BK-Student Activity Dataset (BK-SAD)

Acknowledgements

We express our gratitude to the Ministry of Science and Technology for providing funding for the research conducted under the project number KC-4.0-06/19-25. The BK-SAD dataset is a research outcome of the project.

Introduction

Skeleton-based human action recognition has emerged as a prominent research topic in the field of artificial intelligence due to its broad applicability in a range of domains, including but not limited to healthcare, security and surveillance, entertainment, and intelligent environments. We propose a novel data collection methodology and present BK-Student Activity Dataset (BK-SAD), a new 2D dataset for HAR in smart classrooms. Our dataset contains three classes: hand raising, doze off, and normal activities. The dataset was collected using cameras placed in the real classroom environments and consisted of video data from multiple viewpoints. The dataset contains over 48,000 labeled samples of students raising their hands, over 34,000 labeled samples of students dozing off during class, and over 170,000 labeled samples of normal activities.

Human Activities in the Dataset

Our BK-SAD dataset contains three different activities: (1) hand-raising, (2) doze off, (3) normal

Example images from the 3 activities:

Dataset Description

The dataset consists of a total of 255,159 frames captured from three cameras placed in three different directions of a classroom. Each frame is labeled as one of the three activites: raise hand, doze, or normal. The dataset contains 48,457 raising hand samples, 34,762 dozing samples, and 171,940 normal samples.

The dataset includes the keypoint-labeled image of each student cut from the video and the text file that stores the keypoint coordinates. This allows researchers to perform more detailed analysis and develop more precise algorithms for human action recognition.

The raising hand label indicates that the person in the frame is raising their hand, which could mean they have a question, want to answer a question, or need to use the restroom. The doze label indicates that the person in the frame is dozing off, which could mean they are tired, bored, or not engaged in the class. The normal label indicates that the person in the frame is sitting upright and paying attention to the class.

This dataset provides a valuable resource for researchers and machine learning practitioners interested in developing and testing algorithms related to human action recognition. The large number of frames captured from multiple angles allows for a more comprehensive understanding of the activities taking place within the classroom, making this dataset a valuable tool for developing technologies that can improve student engagement and performance.

Download

The dataset files can be downloaded here

Citation