BUILDING A MODEL FOR DETECTING VIOLENT BEHAVIOR IN CHILDREN

Abstract

Child abuse is a grave and pervasive social problem with profound consequences for both individual victims and society as a whole. However, there are not many models or application that available in order to detect children violence. Proper detection is important not only in recognizing children abuse but also apply appropriate penalties for those who perform violence on children. Efforts to enhance child abuse detection are crucial in addressing this sensitive issue and ensuring the safety and well-being of vulnerable children. As technology and interdisciplinary collaboration continue to evolve, I spend my endeavor to develop a machine learning model which help detecting children violence. In my project, I use machine learning to provide detailed analysis and detect whether there is any children violence is performed in videos. In addition, Deep Learning and Computer Vision are being intensively researched and improved every day. In particular, Google's development of MediaPipe, an opensource framework for building world-class machine learning solutions that provide basic machine learning models for common tasks such as hand tracking, posture recognition, ... Another famous framework that I used in this thesis is YOLOv8 which is well-known for its efficiency in object detection, image segmentation, … This project, "Children abuse detection in videos based on machine learning", is based on the detection of postures by MediaPipe and YOLOv8 which are used to analyze, detect and classify actions of violence. The final experimental results show that the algorithm proposed in this work can effectively identify which actions are performed in videos and in realtime.