軟體產品開發的過程中，通常會使用回歸測試（regression testing）確保釋出的新版本品質。因為回歸測試的過程通常繁複且耗費資源，所以許多測試工具致力於讓整個測試流程自動化。其中應用最為廣泛的方法為Capture/Replay（CR）Approach，只需要一點程式技能就能產生測試案例。過去我們開發出了一套能夠獨立於待測系統（System Under Test，SUT）並且與測試平台無關的自動化測試系統KORAT，但是僅支援鍵盤訊號的錄製與播放，目前應用於非GUI的環境中。
為了增加KORAT測試的能力，在本篇論文我們探討KORAT支援滑鼠設備的方法，讓KORAT也能夠應用於GUI輸入的錄製與重播中。市面上已知的測試工具[1-3]，當使用到滑鼠操作時，通常會因為probe effect，讓錄製以及播放的滑鼠路徑不一致。甚至這些工具根本不提供滑鼠路徑的紀錄。因為KORAT直接錄製與播放滑鼠原始訊號的關係，巧妙的避開probe effect，所以能夠精確的錄製與播放滑鼠訊號。在本研究中整合了一套軟硬體環境，透過STM32F4 DISCOVERY微控制器工程模板，來接收、處理、模擬滑鼠原始訊號。並且我們利用進階的影像處理技術，開發出導航模式，在GUI元件可能變動的情況下，KORAT也能夠引導滑鼠到正確的事件觸發點上。另外，KORAT也設計出追蹤模式，保留滑鼠的原始訊號，精確的紀錄滑鼠的移動路徑。
經實驗結果證實，本研究設計出來的軟硬體環境，用來錄製與重播滑鼠原始訊號，的確能夠精確的紀錄、播放滑鼠的移動路徑。克服過往CR工具都普遍困擾的probe effect。並設計出導航機制以及追蹤模式，幫助測試者錄製以及播放更複雜的滑鼠動作，使之能夠應用於GUI環境的自動化測試上。
;In software development, regression testing is used for ensuring that the quality of new version software is not damaged by the changes. Since regression testing is repetitive, tedious, and time consuming, different approaches have been proposed to automate the task. One popular approach is the Capture/Replay (CR) technique which requires little programming skills to create the tests. In the past, we have developed a testing automation tool called KORAT. It is isolated from system under test（SUT）and allows testing the applications from different platforms. KORAT only supports keyboard device for capturing and replaying. It can be applied to non-GUI testing for now.
To extend KORAT′s testing capability, supporting mouse device is studied in this thesis. Existing CR tools often suffers from the probe effect problem, which causes imprecise mouse path in replay runs. Because KORAT captures and replay hardware mouse signal, the probe effect is kept to minimum. Precise mouse path replay can be achieved in a much more promising way. In this paper, we integrate the necessary software and hardware components for KORAT. KORAT can intercept and emulate mouse signals by a microcontroller unit（MCU）named STM32F4 DISCOVERY. And then, we implement a navigation mode using advanced image recognition technique to navigate the mouse by image analysis. It can help KORAT navigate the mouse to a correct event point, even if location of GUI components changed. KORAT also provides track mode which can keeps the mouse signals for recording the mouse path precisely.
Our experiments indicate that our design and implementation can record and replay mouse signals precisely. KORAT overcomes the probe effect problem which cannot be achieved by existing testing tools.
