Passenger Counting With Face Detection

Teoh, Han Wei (2020) Passenger Counting With Face Detection. Final Year Project, UTAR.

Preview

PDF Download (4Mb) | Preview

Abstract

Passenger counting exhibits a wide variety of applications in the context of smart cities. Such applications range from retail analytics, queue management and space utilizations. Driven by the success of machine learning, this study aims to develop a real-time passenger counting system. Different from existing works, the designed solution is deployed on a resource limited Intel UP Squared (UP2 ) Board, inference of which is handled by an accelerator called Intel Movidius Neural Compute Stick 2 (NCS2). MobileNet-single shot detector (SSD) is chosen as the object detector model since it belongs to a class of efficient models that can execute on mobile and embedded systems. While running detections across every video frames, centroid tracking algorithm tracks every unique person in video streams, where Kalman filter is further applied to reduce the noise. The outcome can be visualized on different type of devices for further analysis through a central cloud server. The performance of the passenger counting system is evaluated in terms of accuracy and frame per second (FPS). Furthermore, the feasibility of the solution is demonstrated in several showcases.

Actions (login required)