Application of Raycast Method for Person Geolocalization and Distance Determination Using Uav Images in Real-World Land Search and Rescue Scenarios

Abstract

The inclusion of drones in Search and Rescue (SAR) operations has enabled the use of computer vision methods to detect persons in aerial imagery automatically. When vertical imagery is used, assuming the correct telemetry of the drone, the location of the detected person thus found can be determined directly by reading the GPS coordinates of each photograph in which the person was detected. However, since the goal of SAR operation is to search the largest area of ​​the territory in the shortest time possible and find a lost or injured person, in practice, when searching by drone, preference is given to oblique photographs. By using them, a larger area can be covered within a single image, reducing the search time. Unlike vertical photographs, oblique photographs have a trapezoidal footprint, and they include a significant change of scale, making it difficult to locate a person in the real world and determine their distance from the drone. Encouraged by previous in silico experiments using the raycast method, we applied the raycast method in vivo to solve this problem. After a series of experiments on terrains of different configurations and complexity, using a custom-made 3D terrain generator and raycaster, along with a YOLOv4 person detector trained on our custom dataset and a low-cost commercial drone, we achieved the accuracy of locating the person within 0.7 m in relation to the ground truth, as the best result. As a conclusion of the analysis of the experiments, we made recommendations for using the raycast method in real-world scenarios.

Keywords: raycasting, drone imagery, Object Detection, odject geolocalization, distance determination, search and rescue missions