A Fast People Counting Method Based on Optical Flow

Abstract

This paper presents a fast people counting algorithm based on the classification of optical flow features. Conventional counting methods often use spatial features which are sensitive to background and illumination. Besides, in order to determine whether a person is entering or exiting a specific area, tracking algorithms are utilized and this brings additional computational burden. In our proposed method, optical flow vectors extracted from a predetermined area are used to count the entering and the exiting. The system requires neither detecting nor tracking the individuals and thus computationally efficient. The performance of the proposed method is tested in a dataset collected from a university campus building, and we find that it achieves %96 accuracy.