Real-Time, Embedded Scene Invariant Crowd Counting Using Scale-Normalized Histogram of Moving Gradients (HoMG)

Parthipan Siva, Mohammad Javad Shafiee, Michael Jamieson, Alexander Wong

Abstract: Automated crowd counting has garnered significant interest for video surveillance. This paper proposes a novel scene invariant crowd counting algorithm designed for high accuracy yet low computational complexity in order to facilitate widespread use in real-time embedded video analytics systems. A novel low-complexity, scale-normalized feature called Histogram of Moving Gradients (HoMG) is introduced for highly effective spatiotemporal representation of crowds within a video. Real-time crowd region detection is achieved via boosted cascade of weak classifiers based on HoMG features. Based on the detected crowd regions, linear support vector regression (SVR) of crowd-region HoMG features is introduced for real-time crowd counting. Experimental results using a multi-scene crowd dataset show that the proposed algorithm outperforms state-of-the-art crowd counting algorithms while embedded on modern surveillance cameras. Thus demonstrating the efficacy of the proposed method for accurate, real-time, embedded crowd analysis.