Projecte llegit
Títol: Computer vision for bird strike prevention
Estudiants que han llegit aquest projecte:
- IBÁÑEZ I BOIX, ADRIÀ (data lectura: 12-07-2023)
- Cerca aquest projecte a Bibliotècnica
- IBÁÑEZ I BOIX, ADRIÀ (data lectura: 12-07-2023)
- Cerca aquest projecte a Bibliotècnica
Director/a: MORA SERRANO, FRANCISCO JAVIER
Departament: DECA
Títol: Computer vision for bird strike prevention
Data inici oferta: 15-02-2023 Data finalització oferta: 15-10-2023
Estudis d'assignació del projecte:
- DG ENG AERO/SIS TEL
- DG ENG AERO/TELEMÀT
- DG ENG SISTE/TELEMÀT
Tipus: Individual | |
Lloc de realització: EETAC | |
Nom del segon director/a (UPC): Alberto Burgos | |
Departament 2n director/a: | |
Paraules clau: | |
Deep Learning, Artificial Intelligence, Computer vision, bird strikes, safety, air transportation | |
Descripció del contingut i pla d'activitats: | |
Overview (resum en anglès): | |
Collisions with birds cause damage to aircraft and in some cases can even cause air travel accidents. According to data from international organizations such as the Federal Aviation Administration (FAA), the radar-based tools currently used to address this problem do not solve it, as there is no indication of a decrease in the number of bird strikes. Early detection and notification to pilots of the presence of birds is key to trying to minimize the possibility that bird impacts can occur.
The objective of this project is to improve bird detection capacity in the airport environment. To achieve this goal, this work proposes that the solution could be the use of artificial intelligence based devices and computer vision. To test this hypothesis, a model based on convolutional neural networks (CNN) is selected, trained and deployed on a device for testing. To do this, research is carried out on the different strategies used to solve problems with artificial intelligence and the performance of pre-trained classifier and detector models available. To select the computer board where the model will be deployed, a discussion of Raspberry Pi¿s market performance is made. A collection of bird images is made for training the model. The prototype will finally consist of deploying the model on a Raspberry Pi that through a script in Python programming language is able to automatically notice birds in the real world using a camera connected to the Raspberry Pi. If any detection occurs, the model is capable of making a notification that could serve to anticipate impacts and thus allow appropriate preventive measures to be taken beforehand. In conclusion, this technology shows great potential to support existing solutions today. Theoretical results with validation images show accuracy and recall parameters above 90% but experimental tests with the prototype do not allow for a conclusive judgment due to limitations regarding the training data set. |