Citation

Yogameena B, Nagananthini C. Int. J. Disaster Risk Reduct. 2017; 22: 95-129.

Copyright

(Copyright © 2017, Elsevier Publishing)

DOI

10.1016/j.ijdrr.2017.02.021

PMID

unavailable

Abstract

Computer Vision (CV) based surveillance in crowded scenes is one of the most significant and promising topics due to the increase of people gatherings in public places where the possibilities of disasters and stampedes is also high. Therefore, it is essential to develop a well established Crowd Disaster Avoidance System (CDAS) for the public safety. In general, CV-CDAS focuses on crowd scene analysis, crowd behavior analysis and crowd management. Either, a stampede may occur due to abnormal behavior of individuals in a crowd or a crowd may behave abnormally after the stampede. Even though behavior analysis of a crowd plays an essential role, it is inadequate to decide whether a stampede may occur or not at that instance. Despite significant progress and various surveys on crowd scene, crowd modeling and behavior analysis over the past few decades, the existing surveys still fail to discuss and delineate the crowd disaster analysis based on many key aspects as a whole. However, CV-CDAS needs to be developed based on the necessary factors such as the number of cameras (single/multiple), target of interest (an individual/group or crowd), crowd scene (structured/unstructured), crowd motion (static/dynamic), crowd behavior analysis, people count and crowd density estimation, person re-identification in crowd, crowd evacuation, forensic analysis on crowd disaster and computations on crowd analysis. Consequently, taking all these factors into an account, this survey aims to deliberate a road map to develop a stable CV-CDAS which includes recent trends and approaches in the crowd disaster analysis. In addition, behavior analysis plays a vital role and hence, an extended version of existing survey with the perspective of single and multiple cameras is elaborated. Besides, it summarizes the crowd disaster problem handled by various CV algorithms in the past and also paves a way to avoid this problem in future. Moreover, this survey presents the existing benchmark datasets with their specifications and performance evaluation metrics with special focus on dataset which will be suitable for specific application and it helps the researchers to select appropriate dataset for evaluation. Finally, this paper concludes with open issues and directions for future research which may further motivate and guide the academic and industrial personnel, involved in developing CV-based crowd disaster management systems.

Language: en

Keywords

Behavior analysis; Computer vision; Crowd density estimation; Crowd disaster avoidance system; Crowd evacuation