Abstract Infrared hand gesture images can capture the human accurate gesture information in the weak illumination environment. In this article, we propose a novel hand gesture recognition model with convolutional neural networks for the human behavior analysis in “double teachers” classroom instruction and learning scenario. The recognized instructors’ hand gestures can be utilized to analyze teachers’ non-verbal behaviors which attract learners’ attention as well as enhance their learning outcomes. To extract the features of the infrared hand gesture images, a non-linear neural networks is constructed with four convolution layers. It aims to learn teachers’ pointing gestures and provides with the accurate gestures information. To achieve robust hand gesture recognition, the supervised convolutional neural network is constructed with three convolution layers to extract the infrared hand images features. Experiment results demonstrate that the proposed method outperforms the state-of-the-art methods in the term of the recognition accuracy and classification error. From the comparing experiments, the results prove that the developed approach can estimate the hand gesture with more than 92% recognition ratio.

中文翻译：

---

双师教学模式课堂中卷积神经网络红外手势识别

摘要 红外手势图像可以在弱光照环境下捕捉人体准确的手势信息。在本文中，我们提出了一种具有卷积神经网络的新型手势识别模型，用于“双师”课堂教学和学习场景中的人类行为分析。经认可的教师手势可用于分析教师的非语言行为，从而吸引学习者的注意力并提高他们的学习成果。为了提取红外手势图像的特征，构建了一个具有四个卷积层的非线性神经网络。它旨在学习教师的指点手势，并提供准确的手势信息。为了实现稳健的手势识别，监督卷积神经网络由三个卷积层构成，用于提取红外手部图像特征。实验结果表明，所提出的方法在识别准确率和分类误差方面优于最先进的方法。从对比实验中，结果证明所开发的方法可以以超过 92% 的识别率估计手势。