Gender-based Detection and Tracking of Child Pedestrians using Machine Learning
Tyler B. Ward · 2024
As the prevalence of fully autonomous vehicles (AVs) on public roads grows in the coming decades, it becomes increasingly important to ensure the safe operation of these vehicles, especially in the presence of pedestrians. A robust perception system in particular is crucial to ensuring safe interactions of these vehicles with humans. It is a noted issue in AV research that these vehicles can sometimes struggle to accurately detect child pedestrians. While there have been advancements made in this area in recent years, there remains a gap in the literature surrounding considerations of a child pedestrian's gender in the detection process. Past research in the field of transportation has shown that the behavior of child pedestrians differs depending on the child's gender, making gender a potentially important consideration in the perception layer of an AV. This paper presents a gender-based pedestrian detection framework based on the YOLOv8 object detection method that is capable of detecting not only whether a child pedestrian is male or female, but also distinguishing them from an adult pedestrian. The system also employs the ByteTrack multiple object tracking technique to track the detected pedestrians through a scene. Performance metrics for this system are included, and the results demonstrate impressive detection and tracking capabilities.