Automatic Drone Patrol of Parking Lots Using YOLO and MiDaS Neural Networks

In recent years, with the rapid development of drone technology, its potential applications across various fields have gradually been unearthed. Particularly in the realm of street and parking lot surveillance, automated drone patrol systems are viewed as an innovative and effective solution. Although numerous studies have discussed the application of drones in road monitoring and obstacle avoidance, few have integrated these aspects, especially in the context of automated patrolling. Existing surveillance systems often rely on static cameras, which struggle to adapt to the complex and dynamic environment of parking lots and fail to update the dynamic information of vehicles in real time. To address this issue, this study proposes an automated drone patrolling method for parking lots, integrating advanced image recognition technology and deep learning models. We employ the YOLOv5 and MiDaS models, where YOLOv5 is responsible for real-time detection of vehicles and license plates, while MiDaS estimates the scene's depth information. This approach not only enhances the drone's ability to autonomously identify and track vehicles but also improves environmental perception through deep learning models, enabling the drone to accurately recognize every vehicle and its license plate within the parking lot and understand the relative position and depth relationship between vehicles. Through multiple simulation tests conducted on Unreal Engine using the Airsim drone simulator, we have demonstrated that our method achieves a recognition success rate of over 80% at low speeds, effectively showcasing its potential and effectiveness in practical applications.