Our Purpose
The purpose of this study is to help make the process for humans of assessing instructional videos less time-consuming and more efficient. Our research design concerns three aspects of using computer vision, machine learning, and deep learning: (a) the type of neural network, (b) the type of video label (i.e., object labels and activity labels), and (c) the subject of instruction. A few types of neural networks have recently proven effective for video classification: convolutional neural networks (CNNs), long short-term memory (LSTM) neural networks, and hybrid CNN-LSTMs. We are using classification accuracy to evaluate the efficacy of each approach while taking account of practical considerations such as the computational burden of each method as well as the training set required. Our design will enable us to address several research questions:
- How accurate is each type of neural network at classifying objects in classroom video?
- How accurate is each type of neural network at classifying instructional activities in classroom video?
- Does the subject of instruction affect classification accuracy?
- What are the practical limitations of using neural networks for video classification?
"Little is known about the feasibility of having a computer discern more complex activities such as representations of content, teacher questioning, modeling of learning strategies, and student engagement. This study is designed to train neural networks to identify such complex activities in elementary videos."