CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Prototipado de visión por computadora para la detección de objetos en drones autónomos

Estudiants que han llegit aquest projecte:

Director/a: VALERO GARCÍA, MIGUEL

Departament: DAC

Títol: Prototipado de visión por computadora para la detección de objetos en drones autónomos

Data inici oferta: 18-07-2023     Data finalització oferta: 18-03-2024


Estudis d'assignació del projecte:
    GR ENG SIS TELECOMUN
    GR ENG SIST AEROESP
    GR ENG TELEMÀTICA
Tipus: Individual
 
Lloc de realització: EETAC
 
Paraules clau:
Dron, tello engineering system, python, roboflow, drones autónomos, visión por computadora
 
Descripció del contingut i pla d'activitats:
 
Overview (resum en anglès):
In the contemporary era, drones have transcended their initial function of capturing aerial images to become versatile tools with applications in multiple fields. This project focuses on exploring the autonomous vision capabilities of drones, particularly in the use of machine learning for object detection. The primary goal is to challenge the negative prejudices associated with the military use of drones and to highlight their transformative capabilities in civilian and recreational contexts.
The project spans from exploring the physical limits of drones to developing adaptable algorithms through the fusion of programming and computer vision technologies. An exhaustive process of prototyping and implementing an object detection algorithm has been carried out, integrating trained models with advanced tools such as Roboflow and using the Tello drone as a test platform. The result is a system capable of capturing real-time images, processing them, and detecting objects accurately, demonstrating the autonomous vision capabilities of these devices.
The project is structured into chapters that detail each stage of development, starting with the introduction of the Tello drone and its engineering ecosystem, followed by programming in Python and the use of specific libraries. A description of the image annotation process with CVAT is included, which is crucial for creating a high-quality dataset. Subsequently, in the model training phase in Roboflow, model configuration, hyperparameter tuning, and performance evaluation are covered.
The integration of the trained model in Python, using PyCharm, addresses real-time video capture and processing from the Tello drone and the use of advanced inference tools. Significant technical challenges were encountered at this stage, such as connecting to two networks and optimising image processing, which were addressed through innovative solutions and constant testing.
In the conclusion, the achievements are reflected upon, and recommendations for future work are provided, establishing a solid foundation for future studies in autonomous drone vision and inviting the community to build upon these foundations. Besides its technical contribution, the project aims to change public perception of drones, promoting their use in accessible and recreational contexts. The innovative synthesis of technologies presented in this project heralds an exciting future where drones and machine learning converge to inspire significant advancements in multiple disciplines.

© CBLTIC Campus del Baix Llobregat - UPC