Real-Time Vehicle Passenger Detection Through Deep Learning

This study investigates the application of Convolutional Neural Networks (CNNs) in object detection and tracking, with specific focus on determining the number of passengers inside moving vehicles as recorded roadside video cameras. The objective is to develop an accurate and efficient method for counting the number of individuals in vehicles. We focus on the one-stage CNN-based model, You Only Look Once (YOLO) applied to two distinct data sets: static images and video streams. The performance of the model under different situations reveals that image quality factors such as lighting conditions, weather conditions and viewing angles play an important part in the task of identifying passengers inside vehicles. The model achieves a mean average precision of over 90% when including weather factors such as fog, rain, snow and spatter.