A Robust, Low-Complexity Real-Time Vehicle Counting System For Automated Traffic Surveillance

Abstract

This paper presents a real-time video based vehicle counting system, the key feature of which is its low computational complexity. The counting algorithm is tailored to be able to run on a low cost hardware platform like Raspberry Pi as part of a smart camera system. Background subtraction forms the basis of day time vehicle counting while night time counting is based on headlight detection and pairing. Both operations are performed only on a pre-specified virtual detection region within each frame. When tested on public traffic datasets, the method outperforms other more complex algorithms under various weather and traffic conditions.