Practical Automated Video Analytics for Crowd Monitoring and Counting

Kang Hao Cheong,U. Rajendra Acharya,Jin Ming Koh,Joel Weijia Lai,Simon Ching Man Yu,Sandra Poeschmann,Kenneth Jian Wei Tang

doi:10.1109/access.2019.2958255

Abstract

Video surveillance is gaining popularity in numerous applications, including facility management, traffic monitoring, crowd analysis, and urban security. Despite the increasing demand for closed-circuit television (CCTV) and related infrastructure in public spaces, there remains a notable lack of readily-deployable automated surveillance systems. In this study, we present a low-cost and efficient approach that integrates the use of computational object recognition to perform fully-automated identification, tracking, and counting of human traffic on camera video streams. Two software implementations are explored and the performance of these schemes is compared. Validation against controlled and non-controlled real-world environments is also demonstrated. The implementation provides automated video analytics for medium crowd density monitoring and tracking, eliminating labor-intensive tasks traditionally requiring human operation, with results indicating great reliability in real-life scenarios.

Highlights

Video surveillance is an integral component of modern urban security, and when coupled with computational analytics, can have greatly expanded functionality including facial recognition, motion detection, traffic and crowd monitoring, and automated hazard alarms [1]–[7]
The controlled studies were conducted with steady artificial lighting and primarily constant environmental parameters; each test lasted a duration of 150 seconds with both the background subtraction (BGS) and single shot detector (SSD) methods, and a manual on-site count was performed simultaneously to match the results against
These results indicate satisfactory counting accuracy for both BGS and SSD methods, with BGS notably achieving perfect accuracy in the idealized single-subject scenario, but is outperformed by SSD in more realistic multiple-subject scenarios

Summary

Introduction

Video surveillance is an integral component of modern urban security, and when coupled with computational analytics, can have greatly expanded functionality including facial recognition, motion detection, traffic and crowd monitoring, and automated hazard alarms [1]–[7]. Amidst the progressing state-of-the-art, integration of automated analytics in commercial video surveillance for crowd monitoring and counting is an area that can be further explored [19]; and there is at present limited literature on demonstrated effective low-cost systems for deployment. Utilizing computer vision and real-time automated analytics in replacement of manual labour reduces operational costs and eliminates human errors and lapses [22]–[24] —we seek to develop a viable deployment-ready implementation in this study. The operation of such a method relies on a known background frame with no present objects. This background reference is subtracted from each frame of the video footage, or subset of frames to reduce computational cost, yielding frames containing only foreground objects. BGS-based methods are presently applied in commercial video surveillance systems for malls and public spaces

Methods

Results

Conclusion