CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Reconocimiento de objetos en tiempo real mediante Deep Learning aplicado en Drones

Estudiants que han llegit aquest projecte:

Director/a: VALERO GARCÍA, MIGUEL

Departament: DAC

Títol: Reconocimiento de objetos en tiempo real mediante Deep Learning aplicado en Drones

Data inici oferta: 17-07-2023     Data finalització oferta: 17-03-2024


Estudis d'assignació del projecte:
    GR ENG SIST AEROESP
Tipus: Individual
 
Lloc de realització: EETAC
 
Paraules clau:
Reconocimiento de objetos, tiempo real, dron, Deep Learning, red neuronal, Drone Engineering Ecosystem, YOLO, YOLOv5
 
Descripció del contingut i pla d'activitats:
 
Overview (resum en anglès):
This project addresses the current challenge of real-time object recognition, focusing on the implementation of Deep Learning techniques and their applicability in unmanned vehicle systems, specifically drones. The study is structured in two phases: a theoretical phase that delves into the fundamentals of Deep Learning and its application in object recognition, and a practical phase where these concepts are implemented and validated in computational and operational drone environments.

The theoretical phase delves into the pillars of Deep Learning, ranging from diverse neural networks to activation and loss functions, also exploring weight initialization techniques, input parameters, and training outcomes. Multiple neural networks are developed and evaluated in Python for demanding tasks such as real-time object recognition, aiming to determine the viability of custom implementations versus pre-existing tools optimized for these tasks.

Subsequently, the integration of the YOLO algorithm, prominent in real-time object detection, is assessed on the drone's onboard computer. If not feasible, execution on a ground station is considered as an alternative. The work culminates in the development of a desktop application enabling route definition for the drone and selection of objects to detect along these paths. The final step involves conducting tests in authorized drone flight spaces like DroneLab to verify the system's proper functionality.

This study not only focuses on the technical aspects of real-time object recognition but also its viability and applicability in operational environments. Additionally, it emphasizes the intent to provide a tutorial facilitating future students' entry into this field within the collaborative ecosystem of Drone Engineering Ecosystem at EETAC.

© CBLTIC Campus del Baix Llobregat - UPC