Parklakohtade monitoorimine ja autode ID-põhine jälgimine
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Parklate jälgimissüsteeme on aegamisi arendatud ning realiseeritud, kuid vähe on süsteeme, kus kasutataks tehisnärvivõrke. Tehisnärvivõrkude arenedes on võimalik neid ka parkla jälgimissüsteemides rakendada. Käesolevas magistritöös uuritakse YOLO närvivõrku, mis suudab pakkuda reaalajas kiirusi. YOLO eelmistest versioonidest ja eelistest teiste närvivõrkude ees tuuakse kokkuvõte. Siin magistritöös uuritakse ka tuvastatud objektide jälgimist unikaalsete ID-ga, kus viimaseid saadakse Kalmani filtrist. Seletatakse Kalmani filtri tööpõhimõtet. Ühtlasi seletatakse lahti käesolevas magistritöös kasutatud tarkvara disain ning pakutakse välja uus jälgimisüsteem parklakoha jaoks, mille kasutamist pole autori teada varem dokumenteeritud. Lõpus uuritakse tulemusi ning pakutakse välja lahendused teatud probleemidele. Ühtlasi kirjeldatakse võimalikke täiendusi, mida antud tööle teha saaks.
Parking space monitoring research has been going on for some time already, but the field is still being widely researched. As neural networks are becoming better, they can be utilised in more areas, including parking space monitoring. This thesis takes an approach with YOLO that is capable of performing detections in real-time. A short summary of YOLO's previous versions and its advantages over other neural networks is given. In addition, this thesis also takes a look at unique ID based detected object tracking by using Kalman filter. The workflow of Kalman filter is explained. The overall design of the software used in the thesis is explained and a method for parking space monitoring that, to the author's knowledge, has not been documented before, is offered. Results of the thesis are analysed and solutions are provided for some of the encountered issues. Some of the future possible improvements are explained.
Parking space monitoring research has been going on for some time already, but the field is still being widely researched. As neural networks are becoming better, they can be utilised in more areas, including parking space monitoring. This thesis takes an approach with YOLO that is capable of performing detections in real-time. A short summary of YOLO's previous versions and its advantages over other neural networks is given. In addition, this thesis also takes a look at unique ID based detected object tracking by using Kalman filter. The workflow of Kalman filter is explained. The overall design of the software used in the thesis is explained and a method for parking space monitoring that, to the author's knowledge, has not been documented before, is offered. Results of the thesis are analysed and solutions are provided for some of the encountered issues. Some of the future possible improvements are explained.