خودنظارت برای حضور فراگیرنده در یادگیری از راه‌دور با استفاده از تصدیق تصویر و شبکه‌های بیزی

Distance learning is one of the common education methods. Its advantage lies in that the student can learn at anytime or anyplace. However, such a learning mode relies highly on the dependence of the student. Under different environments and conditions, not all the students can be attentive. In this research, an auto-detection system has been designed, using image processing and recognition technique, for defining the facial expressions and behavior easily found when a learner is inattentive or in bad mentality under distance learning environment. From the learner’s facial expressions and behavior features, the attentiveness of the student during distance learning can be determined by Bayesian Networks Model. After implementing the system of this research, and performing practical test, it is found that the accuracy of identifying the features of bad mentality and inattentive behavior is high. From Bayesian Networks assessment and inference, the learning attentiveness of the student can be determined precisely to have the teacher control the learning condition of the student explicitly.

Distance learning has been developed rapidly in the past few years. Many information technologies are applied to the tools of distance education. Currently, the most common distance learning method is to learn with personal computer and computer network (Harasim, Hiltz, Teles, & Turoff, 1995). In an attempt to enhance the interaction between the teacher and student, and truly understand the student’s learning process and condition, generally, e-mail and discussion forum are used as interaction tools (Eisenstadt and Vincent, 1998 and Wolfe, 2000). The student can login to the website freely, and decide the time for learning online. As long as the schoolwork given by the teacher is completed, and the related reports are handed in or the evaluation is made, then, the student can earn the credit. Although the student can choose to learn at the time convenient for himself/herself, the time chosen might not be the time when the student can learn with the best mentality and greatest attention. According to the study, effective time management is the best learning strategy, and also a significant factor for academic achievement (Walter & Siebert, 1993). However, in distance learning, how can the teacher know whether the student is attentive, and whether the student has left the seat or not? If the student does not have an active attitude towards learning, it is quite difficult for the teacher to have the student obtain the learning achievement (Kimble, 1967). In an attempt to understand the student’s learning condition, in many researches, focus has been placed on computing the number of times in interacting in the discussion or the hours of participating in the course as the base for evaluating the student’s participation (Hwang and Yang, 2008 and Pena-Shaff et al., 2001). However, the number of times is unequal to the quality of discussion content. A student with a high participation frequency is not equivalent to a student who really participates attentively (Mason, 1997). As the technique for precisely analyzing the discussion content is not mature enough yet (Rourke, Anderson, Garrison, & Archer, 2001), in order to earn the credit hours, many distance learning students might just login online without sitting in front of the computer or participating attentively. However, by doing so, the student can still earn the credit hours. As this does not nurture the attitude of the student towards learning, it makes the credits of distance learning fail to be completely accepted. If the teacher can monitor the learner, and supervise the undesirable learning conditions, for instance, drowsy, distracted, or doing other things, then, the learning process of the learner can be controlled effectively. Image recognition technique, mostly applied to the fatigue detection of drivers, can detect the mentality by facial images (Hamada et al., 2002, Smith et al., 2004 and Ueno et al., 1994). Nevertheless, as the focus is mainly on fatigue features, 8and either no desirable comprehensive assessment is made or other special device is needed, it is not suitable to be applied to distance learning directly. Moreover, learning inattention involves not simply mentality, but also other inattentive behavior. However, by modifying and adding the inattentive behavior detection to such a technique, it can be applied to the distance education for detecting the students’ learning conditions, for instance, the increase of blinking frequency induced by fatigue, yawning, gazing around inattentively, starting other computer programs, or leaving the computer. As the inattentive conditions of the student in learning process is recorded and reported to the teacher, it can help the teacher to control the attentiveness of the student and urge the student to participate attentively for ensuring that the student of distance learning is in good learning conditions (Hwang & Yang, 2009). In this research, the existing defects of distance learning are explored from the perspective of learning attention. In order to help the teacher or parents take hold of the student’s learning to ascertain that the student is learning in good condition, and being attentive to the study while accumulating the participation hours, a mechanism is designed and implemented for determining the student’s learning condition. In this mechanism, no other device is required to be installed. Only a Webcam is needed for capturing the facial image in real time. Then, just detect the student’s mentality and attentiveness in participating the course by auto-detection with image recognition technique. Finally, with Bayesian Networks assessment, the probability of misjudgment can be reduced. The detection performed by the mechanism can help the teacher to control the student’s learning condition, and avoid the student to have a lazy attitude towards distance learning so as to guaranty the effects of distance education. Section 2 is focused on exploring the use of image recognition for identifying the expression features in detecting the possible learning condition of the learner, and on exploring the application of Bayesian Networks assessment. Section 3 is the description on the technique of the design and implementation of the full-time auto-detection mechanism. Section 4 is the practical evaluation of the accuracy of the features detection of image recognition and that of Bayesian Networks assessment. Finally, the conclusion and the directions of future researches are proposed.

In common distance learning, it is hardly possible for the teacher to take hold of the conditions of the student. The student must be self-disciplined to be attentive to the course. Aside from knowledge, learning attitude is also very important in the learning process. In the entire learning process, it is not easy for the teacher to know whether or not the student is lazy or leaving the seat for doing other things. This leads to the doubt about the quality of distance learning. In this research, Webcam was used for capturing the facial image in real time. Then, image recognition technique was applied to detect the student’s mentality and the attentiveness in using the computer via auto-detection. With Bayesian Networks assessment, it can help the teacher to control the student’s learning condition through the detection implemented by the mechanism. From the result of the experiment, it is found that the accuracy in detecting the facial features and feature behavior by Webcam for monitoring the learners is quite high. The Bayesian Networks assessment based on the occurrence probabilities can effectively determine the bad mentality and inattentive condition of the learner and explicitly record the current learning condition of the student so that the teacher can take hold of the student’s condition in real time. In this research, the accuracy of the algorithm in detecting feature behaviors is quite high. However, under certain environment conditions, for instance, the changes of illuminant, background complexity, and computer processing speed, the results might be affected to certain extent. In the future, if the system could adjust the image light and brightness automatically in accordance with the environment background, then, it can make the entire detection mechanism more humanized.