Sensor fusion under unknown associations by particle filters with clever proposal

A new method for sensor fusion under unknown associations among multiple sensors is proposed. Fundamental problem within the sensor fusion situation is huge number of the associations that prohibits enumerating all the combinations within tractable computational time. Proposed method formulates this situation in a state space model, which is highly nonlinear to deal with the unknown associations, and utilizes particle filters to estimate state of the model. Then we obtain state of the target system as well as the associations through the state estimation. We also propose clever proposal in the framework of particle filters that draws efficient particles in a sense of sub-optimality to minimize the variance of particles´ weight. The proposed method is formulated in generic way, so, in principle, it can be applied to various situations for sensor fusion under unknown associations. We show an illustrative example to track sound target in a scene with sensors of two microphones and one camera.