| KEYWORD |
Overcoming Limitations of ARUCO Markers in Close Range Space Navigation through Deep Learning
Thesis in external company
keywords DEEP LEARNING, PROXIMITY NAVIGATION, ROBUSTNESS
Reference persons MARTINA MAMMARELLA
External reference persons Francesco Rossi (AIKO Srl), Federica Paganelli (AIKO Srl), Francesco Evangelisti (AIKO Srl)
Description In the realm of autonomous spacecraft operations, close-range navigation plays a pivotal role, especially in delicate maneuvers such as docking or inspection. The use of marker-based approaches, such as ARUCO markers, has been a prevalent method to optimize navigation in these critical contexts. However, the effectiveness of ARUCO markers is significantly challenged under extreme conditions, impacting the precision of pose estimation and, consequently, the safety and success of navigation tasks. This thesis aims to delve into the limitations of ARUCO markers in close-range space navigation, characterize these limitations, and explore a Deep Learning (DL)-based solution to enhance pose estimation accuracy under stringent conditions.
The focus of this research is twofold:
1. Characterization of ARUCO Marker Limitations:
◦ Identification and analysis of conditions that negatively affect the detection and pose estimation of ARUCO markers, such as lighting variations, partial occlusions, and extreme orientations.
◦ Evaluation of the impact of these conditions on the accuracy of pose estimation in close range navigation contexts.
2. Development of a Deep Learning Solution:
◦ Proposal and development of a Deep Learning model, initially presumed to be a Convolutional Auto-Encoder (CAE), aimed at mitigating the adverse effects identified. This model will preprocess images to enhance ARUCO marker detection and pose estimation.
◦ Exploration of alternative or complementary DL models that could further improve the robustness and accuracy of pose estimation based on ARUCO markers.
◦ Implementation and testing of the developed model(s) to assess improvements in ARUCO marker detection and pose estimation under challenging conditions.
See also thesis proposal_aruco with dl_aiko_polito.pdf
Required skills SKILLS REQUIRED
• Minimum:
◦ Good proficiency in Python programming.
◦ Basic knowledge of Machine Learning and Computer Vision.
◦ Proficiency in spoken and written English.
• Optional:
◦ Familiarity with OpenCV.
◦ Advanced knowledge of Deep Learning.
Deadline 01/04/2025
PROPONI LA TUA CANDIDATURA