應用眼動儀於二維流場可視化方法的使用者研究

Abstract

流場可視化方法被廣泛地應用在許多領域裡面，好的視覺化方法可以幫助人們快速且正確的理解流場的內容，因此如何使流場的結構與特徵更容易讓人理解便是一個很重要的課題。過去相關的流場可視化評測使用者研究多以分析根據作答數據來比較不同方法的效率，本論文則提出了一個以眼動儀資料作為基礎的使用者研究方法，從影像對視覺的影響的方向來評測不同流場可視化方法效率。眼動儀實驗的結果讓我們觀察到一些過去的評測方式無法比較到的地方，這讓我們可以從更多面向瞭解人們在觀看流場影像時候的行為，以及不同流場可視化方法在不同目標下的效率。我們一共對 5 種二維流場可視化方法進行比較，這 5 種流場可視化方法包含了 1 種箭頭形式的方法，2 種流線方法以及 2 種材質紋理方法。從流體專家及以前使用者研究所注重的評估項目，我們知道在流場顯示中臨界點顯示與軌跡預測的重要性，為了評估這兩方面的重要性，同時我們想量測在沒有特別指示之下自然注視的反應，所以我們設計了以下 4 個實驗：(1) 自由注視：我們分析在沒有特定指示下，受測者自由注視流場影像時眼睛凝視位置與流場速度、臨界點位置的關係，同時提出一個預測眼睛注視流場區域的預測模型；(2) 流動路徑預測：由受測者預測流動軌跡，我們比較不同方法之下眼睛軌跡與流場流動軌跡的相似性；(3) 找尋流場特徵：由受測者找尋流場中的特徵，我們根據眼睛凝視點的分布狀態比較不同方法之下流場的特徵是否容易吸引受測者的注意；(4) 分辨流場特徵：由受測者在多個流場中的特徵中分辨出正確的目標，我們分析受測者在找尋正確流場特徵的過程中，在不同待選目標中來回的次數以及目標被找到的順序。除了眼動儀的資料外，我們也對一般作答的結果進行比較。我們將實驗的數據以 2 種分群方式進行分析，第一種是受測者有無流體力學背景，第二種是 5 種流場可視化方法。我們的研究顯示在觀看流場影像的時候，無論是哪種流場可視化方法，人們均會傾向注意到流向變化劇烈的區域。有專業背景的受測者在進行實驗的反應會較快且較注意臨界點，但在判斷的正確性與精準度上面與沒有流體力學背景的受測者沒有差異。而在不同的流場可視化方法之中，材質合成 (texture-base) 的方法在作答的正確性反應速度表現非常優秀，但是在特徵的部分較不容易吸引到受測者的注意。

The flow visualization methods are widely applied in many areas. A good flow visualization method would help people understand the flow quickly and accu-rately, so how to present the structure and featurs of a flow field in a much clearer and more understandable way is a very important issue for evaluating a flow vi-sualiztion method. Previous user studies on flow visualization methods compared the effectiveness of different methods by analyzing the users’ response to ques-tions. In this thesis, we propose an eye-movement-based user study methodology to analyze and evaluate flow visualization methods. We observe some phenom-ena that were not found in the previous questionnaire-based evaluation studies on flow visualization methods. Our experimential results help us understand human subjects’ visual behavior when they view the flow images, and the presentation efficiency of different flow visualization methods. We compared five 2D flow visu-alization methods in our experiments, including a direct flow visualization method, two geometric flow visualization methods and two texture-based flow visualization methods. According to fluid dynamics experts’ suggestions and previous flow visu-alization evaluation studies, display of critical points and flow trajectory prediction are two important functions of flow visualization. In order to evaluate these two aspects, we designed the following four tasks: (1) Free-viewing: we analyze the correlation of fixations with flow velocity, and the fixations with location of crit-ical points while the user is not performing any tasks. We also develop a model to predict the location of the eye gaze under free-viewing. (2) Advection predict-ing: compare subject’s advection predicting performance on different visualization methods. (3) Flow features locating: compare subject’s flow feature finding perfor-mance on different visualization methods. (4) Flow features identifying: compare subject’s flow feature identification performance on different visualization meth-ods. Besides the eye-gaze data, we also compared the result of response time and the location that mouse clicked. Our experiments show that human subjects tend to be attracted by the region where the flow field has large directional variations on all five flow visualization methods. On the other hand, experts attend to the criti-cal points more than non-experts, and the response time of experts is shorter than non-experts, while there are no significant differences in accuracy and correctness between experts and non-experts. Among all the methods used in our experiment, texture-based methods perform well at the correctness and the response time, but their presentation of flow features cannot draw subjects’ attention well.