- 學位論文
機器視覺問卷輸入影像系統開發
Development of the Automatic Questionnaire Input System by Machine Vision
摘要
對許多研究安全衛生的學者來說，透過問卷調查來做定量訊息的收集是相當重要的。研究者通常都希望問卷分析的資料能有較高的準確性和代表性，因此在大型的問卷調查裡數據的收集與管理會相當龐大，也使得問卷輸入成為分析過程裡的重要關卡。問卷輸入可採用人工輸入或利用讀卡機讀取後傳輸至電腦來進行分析，但人工作業需考量問卷版面設計及人力成本，而讀卡機輸入則需使用專用卡紙，事後必須承租讀卡機來進行讀卡，成本相對地也會增加。因此本研究開發一套利用機器視覺進行問卷自動輸入的系統，使得問卷版面設計及紙張材質都無須受限，研究者能自行大量地自動化輸入問卷。 本論文問卷辨識系統開發多項軟硬體模組，當系統運作時，問卷經由送紙裝置送出，透過攝像鏡頭擷取回收問卷的影像，利用圖控式程式語言，手動初定位填答空格，隨後進行即時影像分析處理，並將分析後數據儲存成檔案。本論文進行若干實驗以評估此系統之準確性及效能。在性能實驗中針對填答用筆的顏色及劃記形狀進行準確率測試，經統計分析顯示黑、紅、藍、綠、紫及鉛筆等多種大眾常用顏色之辨識率並無顯著差異；而打勾、劃斜線、劃叉之辨識率顯著高於劃圈。整體而言，各顏色之辨識率均高達98.77%~99.17%，而各形狀之辨識率亦達95.32%~99.21%。此外分別對空白及已填答問卷進行辨識速度測試，其辨識平均時間分別為3119.78ms和3154.85ms，即在3.2秒內軟體可以完成辨識一張填畢問卷。因此本研究的自動化問卷輸入系統確可提高問卷調查辨識的準確性及效率。
並列摘要
For many scholars in the areas of safety and health, it is very important to collect the quantitative information through questionnaire surveys. In a large-scale survey, hoping that the data of the questionnaire analysis can be more accurate and representative, researchers collect and manage a huge amount of questionnaires, which makes the data input become one of the most important steps in the analytical process. Data input of questionnaires can be done manually or through the card reading devices for further analysis in computers. For manual input people need to consider the questionnaire layouts and the labor cost, while using the card reading devices requires much higher cost since those devices are rental and the questionnaires are made of some specified materials. This research are, therefore, to develop by machine vision an automatic input system, which let researchers be able to conduct a large amount of questionnaire input without the limitations of the layout designs and paper materials. The questionnaire input system developed in this thesis integrates a number of hardware and software modules. When the system is in operation, each questionnaire sent by a paper feeder is captured by a camera in real time, and analyzed in the graphical user interface. The data acquired after image processing are then saved as a text file. Several experiments were conducted to assess the accuracy and the efficiency of the system. In the performance experiment, we processed a plenty of questionnaires answered using various pens with the specified colors, as well as some specified marks. It was found by statistical analyses that there was no significant difference in the accuracies of all colors; and accuracies answered with checks, ticks and crosses were significantly better than those answered with circles. Of all the accuracies of the software to recognize the checkboxes filled with checks in various colors ranged from 98.77%~99.17%; and to recognize other checkboxes answered with different marks were around 95.32%~99.21%. In the last experiment, we identified that the average processing time for a blank questionnaire was 3119.78ms, and for an answered one was 3154.85ms, which means that the developed software is capable of processing a questionnaire during 3.2 seconds. It is concluded that this automatic questionnaire input system is indeed able to increase the accuracy and efficiency in questionnaire surveys.