Search Area Prediction Algorithm for Tracking Missing Target in Complex Terrain Environments Using UAVs

This paper presents a novel dynamic-target search area prediction (SAP) algorithm to address the challenge of tracking missing vehicles lost in complex terrain environments. The algorithm aims to improve the effectiveness of prediction after target loss. First, a target motion prediction model based on physical rules is established. This model can accurately predict the potential movement patterns of the dynamic targets. Subsequently, the rule-based model is used as expert demonstrations in the Maximum Entropy Deep Inverse Reinforcement Learning (MEDIRL) algorithm. And the real trajectory data of the target will be used to fine-tune the network structure model obtained by Soft Actor-Critic (SAC) algorithm. This step can quickly fit the driving intentions and preferences of the current target. Furthermore, the algorithm predicts the distribution of target positions and generates a probability-based heatmap, reflecting the likelihood of target presence across the terrain. To determine the area containing the highest probability of target, a spatial sliding adaptive window (SSAW) method is employed. This approach can dynamically adjust the area based on the heatmap, focusing on the most probable region for target presence. Simulation results demonstrate that the proposed algorithm effectively predicts target areas with a higher probability of containing the targets. These predictions provide valuable reference for target tracking in mountainous terrain environments.