深度學習於學生專注度分析之應用
摘要
隨著科技日新月異的進步,學生在上課時往往會使用筆記型電腦和手機做筆記或查資料,不過在這種情境下,自制力差的學生容易會被電子產品影響而導致分心,而專注力差的學生會有心不在焉或東張西望的情形發生。因此本論文的目的為希望能透過彩色影像分析學生的專注度,幫助老師了解學生們的學習狀況。本論文使用人臉偵測和類神經網路找到人臉特徵,判斷臉部資訊和估測人臉朝向的方向,也從姿態評估系統的骨架資料中擷取特徵,透過類神經網路和物品辨識判斷出目前的姿態,根據以上的結果分析學生的專注度。 本論文在臉部資訊上設計了兩種疲勞行為,以人臉特徵判定行為的發生,也擷取特徵用以訓練類神經網路預測人臉朝向角,平均角度誤差在10度以內;在動作辨識上設計了八種學生常見的姿態,其中四種姿態為使用物品的情境,並以此蒐集資料集訓練和測試類神經網路,在不同人間和角度推廣性測試的辨識率將近八成,在實際情境測試下也有不錯的辨識率,證明本系統在臉部資訊和動作辨識上能提供準確的資訊。 關鍵字:學生專注度、電腦視覺、行為識別、類神經網路

關鍵字: Kinect、視覺障礙者、卷積神經網路、物品偵測

 

 

Applications of Deep Learning in Student Concentration Analysis
Abstract
With the ever-improving of the technology, students usually use the laptop and cellphone to take notes or look up information. However, in this situation, students with poor self-control are easily distracted by electronic products. Also, students with poor concentration can be absent-minded and wandering. Therefore, the purpose of this paper is to analyze students’ concentration through color images and help the teacher understand students’ situations. In this paper, we use the face detector and neural network to find facial landmarks in order to determine the facial information and estimate the face orientation. In addition, we also use the neural network and object recognition to classify these skeleton features extracted from the skeleton data of the pose estimation system. Finally, we analyze the students’ concentration according to the above results. In this paper, two kinds of fatigue behaviors are designed on the facial information and the occurrence of fatigue behaviors is determined by facial landmarks. Also, we use the features extracted from facial landmarks to train the neural network to estimate the face orientation angles with an average angular error of less than 10 degrees. In motion recognition, we design eight kinds of common postures of students, four of which are human-object interaction, and to collect the dataset to train and test the neural network. In the experiments, the recognition rate of person and angle independent experiments is nearly 80%. Also, the recognition rate of the real situation is good. With these experiments, it is proved that the system can provide accurate information in facial information and motion recognition. Keywords: students’ attention, computer vision, activity recognition, neural networks