License Plate Detection Using OCR Method with Raspberry Pi

Abstract

The application of The Automatic License Plate Recognition (ALPR) to overcome the weakness of reading vehicle license numbers manually is largely determined by the choice of segmentation techniques in processing the detected object image. This study shows a comparison of the performance of two segmentation methods in detecting license plate edges, namely Canny Edge and Otsu Thresholding. The license plate image data is processed through several stages before the license plate text is detected by the OCR and Teserract Library methods. Data processing is done using Rasberry Pi. The performance test of the two methods compared was carried out on 30 samples of vehicles in three time segments, namely morning, afternoon, and evening. From the results of the experiments carried out, Canny Edge shows better performance because it can detect 100% of edges compared to Otsu Thresholding which is only able to detect 70% of edges of the entire data. In general, the system built successfully detects number plates with an average accuracy of 72%.