Aristotle University Of Thessaloniki (AUTH)
Ανθρωποκεντρικές τεχνικές υπολογιστικής όρασης για αυτόνομα συστήματα
Abstract
dc:descriptionThe aim of this PhD thesis was the development and study of computer vision and machine learning techniques related to the analysis of images and image sequences depicting people (individuals or in crowds). These techniques, with the exception of a technique involving the propagation of person identity labels in facial images, are intended to assist in the operation and navigation of autonomous systems (robots, unmanned aerial vehicles, UAVs) or involve such systems for more efficient data collection (active vision). The first chapter tackles the subject of face recognition, and deals with speeding up and increasing the performance of a (published in the literature) person identity label propagation algorithm, namely the Multiple-graph Locality Preserving Projections - Cluster-based Label Propagation, MLPP-CLP algorithm. An incremental label propagation technique is also proposed which can reduce the computational complexity of facial image label propagation when applied to images whose number increases with time, e.g. frames from a video feed. The second chapter provides an overview of computer vision techniques (e.g., for landing site detection, semantic mapping, semantic segmentation, obstacle avoidance etc.) that contribute to flight safety, a critical issue for the navigation of unmanned aircraft. A summary of relevant legislations and regulations around the world is also provided. These techniques are relevant to any type of autonomous UAV flight (e.g., for film/television production, search/rescue, inspection, mapping, crowd management) making this topic of broad interest. The third chapter presents tools that can be included in algorithmic UAV flight safety pipelines. First, a technique is presented to locate a potential (UAV) landing site using terrain information through flat area detection. Also, a method for the labeling of 3D maps with crowd areas is proposed. The method projects 2D crowd annotations derived from UAV video analysis on a 3D map using the UAV camera parameters. The fourth chapter presents a method for robust, on-the-fly 3D crowd detection and mapping in largescale outdoor environments. The method utilizes and fuses the independent crowd detections, by the corresponding detectors, in videos captured from the cameras of multiple UAVs. The final chapter tackles a topic of active computer vision involving robotic autonomous systems. Specifically, by trying to address the lack of active face recognition algorithms, in this chapter, we propose such an approach that uses photorealistic synthetic facial images in order to guide a robot towards view points that provide increased recognition accuracy.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kakaletsis, Efstratios
- Κακαλέτσης, Ευστράτιος
Subjects
dc:subject × 22- Υπολογιστική όραση
- Μηχανική μάθηση
- Αυτόνομα συστήματα
- Γραφικά υπολογιστών
- Επεξεργασία εικονοσειρών
- Computer vision
- Machine learning
- Autonomous systems
- Computer graphics
- Digital video processing
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Γραφικά υπολογιστή και Σχεδιασμός με χρήση υπολογιστή
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Μηχανική και συστήματα επικοινωνιών, Τηλεπικοινωνίες
- Computer and Information Sciences
- Natural Sciences
- Computer Graphics and Computer-Aided Design
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Communication engineering and systems, Telecommunications
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57171