Crowd anomaly estimation and detection: A review

Abstract

Abnormal crowd detection and estimation are critical in video surveillance for ensuring public safety and preventing stampedes. Owing to occlusions and blind spots, traditional video surveillance methods cannot detect, estimate, or locate people in dense moving crowds with acceptable accuracy, posing a major challenge. Therefore, this study aims to provide an in-depth analysis of the most recent advancements in recognizing abnormal behaviors in large crowds. We present a comprehensive literature review on crowd anomaly detection using disruptive technologies such as radio frequency identification, wireless sensor networks, Wi-Fi, and Bluetooth low energy, employing device-free noninvasive algorithms based on received signal strength indicator variations to detect the speed and direction of a moving crowd to predict the onset of a stampede. Furthermore, this study presents the most recent findings on mobile crowdsensing based on edge computing, urban dynamics, optical flow, and machine learning techniques. Finally, we critically analyze the major challenges, shedding light on opportunities and directions for future work.