Scene Representation and Anomalous Activity Detection using Weighted Region Association Graph

Abstract

In this paper we present a novel method for anomalous activity detection using systematic trajectory analysis. First, the visual scene is segmented into constituent regions by attaching importances based on motion dynamics of targets in that scene. Further, a structured representation of these segmented regions in the form of a region association graph (RAG) is constructed. Finally, anomalous activity is detected by benchmarking the target's trajectory against the RAG. We have evaluated our proposed algorithm and compared it against competent baselines using videos from publicly available as well as in-house datasets. Our results indicate high accuracy in localizing anomalous segments and demonstrate that the proposed algorithm has several compelling advantages when applied to scene analysis in autonomous visual surveillance.

Video synopsis generation using spatio-temporal groups

“…It is visually less confusing too. group the trajectories based on various other criteria such as interest area based [18]- [20], movement graph based [21], by supervised or unsupervised machine learning [22], [23], or by using deep learning to understand the activities based on region(s) of interest [18].…”

Motion anomaly detection and trajectory analysis in visual surveillance

“…More recently, Brun et al [11,15] have proposed a scene partitioning approach to classify trajectories using unsupervised learning. Main idea in their method is to divide the scene into non-overlapping zones and to represent it using a graphical structure similar to the method proposed in [43]. Brun et al have clustered car trajectories recorded from parking zones [44] and human trajectories obtained inside a busy railway station [45].…”

“…Brun et al have clustered car trajectories recorded from parking zones [44] and human trajectories obtained inside a busy railway station [45]. The method proposed in [43] has been used to detect very simple anomalies such as a target visiting abandoned or inaccessible regions. Further, a similar method, often refereed as non-conformal recognition technique has also been recently introduced for maritime applications [46].…”

“…Finally, the use of spatio-temporal context analysis for the accurate anomalous event detection and localization has been proposed in [56]. Context-aware anomalous activity detection has also been studied in [43,57]. Here, it has been demonstrated that in order to determine scene dynamics that relate to target interactions, it is critical to analyze the target motion saliency for reliable anomalous activity detection.…”

“…Despite recent developments in this research domain, the problem of motion anomaly detection still remains to be challenging due to the following reasons: a) To identify what constitutes anomalous within the surveillance context and distinguishing it from normal behaviour, b) To propose a scene independent model that can work under varying scene conditions, c) To deal with the complexity in behavioural patterns of moving targets, and d) To operate within reasonable computational overhead. Majority of the above mentioned state-ofthe-art methodologies do not address the first challenge, whereas the method proposed in [43] has been shown to handle very simple motion anomalies. Although the unsupervised approach proposed by Burn et al [11,15] can successfully classify trajectories, this method does not provide any insight on the type and category of the anomaly.…”