Computer Vision Models for Object Detection Based on Video Data from Unmanned Aerial Vehicle

This paper solves the problem of finding lost people in sparsely populated areas during search and rescue operations using images from unmanned aerial vehicles and machine vision techniques. To accomplish this, a two-step approach is proposed, consisting of identifying areas of interest and then refining their location. Various machine vision architectures are considered to identify areas of interest. These include segmentation and detection models, as well as modifications to the RetinaNet model and combinations thereof. An in-depth analysis of the quality of predictions and model performance is conducted to determine the best architecture. The paper proposes a modified version of the RetinaNet architecture that outperforms advanced detection models in terms of accuracy and speed, making it a promising solution for this task.