Multi sensor data fusion based on Kalman filter and D-S evidence theory
Ying Liu, Meng Gao, JiaJun Gao · 2024
With the advent of the big data era, data fusion technology has garnered increasing attention. Multi-sensor data fusion technology addresses the limitations of single sensors, significantly enhancing system reliability and accuracy. This article proposes a multi-sensor data fusion technique based on Kalman filtering and D-S evidence theory for gyroscope data fusion in multiple inertial measurement units. Initially, dynamic threshold adjustment is employed to detect abnormal data and make real-time threshold modifications. Following this, Kalman filtering is used to predict and update the system state, enabling dynamic adjustments to the state estimation and generating the predicted values. For each sensor measurement, D-S evidence theory is used to compute the basic probability assignment (BPA) relative to the predicted value, evaluating the credibility of outputs from various sensors. The BPAs from different sensors are then fused according to the D-S rule. The fusion result is subsequently obtained by calculating the weighted sum of the Kalman filter’s estimated values and the weighted median of the fused results. To ensure stability, a smoothing factor is introduced to mitigate drastic fluctuations in the fusion results. Experimental analysis demonstrates that the fusion algorithm proposed in this paper achieves superior performance and accuracy compared to several other comparison algorithms.