Abstract

With the acceleration of urbanization, the limitations of traditional fixed monitoring systems in terms of field of view and flexibility are becoming increasingly apparent. Low-altitude Unmanned Aerial Vehicles (UAVs), leveraging their aerial perspective and high mobility, provide a new technological pathway for constructing a dynamic and intelligent security patrol system. This study aims to build an integrated low-altitude UAV intelligent patrol system encompassing perception, identification, localization, and early warning, focusing on solving the key problems of accurate identification and rapid localization of abnormal behaviors in complex scenarios. Methodologically, to address challenges such as significant scale variations of targets and complex backgrounds from the UAV perspective, mainstream object detection algorithms are optimized by introducing attention mechanisms to enhance robustness against small and occluded targets. A multi-modal behavior recognition strategy based on feature-level fusion is designed to improve the system’s adaptability during nighttime and under adverse weather conditions. For localization, a collaborative positioning scheme integrating the Global Positioning System (GPS), Inertial Navigation System (INS), and visual information is proposed to achieve high-precision geographic coordinate calculation and trajectory tracking of identified targets. Through the combined application of deep learning model optimization and multi-source information fusion technology, this study effectively enhances the performance of the UAV patrol system in terms of abnormal behavior recognition accuracy and target localization precision, holding significant theoretical value and practical importance for improving the intelligence level of social security governance.