Fast and automated recognizing of abnormal behaviors in crowded scenes is significantly effective in increasing public security. The traditional procedure of recognizing abnormalities in the Web of Thing (WoT) platform comprises monitoring the activities and describing the crowd properties such as density, trajectory, and motion pattern from the visual frames. Accordingly, incorporating real-time security monitoring based on the WoT platform and machine learning algorithms would significantly enhance the influential detection of abnormal behaviors in the crowds. This paper addresses various automatic and real-time surveillance methods for abnormal event detection to recognize the dynamic crowd behavior in security applications. The critical aspect of security and protection of public places is that we cannot manually monitor the unpredictable and complex crowded environments. The abnormal behavior algorithms have attempted to improve efficiency, robustness against pixel occlusion, generalizability, computational complexity, and execution time. Similar to the state-of-the-art abnormal behavior detection of crowded scenes, we broadly classified methods into different categories such as tracking, classification based on handcrafted extracted features, classification based on deep learning, and hybrid approaches. Hybrid and deep learning methods have been found to have more satisfactory results in the classification stage. A set of video frames called Motion Emotion Dataset (MED) is employed in this study to examine the various conditions governing these methods. Incorporating an appropriate real-time approach with considering WoT platform can facilitate the analysis of crowd and individuals’ behavior for security screening of abnormal events.

快速、自动化地识别拥挤场景中的异常行为，对于提高公共安全具有显着的作用。在 Web of Thing (WoT) 平台中识别异常的传统程序包括监控活动并从视觉框架中描述人群属性，例如密度、轨迹和运动模式。因此，结合基于 WoT 平台和机器学习算法的实时安全监控将显着增强对人群中异常行为的影响检测。本文介绍了用于异常事件检测的各种自动和实时监视方法，以识别安全应用中的动态人群行为。公共场所安全和保护的关键方面是我们无法手动监控不可预测和复杂的拥挤环境。异常行为算法试图提高效率、对像素遮挡的鲁棒性、通用性、计算复杂度和执行时间。类似于最先进的拥挤场景异常行为检测，我们将方法大致分为不同的类别，例如跟踪、基于手工提取特征的分类、基于深度学习的分类和混合方法。已经发现混合和深度学习方法在分类阶段有更令人满意的结果。本研究采用一组称为运动情感数据集 (MED) 的视频帧来检查控制这些方法的各种条件。