Inimeste loendamine autonoomsete droonide ümber

Abstract

The goal of this bachelor’s thesis was to develop a prototype for indoor people counting using a mobile drone platform based on machine learning and computer vision models. The system integrates four thermal cameras, various sensors, and ESP32-based microcontrollers that collect real-time data and transmit it to a web-based user interface. To train and evaluate the models, three different experiments were conducted: in a controlled environment, in a real office setting, and under conditions where the drone was in motion. Based on the results, the most suitable models were selected, with YOLOv9s standing out for its high accuracy in real-time detection. This work shows that combining energy-efficient sensors with thermal image-based computer vision enables a mobile, accurate, and privacy-friendly indoor people counting system, offering an alternative to traditional cameras.