Article Preview
TopIntroduction
Artificial intelligence is a discipline that enables computers to simulate human thought processes or intelligent behaviors. In recent years, artificial intelligence technology has been widely used in the field of education (Holmes et al. 2019; Verma 2018). It has become a new trend to construct an education system by using intelligent learning and interactive learning.
Classroom behavior analysis (Twyman & Heward 2018) aims to study the internal mechanism of teachers' teaching activities and students' learning and development in the classroom. It can help teachers and students to rethink their classroom behaviors and promote the quality of classroom teaching (Owens 2018). Most of the traditional classroom teaching behavior analysis methods are based on self-evaluation, manual supervision, manual coding et al. These methods are time-consuming, laborious and have the disadvantages of strong coding subjectivity, small sample size. They lack interpretability and scalability. With the help of intelligent technology, we can comprehensively collect and analyze the classroom data to identify classroom behavior and gain insight into the teaching and student learning status in time (Chassignol et al. 2018). Thus, the teaching quality can be strongly enhanced.
Compared with other disciplines studying in the classroom, physical education classroom teaching behavior has its particularity characteristics (Morgan & Hansen 2008). Therefore, the physical education classroom teaching behavior observation system is important to improve the observation and communication effect of physical education classroom teaching and improve the quality of physical education. The physical education classroom teaching behavior observation system also plays an important role in promoting the development of physical education teacher’s professional skills (Richards et al. 2018). The previous research paid more attention to teacher’s behavior and less attention to student’s learning behavior. The attention to the behavior and interaction between teachers and students is insufficient. The teaching and learning behaviors are separately researched. The classification of physical education teaching behavior is not systematic. For complex, too detailed and too cumbersome behavior coding, it also affects the analysis of front-line teachers in teaching and research activities, on-site observation records and analysis.
The early researches on human behavior analysis typically adopted 2D video as the source of perceptual data. However, 2D video can only provide limited information (Hsueh et al. 2020). Even there is no occlusion, the teaching behavior is still difficult to recognize with high accuracy. In this paper, we adopts Microsoft Kinect which utilizes RGB-D sensors to capture RGB video and depth video to analyze and identify teaching and learning behavior in physical classroom. The proposed behavior analysis method can utilize the symbiotic relationship and geometric constraints between human posture and interacting objects to identify the teaching and learning behaviors in physical classroom. First, the human body video sequence is captured by using Microsoft Kinect camera. Second, the video sequence is converted as structural representation by using recursive hierarchical conditional random fields. Third, the structural features of video sequence are used to learn a structural support vector machine which is used for structural data classification. Lastly, the behavior in future video sequence is identified by the trained structural classification model. The proposed behavior recognition framework can be further used in physical education analysis. The structure of behavior recognition induced physical education system is shown in the following figure.
Figure 1.
The architecture of behavior recognition for physical education
In Figure 1, the behavior video is input into a recognition model. The recognition result is finally output in a screen. The main contributions of this paper are summarized as the following three bullets.