自動化車牌辨識系統

Abstract

車牌辨識系統已廣泛應用於生活各方面, 諸如: 停車場管理系統、高速公路收費系統等。然而探討現有的辨識系統, 在處理畫面上皆有著相當的限制, 為了達到及時運算的效果, 大多都需要加裝額外的感應器, 又或者僅只能辨識單一車道。 本論文期望能直接利用各路口的監視器畫面, 達到多車道的辨識效果, 且無須額外的輔助裝置。利用主成分分析判斷監視器畫面中的背景以及前景車輛位置, 藉此大量降低搜索車輛所需時間。同時本研究蒐集各種不同狀況的車牌樣本, 包含反光、髒汙、陰影等因素, 因應不同狀況開發穩健的車牌辨識模組。模組中以索貝爾邊緣偵測以及高斯遮罩擷取車體中的車牌區塊, 接著分析連接像素區以切割車牌中的字元, 最後採用支持向量機辨識字元。 最後以一段高畫質的模擬影片檢驗本系統成效。結果顯示本系統可達即時運算,並且達到91.35% 的準確率。

Car plate recognition system is widely used in all aspects of life, such as parking lot management system and highway toll collection. However, existing recognition systems are limited by image processing. To recognize car plates instantly, extra sensors are installed on most of systems, or a region of interest is defined on single lane. Our aims of research are to process surveillance video directly without additional devices, and to achieve multi-lane recognition in real time. By principal component analysis, we can separate the cars from the background, therefore plenty of processing time has been saved. We also develop the robust recognition module based on different samples, which are influenced by environmental factors such as reflection, dirt and shadow. The module implements plate segmentation by Sobel edge detector and Gaussian filter, analyzes connected component in plate to segment characters and uses support vector machine to recognize characters. In the last chapter, we examine the system with a high-definition video. The result show that our system can recognize the simulation video in real time, and the overall rate of success is 91.35%.