CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Models de detecció d'objectes per a identificar debris a pistes d'aterratge mitjançant l'ús d'un dron

Estudiants que han llegit aquest projecte:

Director/a: FORNÉS MARTÍNEZ, HECTOR

Departament: FIS

Títol: Models de detecció d'objectes per a identificar debris a pistes d'aterratge mitjançant l'ús d'un dron

Data inici oferta: 09-07-2024     Data finalització oferta: 09-07-2024


Estudis d'assignació del projecte:
    DG ENG AERO/TELEMÀT
Tipus: Individual
 
Lloc de realització: ERASMUS
 
Paraules clau:
Object Detection models, Labelling, YOLO, CNN,
 
Descripció del contingut i pla d'activitats:
 
Overview (resum en anglès):
This final degree project has investigated the feasibility of an object detection system using the YOLOv7 algorithm and the LabelImg labelling tool. The main objective was to analyse how this technology can be applied to object detection in airport environments. Despite practical limitations due to adverse weather conditions that prevented sufficient data collection, a comprehensive theoretical analysis has been conducted.
The project initially planned to use drones for image capture, leveraging their unique perspectives. However, operational and safety restrictions led to the exploration of alternative methods such as static cameras for data acquisition. These alternatives can offer greater robustness and efficiency to the detection system.
The theoretical study focused on the configuration and training of the YOLOv7 model, highlighting its ability to process images in real-time with high accuracy. Emphasis was also placed on the importance of high-quality labelled data, which is essential for obtaining a valid object detection model, from the initial to subsequent iterations to perfect the model. Throughout the project, the use of Roboflow for dataset management was considered due to its use in a previous practice, but LabelImg was ultimately chosen due to the tutor's familiarity with this tool.
Despite not being able to carry out the practical part of the project on a large scale, the theoretical results obtained provide a solid foundation for future research. This work represents a theoretical contribution to the field of computer vision, offering a clear guide on how to approach the implementation of YOLOv7 in a real environment and highlighting the necessary technical considerations.
Although no debris detection system has been implemented in any Schengen country, EASA has already highlighted this use case of artificial intelligence in its latest AI Roadmap for Aviation. A similar project, scaled appropriately and with the necessary resources, could represent a significant innovation compared to current systems based on radar technology and optical cameras.

© CBLTIC Campus del Baix Llobregat - UPC