Using Artificial Intelligence for Obtaining Vehicle Occupancy Using Security Cameras

Controlling vehicle occupancy is very useful for detecting overcrowding situations in public transportation. In a pandemic, it becomes essential to prevent disease spread by dynamically adapting the route plan and traffic scheduling and allowing passengers to choose vehicles that carry fewer passengers in real-time. However, given the size and complexity of the system, this task is not trivial. In this paper, a classification method for passenger occupancy analysis is proposed. The classifier is based on a Convolutional Neural Network (CNN) model. As input to the proposed model, images from Closed-Circuit Television (CCTV) cameras, already present in the vehicles for security reasons, were used. The addressed problem is especially challenging due to the non-standardization of the equipment, poor quality of the images, and cameras’ variable positioning inside the vehicle. Besides, as there is no pre-existing image dataset for training, one must deal with a small number of labeled images. Two datasets are created, varying on the assessed quality score. Three CNN architectures were evaluated on the datasets, to validate the feasibility of the proposed method. Experiments on the created public transportation dataset show that the proposal of this work achieves promising results.