架構於數位訊號處理器之車牌辨識系統

Abstract

車牌辨識系統已經漸漸成為許多交通應用技術方面的關鍵。 其中一些系統已經可以運用在實際的情形下，例如管理私人的停車場，高速公路，以及管制工廠的車輛的進出等等。然而，由於車牌辨識的演算法則都需要相當龐大和複雜的運算，以至於大部分的車牌辨識系統都必須架構於電腦上，再由攝影機所擷取到的車輛影像透過網路後傳到電腦作處理。顯然易見地，由於網路上傳輸的限制，要從解析度低的影像中得到可靠的辨識結果是困難的。 既然要拆卸現有的系統來建造更新、更快的傳輸網路是不實際的，那麼將擷取影像的攝影機和車牌辨識系統結合的方式未嘗不是另一種有發展潛力的方法. 因此，本篇論文的目的便在於驗證嵌入式車牌辨識系統的可行性與潛力。透過將車牌辨識的演算法則以及車輛影像放入DSP TMSC6701 實驗板的方式，經過重複的實驗得知，其辨識率約為95%，辨識時間為0.95秒。證明了在未來，嵌入式的車牌辨識系統是具有發展潛力與可行性的。

Vehicle license plate recognition system becomes the key to many traffic related applications. Several systems have been applied practically, such as for managing a private parking lot, highways, and monitoring vehicle entry and exit at a factory. However, most of vehicle license plate recognition systems are developed based on PC because the algorithms to recognize license plate is complex and need a number of mathematical operations. Images containing the vehicle are acquired by the sensor like machine vision, and transferred to the image processor through the communication network between them. It is obvious that getting reliable recognition results from poor-quality images mainly caused by restriction on the provision of the efficiency and correctness is difficult especially for the system to identify license plate. Since scrapping existing road-vehicle systems and building new transportation systems from scratch are not practical, there is potential for combining the system to identify the license plate with the sensor capturing the image containing the vehicle. The object of this paper is to verify the potential and practicability of embedded vehicle license plate recognition system. It is to recognize license plate based on DSP TMSC320C6701 EVM board. And the image includes a car is loaded into the memory on the board. As experiment results present, the recognition rate efficiency reach up to 95% and the recognition time is 0.95 second. It conforms that embedded vehicle license plate recognition system has potential and is practicable in the future.