由於單一鏡頭所能拍攝的視野有限，在需要大範圍視野的影像監控應用上是一大瓶頸。本研究基於雙攝影機架構，設計及實作一個適合應用於大型監控場景的即時連續影像接合系統，其目標在於達到串流視訊的即時效能同時確保影像接合區域的連續性。 本系統將繁複影像接合程序分割為離線校正階段與影像接合階段，用以減輕連續影像接合時的效能負擔；在影像扭曲校正方面，我們提出一區域性校正扭曲模型，可應用於非徑向扭曲影像。另外，本研究也提出一個方法，成功克服前景移動物於影像重疊區域出現鬼影的現象。最後，我們合併兩次座標轉換提出一次性影像插補方法以及加入離散小波轉換，除了大幅改善接合影像品質，也同時提升了系統效能。 影像接合階段演算法於個人電腦以及嵌入式處理器NIOS平台完成了軟體驗證。隨後我們分析各軟體模組執行時間找出系統瓶頸，再進行嵌入式硬體設計。我們以實驗室提出的MIAT高階系統硬體設計方法論將複雜計算模組合成至硬體電路，最後於元件可程式邏輯閘陣列(Field-Programmable Gate Array, FPGA)進行硬體驗證。 經由軟硬體驗證和實驗結果，顯示本系統不僅能針對具有移動物的視訊進行高品質的影像接合，軟硬體整合後更可達到即時需求。Because of the limited vision, a single lens is not appropriate for video surveillance applications in wide-angle vision. Based on the usage of dual cameras, a real-time continuous images mosaic system which is suitable for wide-angle monitoring is proposed in this study. The system can provide satisfactory real-time streaming video performance while ensuring the continuity of bonding area. To reduce the computational burden, the complex image bonding procedure is divided into off-line calibration and image stitching phases. In addition, a regional distortion model for calibrating non-radial distortion images is developed. Moreover, the ghost image of overlapping region is removed by a new measure. Finally, both the one-time interpolation approach proposed in this study and the discrete wavelet transform can significantly improve image quality and system efficiency. The system software has been verified in a personal desktop and an embedded NIOS processor, followed by an analysis of the execution time of each software module to identify and resolve system bottlenecks. The software module is then synthesized into a hardware circuit module by the MIAT hardware synthesis methodology and verified on a FPGA(Field-Programmable Gate Array). The module verifications and experimental results show that the proposed system can provide a high-quality and high-performance solution, even in the case of moving objects.
