Enhancing Security of Smart Cities with “Signal for Help” Recognition System

The COVID-19 pandemic brought an alarming surge in violence against women and children, referred to as The Shadow Pandemic. In response, a Canadian foundation introduced the "Signal for Help" gesture, a discreet way for individuals in danger to alert others. However, the success of this gesture hinges on its recognition and appropriate reaction by bystanders. This paper introduces an innovative real-time system designed to detect these silent pleas for help within surveillance footage. The system integrates three key components: a person tracking mechanism utilizing YOLOv7 and Deep SORT to identify and follow individuals in videos; a hand feature extraction module based on MediaPipe to capture hand-related data; and a machine learning classification model to discern the presence of a help request. Our proposed model and pipeline architecture deliver real-time inference speeds without compromising on prediction accuracy, offering a potent tool to enhance safety in smart cities.