Research On Learning States Based On The Facial Features

With the rapid development of the electronic information technology, the online learning system came into being. The emergence of the online learning system fully meets people’s learning needs wherever, whatever. However, there are also some disadvantages in the development of this system compared with traditional learning mode. Mostly the system evaluates the effect by the learning time and it lacks the process monitor and status feedback, which leads to bad learning effect. Therefore, the paper puts forward a method which makes use of the camera to capture the real-time video streaming, analysis learners’fatigue, attention and learning state based on the image from the video and reminds the learner when it seems to be fatigue, inattention as well as non-learning case by audio and dialog to achieve response. The paper’s work includes:(1) Put forward a new state identification method which consider the learner’attention, fatigue as well as the learning state judged by the criteria of concentration, fatigue, safety and reminding time, then describe the algorithm by the finite state automata.Finally verify the feasibility the method theoretically.(2) Put forward five kinds of face detection methods. The paper first introduces the existing face detection methods and decides to use the AdaBoost algorithm in accordance with the requirements of real-time and accurate detection. However the AdaBoost algorithm can’t detect the face which the rotation angle is too large, so the face detection method based on the skin color is considered. According to the characteristics of the two methods, the paper proposes five kinds of face detection methods and compares the detection time and false detection rate on the MobBIO database. Finally one face detection method is selected.(3) Study first and detect again to achieve the face features of the specific person, then recognize the face based on the features and effectively reduce the detection time. When various parameters are selected to measure the learner’s state, the same standard for different people can’t fully describe the learner’s state and even occurs some errors. So the paper proposes to learn first from the streams then recognizes the face and fixes the parameters at the same time.(4) Put forward the method of multi-parameter to analysis the learner’s state. In order to analyze the learner’s state preferably, the paper uses multiple parameters instead of one to improve the accuracy of the analysis results.The experimental results show that this method can be accurately recognize the learning state relatively.