CHILD VIDEO DATASET TOOL TO DEVELOP OBJECT TRACKING SIMULATES BABYSITTER VISION ROBOT

This study presents a Child Video Dataset (CVDS) that has numerous videos of different ages and situation of children. To simulate a babysitter’s vision, our application was developed to track objects in a scene with the main goal of creating a reliable and operative moving child-object detection system. The aim of this study is to explore novel algorithms to track a child-object in an indoor and outdoor background video. It focuses on tracking a whole child-object while simultaneously tracking the body parts of that object to produce a positive system. This effort suggests an approach for labeling three body sections, i.e., the head, upper and lower sections and then for detecting a specific area within the three sections and tracking this section using a Gaussian Mixture Model (GMM) algorithm according to the labeling technique. The system is applied in three situations: Child-object walking, crawling and seated moving. During system experimentation, walking object tracking provided the best performance, achieving 91.932% for body-part tracking and 96.235% for whole-object tracking. Crawling object tracking achieved 90.832% for body-part tracking and 96.231% for whole object tracking. Finally, seated-moving-object tracking achieved 89.7% for body-part tracking and 93.4% for whole-object tracking.