| Summary: | In this paper, we use computer techniques to extract the features of each convolutional layer of CNN, analyze the feature variations of different depth convolutions, and propose a new network framework to improve the head pose estimation task performance by combining the task characteristics of head pose estimation. Multi-scale feature information fusion is the basis of the proposed head pose estimation method (IRHP-Net), which consists of a feature extraction module and a multi-scale feature information fusion module. In the smart classroom learning environment, the algorithm is used to identify students’ attention areas and construct the distraction index and threshold parameters to determine the inattentive state and provide relevant teaching measures. Smart classroom teaching resulted in a 39.69% increase in students’ attention, as shown in the results. Students in the traditional teaching mode showed a 16.7% lower level of learning engagement. A 17.24% increase in academic performance was also a result of the increase in attention and learning engagement.
|
|---|