Sensor fusion for relative altimetry using a hybrid Gaussian mixture filter

We consider the problem of merging measurements when the sensors are likely to be affected by multiple malfunctions. An hybrid model is introduced which describes the dynamic behavior of the various sensors not only under each operating mode, but also during mode transitions. The data fusion problem is then written within the Bayesian probabilistic framework, as an estimation problem. Its optimal solution can be approximated numerically in multiple ways. The Gaussian mixture approach is well adapted because of the multiple hypothesis context. This led us to the development of an hybrid Gaussian mixture filter. The application we dealt with is the improvement of the relative altitude sensing function onboard aircraft. It aims to deliver the height of the aircraft above the ground. Currently this function is only supported by the radio-altimeters system along with specific thresholding and voting logics. Our approach leads to a joint estimation of the relative height and of the sensors operating modes. It thus allows to combine various sensors while monitoring their modes easily. We particularly focus on the case of two radio altimeters, which is the current situation.