Long distance Automatic Number Plate Recognition under perspective distortion using zonal density and Support Vector Machine

摘要

Automatic Number Plate Recognition (ANPR) is one of computer vision applications to extract information in vehicles plate number. Nevertheless, perspective distortion is unavoidable when taking pictures of the plate number. Another factor that causes inaccuracy is the distance of the camera from the plate number. To solve these problems, we propose a new method to automatically detect and recognize vehicle plate number with regards to perspective distortion and distance of capturing plate number. We used zonal density with Support Vector Machine (SVM) as a classifier. We tested our algorithm on 21 vehicles plate number with 1, 3, and 5 meter of capturing distance. Our method yields an accuracy of 89.77%, 82.86%, and 65.22% for 1, 3, and 5 meters capturing distance, respectively. Compared with previous work, our method is able to preserve high accuracy when segmenting characters of plate number taken from 5 meter distance.