|摘要: ||本研究使用低價的CMOS視覺感測元件製作機器視覺系統，修改市售CMOS視覺模組的部份限制，發展適合機器視覺系統的影像擷取技術。修改重點包括影像擷取時的環境設定與影像編碼格式，以及影像擷取後的影像處理程序等。首先以改變視覺模組工作頻率的方式，產生不同的曝光程度以得到不同的影像品質。並將影像劃分為數個子影像加以分析，選擇出最佳品質的子影像，組合後做為後續影像處理的原始影像。其次，使用感測元件輸出的影像編碼格式進行影像傳輸與處理，具有影像資料量小且訊息完整的優點。最後，針對原始影像運用影像分割技術找尋所需要的物件，再輸出物件位置給控制系統做後續處理。視覺模組工作頻率的變動方面，將以數位邏輯電路形式實現在Altera的FPGA晶片中，驅動義統電子低價位的ET21X110A視覺模組。並在Altera的FPGA晶片中自行設計Nios CPU，以實現品質測定、影像邊緣強化、影像放大與影像分割等程序。最後利用UART串列傳輸輸出到控制系統。發展的機器視覺系統應用在光碟拷貝機的感測與檢測程序，包括有無碟片偵測、托盤伸出長度量測、與碟片中心點位置量測等。未來也可應用在動態環境的感測與檢測，例如當做機器人的視覺系統以回授周遭訊息。|
In this research, a Machine Vision System (MVS) is developed based on a low-cost CMOS vision module. Many limitations on image processing for the CMOS vision are investigated and improved in order to design an applicable MVS for industry applications. These problems include the environment setting during the process of image acquisition, the image encoded format, and the image processing after the image is acquired. First, in image acquisition setting, the working frequencies of the vision module are varied to perform different exposure time and obtain different quality of image. After that, an image quality selection procedure is executed by analyzing the sub-images. Sub-images with best quality are collected to reform a full image. Second, the original image encoded format of the CMOS module is utilized for image transmission. Therefore, the transmitted image will preserve complete but small-in-volume information. Finally, utilize the technique of image segmentation to separate the image pixel of the tracked object from those of the environment. Then the information about the tracked object can be abstracted and fed back to the mechanism control system.
In hand-on practice, all of the mathematical algorithms for image acquisition and processing are programmed in an Altera FPGA chip using SOPC design tools. Four functional modules, namely working frequency selection, image acquisition, signal hand-shaking, and UART transmission, are designed based on VHDL. Furthermore, a Nios CPU is designed to perform the procedures for image selection and processing. The developed MVS is utilized in the automatic sensing and inspection processes of a CD/DVD duplicator. Three experiments are implemented, including logic detection of the existence of a compact disc, one-dimensional measurement of stretch length of a CD tray, and two-dimensional measurement of the center position of a compact disc. In the near future, an attempt will be made to apply this vision system to a dynamic environment, for example, the sensing device for a mobile robot.