Sistem za razpoznavo registrskih tablic na platformi Raspberry Pi

Rebernik, Jure (author), Žemva, Andrej (mentor)

Sisteme za razpoznavo registrskih tablic vse pogosteje srečujemo na javnih parkirnih mestih, kjer namesto klasičnega odpiranja zapornice z listki avtomobile raje identificiramo preko njihovih registrskih tablic. V tem magistrskem delu je raziskana možnost poenostavitve, posledično pa zmanjšanja cene tovrstnih sistemov, s tem pa bi lahko le-ti prodrli v domačo oz. gospodinjsko rabo. V ta namen je bil razvit sistem, ki temelji na platformi Raspberry Pi 4. Za zaznavo lokacije registrskih tablic na slikah avtomobilov je bil uporabljen model nevronskih mrež SSD MobileNet V2 FPNLite 320x320, za katerega smo izdelali lastno učno množico in ga naučili prepoznavati slovenske registrske tablice. Obrezana slika, ki vsebuje samo tablico, je bila obdelana in pripravljena za postopek optičnega razpoznavanja znakov, ki je bil izveden s pomočjo knjižnice PyTesseract. Izdelan je bil spletni vmesnik, s katerim ima uporabnik pregled nad celotnim sistemom. Predstavljena je bila tudi možnost integracije izdelanega sistema na že obstoječi sistem dvoriščnih vrat. Razviti sistem uspe obdelovati sličice videa v realnem času s hitrostjo 1,2 sličice na sekundo. Skupna cena sistema je 67 EUR.

• Language: Slovenian
• Keywords: strojno učenje, nevronske mreže, pametni dom, avtomatizacija doma, robno procesiranje
• Work type: Master's thesis/paper
• Organization: FE - Faculty of Electrical Engineering
• Year: 2023
• PID: 20.500.12556/RUL-145555
• COBISS.SI-ID: 150326275
• Publication date in RUL: 21.04.2023

Secondary language

• Language: English
• Title: License plate recognition system using Raspberry Pi

Abstract

License plate recognition systems are becoming more and more common in public parking lots, where instead of using the traditional ticket barrier system, cars are identified based on their license plates. This master's thesis explores the possibility of simplifying and subsequently reducing the cost of such systems, which could allow them to enter into domestic use. A system based on the Raspberry Pi 4 platform was developed. The SSD MobileNet V2 FPNLite 320x320 neural network model was used to detect the location of license plates in car images. A custom dataset for model learning was created, containing images of Slovenian license plates. Cropped images that contain only license plates were then processed and prepared for optical character recognition process, which was done using PyTesseract library. A web interface that allows users to have an overview over the whole system was developed. Integration into an already existing household gate system was also explored. The developed system is capable of processing video frames in real-time at a rate of 1.2 frames per second. The total cost of the developed system is 67 EUR.

• Keywords: machine learning, neural nets, smart home, home automation, edge computing