Paper Title: "Video Classification Based on the Behaviors of Children in Pre-school Through Surveillance Cameras"
Authors:
Nguyen Tran Gia The - 20521940 – KHDL2020 – Lead Author.
Do Pham Phuc Tinh – 20522020 – KHDL2020 – Co-Author.
Cao Dinh Duy Ngoc - 20521661 – KHDL2020 – Co-Author.
Nguyen Huu Minh Tam – 20521871 – KHDL2020 – Co-Author.
Ngo Huynh Truong – 20522085 – KHDL2020 – Co-Author.
Supervisor:
Dr. Do Trong Hop
Abstract:
In preschool, children are active and curious about the world around them. Their tendency to engage in unusual or dangerous behaviors poses a significant risk to their safety. Constantly monitoring surveillance camera footage and analyzing it for abnormal behavior requires substantial attention and effort from human observers. Therefore, applying technology to monitor children's abnormal behavior in preschool settings is crucial. With technological advancements, classifying videos through surveillance cameras can be achieved. However, there is a scarcity of datasets for this task, especially with video data in preschool environments due to challenges in data collection and concerns regarding children's privacy and safety. In this paper, we propose the Behaviors of Children in Preschool (BCiPS) dataset to address this gap. BCiPS comprises 4268 videos ranging from 3 to 6 seconds in length. We evaluate various machine learning and deep learning models on BCiPS, with the CNN-LSTM model achieving the highest performance, surpassing 75%, in terms of accuracy, recall, precision, and F1_score. Additionally, we analyze cases where the models fail to identify abnormal behavior to understand the reasons behind these failures, identify dataset weaknesses, and enhance dataset accuracy, creating a reliable tool for constructing an effective warning system in real-life situations.
"We express our gratitude to Dr. Do Trong Hop for his guidance and support throughout our research and the publication of this international scientific paper."
The ICIT2023 conference is the International Conference on Intelligence of Things sponsored by Springer. This conference focuses on the latest technologies, applications, and research in the fields of the Internet of Things, Artificial Intelligence of Things (IoT2/AIoT), and related information technology areas. Accepted and presented papers at the conference will be published by Springer in the Lecture Notes on Data Engineering and Communications Technologies series, indexed by SCOPUS (Q3).
Detailed Information: https://www.facebook.com/UIT.Fanpage/posts/pfbid0haBJhmwbj6VrXuFyMvomvLegBQytUzkg5gmL5xmDJau4jnvPbPtTKKRz27BRyRmKl
Hai Bang - Communication Collaborator, University of Information Technology
English version: Phan Huy Hoang
