Αναγνώριση πεζών από αυτόνομο ρομποτικό όχημα με χρήση τεχνητής νοημοσύνης και τεχνικών βαθιάς μάθησης
Autonomous robotic vehicle pedestrian detection using artificial intelligence and deep learning techniques
KeywordsΡομπότ ; Robot ; Autonomous Robot ; Autonomous Robotic Vehicle ; NN ; CNN ; DL ; ΑΙ ; RCNN ; Αυτόνομο αυτοκίνητο ; Αυτόνομο όχημα ; Self-driving car ; Artificial Intelligence Robot ; Αναγνώριση πεζών ; Μηχανική όραση ; Μηχανική μάθηση ; Τεχνητή νοημοσύνη ; Νευρωνικά δίκτυα ; Βαθιά μάθηση ; Αναγνώριση προτύπων ; Βαθιά συνελικτικά νευρωνικά δίκτυα ; Ρομποτικό όχημα ; Pedestrian detection ; Machine vision ; Machine learning ; Artificial intelligence ; Neural networks ; Deep convolutional neural networks ; Autonomous robot ; Autonomous robotic vehicle ; Artificial intelligence robot ; Deep learning robot
The adoption of modern technologies with the aim of improving and eliminating all the basic problems that afflict today's society, is a non-negotiable need for the modern scientific community. In the context of implementation of this philosophy, research and development of autonomous robotic vehicles will potentially result in the elimination of a multitude of problems facing humanity, while actively contributed decisively to the standard of living and the environment protection. The rapid progress made in recent years, both in the technology of autonomous robotic vehicles, and technology in general, due mainly to the equally rapid evolution that has made the field of Artificial Intelligence. Artificial Intelligence, including a multitude of sophisticated algorithms, techniques and methods, achieves the application of the human brain operation through algorithms to common computer systems. Therefore, the imitation of human cerebral function during a vehicle driving procedure, by a computer, requires the implementation and integration of complete and sophisticated modeling algorithms which implement all the required activities using algorithms, techniques and methods of Artificial Intelligence, particularly Deep Learning, which is the most advanced field of Artificial Intelligence that contributes to the effective identification of prototypes in visual environments. The aim of this dissertation is both the presentation and analysis of technology used in autonomous robotic vehicles, and secondly, the implementation of an advanced Deep Learning algorithm that is able to detect - through appropriate optical sensors used into an autonomous robotic vehicle - pedestrians within the analysis range of optical sensors. The first chapters of this dissertation have been implemented at bibliographic level, and include topics that present the basic concepts, historical evolution elements and structure systems for an autonomous robotic vehicle. Thereinafter, the following chapters are implemented through a combination of bibliographic and example-based content, that contributes to the further understanding of the presented concepts. In conclusion, the last part of this dissertation presents the development of a pedestrian detection Deep Learning algorithm, the results produced during algorithm testing procedure and also the proposal of possible further improvement points of this algorithm, based on the modern Artificial Intelligence tools.