Real-Time Aerial Object Detection for Collision Prevention Using Roboflow Image Detection
Ashwin Varier, Navdeep Malik, Manju Khanna · 2025
Air traffic collision avoidance is essential to ensure the safety of increasingly congested airspace. This paper introduces an innovative, real-time aerial object detection system to prevent collisions in increasingly crowded airspace. Utilizing a pre-trained YOLO model coupled with Roboflow for efficient image annotation, the system effectively classifies and identifies various aerial objects such as airplanes, helicopters, hot air balloons, and birds from a personal dataset of 700 images. During the detection stage, bounding boxes are used to calculate spatial coordinates and dimensions, and the Closest Point of Approach (CPA) algorithm calculates collision risk. The methodology adopts a dual-framework approach where Python is utilized for object detection and data export to Excel, and MATLAB is used for subsequent visualization and accurate spatial analysis by plotting object centers and inter-object distances. Experimental findings show high precision and recall in detection, with the system having mAP values of over$\mathbf{8 8 \%}$in nearly all classes, while some categories like hot air balloons experience higher rates of misclassification. Results demonstrate the feasibility of this methodology for enhancing airspace safety, with potential applications in surveillance, navigation, and autonomous vehicle systems.