Real-time object detection on DJI Ryze Tello quadrotor drone with YOLO and SSD algorithms

Drones equipped with real-time object detection offer a novel perspective on our environment, yet they encounter challenges related to hardware limitations and the demands of complex applications. These challenges necessitate innovative solutions, particularly given the potential benefits in areas such as search and rescue operations, agriculture, infrastructure monitoring, and environmental conservation. This study investigates the feasibility of real-time object recognition on the DJI Ryze Tello drone using two advanced algorithms: YOLOv3 and Single Shot Multibox Detector (SSD). The evaluation of these algorithms focuses on three key metrics: accuracy, processing speed, and resource consumption. Experiments were carried out in a controlled indoor setting and in the field, with changing lighting as well as background complexity. The results showed that SSD outperformed YOLOv3 in precision, recall, and F1-Score (60-70%), making it more suitable for accuracy-focused applications. In contrast, YOLOv3’s faster processing speed (50-60 FPS) is better suited for scenarios requiring quick decision-making. This research offers valuable guidance for selecting these models based on specific application needs, particularly in agriculture, search and rescue, and surveillance.