A border surveillance operation requires sophisticated sensor planning, as sensors are usually scarce and cannot cover an entire region simultaneously. Border patrol agents act as moving sensors in a border region, and the border patrol agents’ coverage moves around the region dynamically, increasing the chance of approaching a trespassing agent. Typically, the locations of trespassing agents cannot be fully observed due to the size of the border region and obstacles. In addition, intelligent trespassing agents may dynamically adjust their traveling paths so that the border patrol agents cannot predict their locations easily. Trespassing agents are assumed to leave traces that indicate their footprints, providing their estimated locations. Border patrol agents may use the trespassers’ footprints as partial information to leverage patrol path planning. We propose an adaptive border patrol process as a partially observable Markov decision process (POMDP), in which an individual border patrol agent’s decision is determined dynamically on the basis of trespassing agents’ partially observed locations. The observations are shared among individual border patrol agents, allowing the border patrol agents to cooperate. The zoning technique is used to limit the planning scope of an individual border patrol agent, and Monte Carlo simulation is applied to reduce the complexity of the POMDP planning problem. Empirical experiments are conducted by means of simulated agents. The simulation parameters are derived from the interviews with a group of border patrol experts. The results in different scenarios show that the proposed patrol path planning scheme outperforms other patrol path planning schemes in terms of the trespasser detection rate. The simulation results are validated with respect to subject matter experts (SMEs), where SMEs are the same border patrol experts who had given the interviews. The proposed method has potential in border surveillance as an assisting system for human border patrol or an automated guidance system in robots or drones.

边境监视行动需要复杂的传感器计划，因为传感器通常很稀少，无法同时覆盖整个区域。边境巡逻人员充当边界区域中的移动传感器，并且边境巡逻人员的覆盖范围动态地在该区域周围移动，从而增加了接近侵入人员的机会。通常，由于边界区域和障碍物的大小，无法完全观察到侵入剂的位置。另外，智能闯入人员可以动态地调整其行进路径，以使边境巡逻人员无法轻易预测其位置。假定侵入者留下痕迹，以表明其足迹，并提供其估计的位置。边境巡逻人员可以将闯入者的足迹用作部分信息，以利用巡逻路径规划。我们提出了一种自适应的边境巡逻过程，作为部分可观察的马尔可夫决策过程（POMDP），在该过程中，根据越境人员的部分观察位置动态地确定了单个边境巡逻人员的决策。这些观察结果在各个边防巡逻人员之间共享，从而使边防巡逻人员可以合作。分区技术用于限制单个边防巡逻人员的计划范围，而蒙特卡罗模拟则用于减少POMDP规划问题的复杂性。通过模拟代理进行经验实验。模拟参数来自与一组边境巡逻专家的访谈。在不同场景下的结果表明，在侵入者检测率方面，所提出的巡逻路径规划方案优于其他巡逻路径规划方案。模拟结果已针对主题专家（SME）进行了验证，其中SME与接受采访的边境巡逻专家相同。所提出的方法在边界监视中具有潜力，可以作为人类边界巡逻的辅助系统或机器人或无人机中的自动引导系统。