本文以複雜環境背景的影像串流為參考對象,攝影機架設在天橋與一般道路取樣,提出一套多目標車牌即時定位的技術。可結合車牌辨識技術,達到實際道路車輛之車牌辨識,作為一般道路車輛之自動監控系統。 本文所提的方法可於多環境的車道下進行多目標車牌定位,包含汽機車定位,更可在任意天候情況下定位。主要方法分為兩個步驟:移動物體偵測多物件處理與多目標車牌定位。移動物體偵測多物件處理部分,利用連續影像相減、候選區域分割,精準地分割出移動物體位置,於候選區域影像作空間域分析與二值化分析,使用形態學與線段連通成份標記達到多物件。應用兩種分析法中的多物件包含車牌物件正確率可達98.2%。 多目標車牌定位部分,篩選多物件的方法:車牌長寬比、車牌密度分析、字元交越之特徵、字元密度分析等,利用這些車牌的特徵來篩選車牌,最後確認出車牌的位置。應用兩種分析法的多物件,篩選出汽車車牌與機車車牌的物件,提高多環境下的多目標車牌定位正確率,使得本系統實作的定位正確率達91.7%。處理速度上:移動物體與車牌定位平均處理時間為15ms、總平均處理時間為31ms,顯示在運算速度達到即時性。以上證明我們提出的方法在實務系統上有不錯的結果。故此系統可整合現有道路監視系統資源,即時達到失竊汽車協尋以及問題車之監控。
In this thesis, the Multi Objective License Plate Locating (MOLPL) automatically system is researched and implemented on complex background video streaming in real-time systems. Cameras set up on the flyover and general roads record the sampling. The technology of Moving Vehicular License Plate Locating (MVLPL) and License Plate Recognition (LPR) for vehicular will integrate into the Vehicular Locating and Monitoring (VLM) system. The VLM system is active and can be employed on the intelligent surveillance system for the running vehicle. Moving Object Detection processing multi object use a Continuous Image Subtraction and a Candidate Region Segmentation to find moving object location. In the Candidate Region for Spatial Analysis and Binarization Analyzed is using morphological and the Line Connected Component Markers to implemented multi objects. Two analysis methods in multi object include the license plate (LP) can make the accuracy rate of 98.2%. MOLPL screening multi object of method include LP aspect ratio, LP density analysis, character crossover characteristic and character density analysis the use of LP characteristics to select LP object. Finally, confirm the position of the LP. The results of the experiments show that the performance of the MOLPL rate is 91.7%.The average processing time is 31ms, displayed in real-time computing speed. Therefore the system could combine all of the monitoring systems on road and detect the license plate of running vehicles.