Multi-step Mutual Information Prediction for Fire-Spots Tracking in Active Sensing of Wildfires

In this paper, the problem of fire-spots tracking in wildfire sensing tasks is studied. Aiming at the time-effectiveness problem not considered in traditional fire-spot detection methods, an active sensing scheme of fire-spots based on the prediction of multi-step mutual information (MI) is proposed. The method combines with the particle filter algorithm to estimate the location of the fire-spot and establishes the long-term cost function in the process of active perception based on mutual information. A multi-step MI approximate calculation method based on observation sequence sampling is designed to reduce the computational complexity. Considering the change of the uncertainty of fire-spots during the task, a single-step/multi-step mode switching strategy was proposed, which combined the advantages of the two modes and provided a more efficient scheme for the fire-spots perception task. Simulation results indicate that the proposed method can alleviate the local optimization problem of the single-step MI method, effectively reduce the target uncertainty, and improve the tracking accuracy.