Surveillance in Daycare Centers Using 3D Depth Sensor

According to the United States Department of Health and Human Services, more than 3 million reports of child abuse are made in the nation, involving more than 6 million children. This is a huge issue not only in the U.S. but also all around the world, specifically in areas with a high children population such as daycare centers. While daycare centers are considered to be relatively safe, these environments present one of the biggest concerns against of safety of children. In this work, we focus on detecting child abuse that are portrayed by teachers in daycare centers. We use streaming 3D skeleton joint coordinates obtained from Kinect sensors to (1) classify a teacher from a group of children and (2) detect physically abusive behaviors such as such as hitting, kicking, slapping, shaking, and pushing that are performed by the adult. Specifically for age group classification, we analyze the histograms of training samples and implement a bin-based classification method that represents bin-boundaries. For abusive behavior detection, we use a combination of an array of supervised learners to recognize a predefined set of abusive actions. By designing such a system, we aim at opening the possibilities for developing applications that contribute to preserving the safety of children in daycare centers.