Recently, intelligent video surveillance applications have become essential in public security by the use of computer vision technologies to investigate and understand long video streams. Anomaly detection and classification are considered a major element of intelligent video surveillance. The aim of anomaly detection is to automatically determine the existence of abnormalities in a short time period. Deep reinforcement learning (DRL) techniques can be employed for anomaly detection, which integrates the concepts of reinforcement learning and deep learning enabling the artificial agents in learning the knowledge and experience from actual data directly. With this motivation, this paper presents an Intelligent Video Anomaly Detection and Classification using Faster RCNN with Deep Reinforcement Learning Model, called IVADC-FDRL model. The presented IVADC-FDRL model operates on two major stages namely anomaly detection and classification. Firstly, Faster RCNN model is applied as an object detector with Residual Network as a baseline model, which detects the anomalies as objects. Besides, deep Q-learning (DQL) based DRL model is employed for the classification of detected anomalies. In order to validate the effective anomaly detection and classification performance of the IVADC-FDRL model, an extensive set of experimentations were carried out on the benchmark UCSD anomaly dataset. The experimental results showcased the better performance of the IVADC-FDRL model over the other compared methods with the maximum accuracy of 98.50% and 94.80% on the applied Test004 and Test007 dataset respectively.

最近，通过使用计算机视觉技术来调查和理解长视频流，智能视频监控应用在公共安全中变得必不可少。异常检测和分类被认为是智能视频监控的主要元素。异常检测的目的是在短时间内自动判断异常的存在。深度强化学习（DRL）技术可用于异常检测，它集成了强化学习和深度学习的概念，使人工智能能够直接从实际数据中学习知识和经验。基于此动机，本文提出了一种使用具有深度强化学习模型的 Faster RCNN 的智能视频异常检测和分类，称为 IVADC-FDRL 模型。提出的 IVADC-FDRL 模型在两个主要阶段运行，即异常检测和分类。首先，将 Faster RCNN 模型用作对象检测器，以 Residual Network 作为基线模型，将异常检测为对象。此外，采用基于深度 Q 学习 (DQL) 的 DRL 模型对检测到的异常进行分类。为了验证 IVADC-FDRL 模型的有效异常检测和分类性能，在基准 UCSD 异常数据集上进行了大量实验。实验结果展示了 IVADC-FDRL 模型的性能优于其他比较方法，在应用的 Test004 和 Test007 数据集上的最大准确率分别为 98.50% 和 94.80%。