A Deep Learning Approach to Count people Using Facenet Architecture

This paper proposes a unique methodology to count people using a vision based approach. Facenet, a deep learning architecture is the backbone of this system. We have designed a simple yet unique algorithm to count people and store the pictures of individuals in a database for later investigation if needed. Our vision based people counter rely on face detection or pedestrian detection. These types of algorithms do not serve the need while using it for security. Keeping that in mind we have designed our system in a manner which can be used for security purposes. Our system is also able to identify the number of people present in a particular room and how many have left that room in a particular time. The system not only counts people but also tracks each individual, when he/she enters and leaves. For this paper we have tested four different face detection methodologies and chose the best one. The whole system can be implemented in any computer. We have tested the system with two cameras. One is for entrance, another is for exit. SQLite is used here for storing and managing data throughout the system. In this paper we able to achieve about 88% accuracy with the ability of storing each individual in the database.