本文關(guān)鍵詞：BP神經(jīng)網(wǎng)絡(luò)聯(lián)合模板匹配的車牌識別系統(tǒng)，由筆耕文化傳播整理發(fā)布。

BP神經(jīng)網(wǎng)絡(luò)聯(lián)合模板匹配的車牌識別系統(tǒng)

License plate recognition system using a BP neural network and template matching

[1] [2] [3]

GUO Runhua, SU Tingting, MA Xiaowei （Department of Civil Engineering, Tsinghua University, Beijing 100084, China）

清華大學(xué)土木工程系,北京100084

文章摘要：車牌識別是智能交通系統(tǒng)的重要部分，對實時自動監(jiān)控具有重要的意義。用科學(xué)的方法識別字符，同時提高識別準確率是改善車牌識別系統(tǒng)的核心問題。由于采集的車牌圖像會存在噪音和干擾，，現(xiàn)有方法存在識別效率低下的問題。本文基于Matlab平臺，聯(lián)合應(yīng)用BP（backpropagation）神經(jīng)網(wǎng)絡(luò)和模板匹配方法優(yōu)化車牌字符識別方法，在神經(jīng)網(wǎng)絡(luò)收斂的情況下嵌入模板匹配方法精確識別。針對車牌的特點提出一種高效的神經(jīng)網(wǎng)絡(luò)字符特征提取方法，從單字符的800個像素特征中僅提取211個特征向量。該方法識別率高（97．2％）、識別時間短（單字符0．02S和全車牌0．39s）、抗干擾和容錯性強。在理論創(chuàng)新的基礎(chǔ)上設(shè)計了LPR2011車牌識別系統(tǒng)，可應(yīng)用于實際交通管制。

Abstr：License plate recognition is an important part of intelligent transportation systems with great significance for real time automatic road monitoring. The recognition accuracy rate is the core issue of the system. Existing methods are not effective due to noise and interference. This research describes license plate recognition method that combines a BP （back propagation） neural network and template matching. The method embeds template matching in the convergent neural network for further identification. The algorithm uses an efficient feature extraction method for the neural network which extracts only 211 features out of 800 pixels. The method achieves a high recognition rate of 97.2~ with rapid processing （0.02 s for single character and 0.39 s for an entire license plate）, resists interference and as tolerant to various faults. The system can be applied to actual traffic control.

文章關(guān)鍵詞：

Keyword：:traffic control license plate recognition character recognition BP （back propagation） neural network template matching feature extraction

課題項目：國家自然科學(xué)基金資助項目（51008169）

作者信息：會員可見

  本文關(guān)鍵詞：BP神經(jīng)網(wǎng)絡(luò)聯(lián)合模板匹配的車牌識別系統(tǒng)，由筆耕文化傳播整理發(fā)布。