Examinee Behavior Description Method Based on Image Captioning
Miao Ma, Ziang Gao · 2019
Aiming at the problem of missing report of abnormal situation caused by manual observation of examination room monitoring video, this paper introduces image description in deep learning to construct a data set “Examinee Activity Captioning Dataset” suitable for examination scene, and proposes a method to describe examinee's behavior based on CNN+LSTM. This method employs a convolutional neural network to automatically extract the characteristics of examinee's behavior, and further generates sentences to describe examinee's behavior through long short-term memory network. The experimental results show that the proposed method can accurately describe five kinds of behaviors, including writing test papers, turning over test papers and looking around, which may play an important role in future smart examination room.