large data volume, so it is difficult to realize;

3) Hybrid: In the hybrid multi-sensor information fusion framework, some sensors adopt centralized fusion mode, and the rest adopt distributed fusion mode. The hybrid fusion framework has strong adaptability, takes into account the advantages of centralized fusion and distribution, and has strong stability. The structure of hybrid fusion mode is more complicated than that of the first two fusion modes, which increases the cost of communication and calculation.

Kalman filter (KF)

The process of information processing by Kalman filter is generally prediction and correction. It is not only a simple and concrete algorithm, but also a very useful system processing scheme in the role of multi-sensor information fusion technology. In fact, it is similar to many systems' methods of processing information data. It provides an effective statistical optimal estimate for the fused data by means of mathematical iterative recursive calculation, but it requires little storage space and calculation, so it is suitable for the environment with limited data processing space and speed. KF can be divided into two types: distributed Kalman filter (DKF) and extended Kalman filter (EKF). DKF can make data fusion completely decentralized, while EKF can effectively overcome the influence of data processing errors and instability on information fusion process.