HOG based Infant Anti-Abduction Surveillance System for Maternity Hospitals

Child abduction is a pervasive concern that poses a serious threat to the safety and well-being of infantsworldwide. In response to this pressing issue, this researchpaper introduces an innovative Infant Anti-Abduction Surveillance System designed to enhance infant security through the implementation of advanced image and video processing technologies. The primary objective of the system is to provide hospitals and caregivers with a reliable and efficient tool for preventing and responding to potential abduction scenarios in real-time. The proposed system is a start to be able to capture activity in hospitals that may suggest abduction or a threat to infants. The system considers three step classification to capture suspects. It uses crowd detection and speed detection as factors to determine whether the person with the baby is a suspect or not. For these factors an image and video processing pipeline was incorporated, which used various machine learning algorithms namely Random-forest, decision trees, Naive Bayes, and KNN for classifying "with baby" and "without baby", showing a high accuracy of 85%; a deep learning LSTM model for action recognition that showed accuracy reaching approximately 78% in 100epochs was employed, lastly, crowd detection was achieved via the use of HOG and SVM. Through these detection techniques, the security concerns for families can be significantly reduced.