Projecte llegit
Títol: Deep-learning based pipeline for high voltage tower detection
Director/a: MESEGUER PALLARÈS, ROC
Departament: DAC
Títol: Deep-learning based pipeline for high voltage tower detection
Data inici oferta: 09-02-2021 Data finalització oferta: 09-10-2021
Estudis d'assignació del projecte:
- MU DRONS
Tipus: Individual | |
Lloc de realització: Fora UPC | |
Supervisor/a extern: Miquel Mulet | |
Institució/Empresa: Venturi Unmanned Technologies SL | |
Titulació del Director/a: Aerospace Engineer (degree) | |
Paraules clau: | |
drone, UAV, RPAS, UAS, tensorflow, detection, IA, machine learning, darknet, YOLO | |
Descripció del contingut i pla d'activitats: | |
Implement an own version of image-filtering step in python-opencv
to extract high probability areas containing electric towes on high resolution images. Do a benchmark of available CNNs to select the most suited for this usecase to run on the cropped images. Expected result is a fast and low-latency method for the tracking of electric towers taken from aerial images that can run on a NVIDIA Jetson NANO embedded platform. The student will be working full time (9 AM to 6 PM) mostly dedicated at writing the master thesis. |
|
Overview (resum en anglès): | |
The aim of this thesis is to build a complete system for automatize the regular inspection of power lines and to ease its maintenance. It will be carried in a flying drone. It should be able to detect the electric towers, track them and facilitate the data extraction from the images obtained.
This master thesis develops a complete prototype for detecting and tracking electric towers from a flying drone. It means, the design of the object recognition system, the tracking mechanism and the final embedded system that will be carried on a drone. In order to achieve the objective some tools will be analysed, discussed and checked its functionality, to ensure the viability of the entire structure. The algorithm for the prototype will be fully written in python, some useful libraries will be used to help building the model, such as OpenCV, simpleBGC, pyyolo, etc. The hardware set up will be, a gimbal camera for obtaining the images and visually tracking the power lines and an SBC (Single Board Computer) where all the algorithms for detecting and tracking will run. Some experimental designs will be presented, from a less complex environment to a more complicated one. Finally, it will end with a demonstrated simulation of the complete prototype for detecting and tracking electric towers. Although the system could not be tested on a real situation, the results obtained in realistic simulations are considered good and feasible for switching to real life as soon as possible. The prototype works as expected and detects the towers with low error range. |