Optimized charging-station placement and UAV trajectory for enhanced uncertain target detection in intelligent UAV tracking systems

Unmanned Aerial Vehicle (UAV) technology is proposed to improve social safety, provide specialized services, and improve overall well-being in crowded indoor spaces. The deployment of drones in indoor environments can improve emergency response time, offer various wireless services, allow efficient tracking, and improve awareness in crowded scenarios. In this paper, we propose a UAV-based tracking framework that relies on energy-limited UAVs that attempts to determine the appropriate placement of UAV charging stations (CHSs) and design a UAV path planning strategy to effectively carry out detection/tracking tasks of uncertain phenomena. The proposed framework comprises a CHS placement method and a UAV path planning algorithm. The CHS placement method attempts to find the optimal placement of a given number of available CHSs so that the energy consumed by a UAV to reach the nearest CHS is reduced. This, consequently, preserves the UAV's energy, reducing the time required to return to the CHS and the period of none-tracking during the return time to the CHS. This can extend the tracking mission time and enhance detection performance. Based on the obtained optimal CHS placement, we design a reinforcement learning (RL)–based UAV trajectory algorithm to effectively detect and track a target (event of interest) with unknown behavior. The proposed RL-based UAV trajectory algorithm leverages long-term spatio-temporal behavior knowledge of uncertain targets (i.e., observed and learned events) to improve detection accuracy. Improving the detection of uncertain targets leads to better decision-making, faster responses, and improved security, safety, and efficiency in applications such as surveillance, defense, and search and rescue. The simulation results demonstrate the superior detection accuracy achieved by the proposed framework. Compared to a reference RL-based approach, the proposed algorithm achieves up to 65% higher detection accuracy in symmetric monitored areas and 20% increased accuracy in asymmetric monitored areas.