Abstract:Anomaly detection has significant importance for the development of autonomous monitoring systems. Real-world anomalous events are complicated due to diverse human behaviors and class variations. Anomalous activities depend upon speed, length of activity, and motion features to comprehend suspicious behaviors. Fast activities are captured quickly within a few video frames, whereas slow actions may take several hundred video frames to define an anomalous action. Furthermore, a video is more than just a stack of frames with spatiotemporal representations. Most of the existing approaches suffer from learning variable speed fast and slow activities simultaneously and primarily focus on learning spatiotemporal features only. Modeling the spatiotemporal and motion relationships between frames together can help understand the actions better. Motion features when combined with spatiotemporal representations perform higher. Our contribution is two-fold in this research work. Firstly, a novel dynamic frame-skipping approach is proposed to duly generate meaningful representations of spatiotemporal frames and optical-flow-based motion representations for variable speed anomalous actions. Secondly, AnomalyNet, as a new end-to-end deep architecture, is designed to simultaneously learn both spatiotemporal and motion features in image sequences. AnomalyNet is evaluated on the challenging real-world anomaly detection datasets. The results confirm that the proposed model has achieved a competitive AUC of 86.1% on the real-world UCF-Crime dataset and has achieved a superior AUC score of 99.87% compared to state-of-the-art methods on challenging ShanghaiTech dataset in the domain of unsupervised, weakly-supervised, and fully-supervised anomaly detection. Furthermore, the model achieved the highest F1 score for both fast and slow variable speed anomalous activities, such as explosions, road accidents, robbery, and stealing for real-world autonomous surveillance.

Learning Motion Regularity for Temporal Video Segmentation and Anomaly Detection

Learning Attention Augmented Spatial-temporal Normality for Video Anomaly Detection

Video anomaly detection and localization based on appearance and motion models

Learning Appearance-motion Normality for Video Anomaly Detection.

Motion-Aware Feature for Improved Video Anomaly Detection

Real-world Video Anomaly Detection by Extracting Salient Features in Videos

Consistency-based Self-supervised Learning for Temporal Anomaly Localization

Self-Supervised Representation Learning for Visual Anomaly Detection

Motion anomaly detection and trajectory analysis in visual surveillance

Spatio-Temporal-based Context Fusion for Video Anomaly Detection

MTFL: Multi-Timescale Feature Learning for Weakly-Supervised Anomaly Detection in Surveillance Videos

AnomalyNet: a spatiotemporal motion-aware CNN approach for detecting anomalies in real-world autonomous surveillance

Anomaly Detection Via Midlevel Visual Attributes

Multi-Scale Temporal Relations and Segmented Channel Attention for Video Anomaly Detection

Video Anomaly Detection and Localization Using Motion-Field Shape Description and Homogeneity Testing

Unsupervised Video Anomaly Detection with Diffusion Models Conditioned on Compact Motion Representations

Learning Appearance-Motion Synergy Via Memory-Guided Event Prediction for Video Anomaly Detection

Memory-Augmented Spatial-Temporal Consistency Network for Video Anomaly Detection.

Drastic Anomaly Detection In Video Using Motion Direction Statistics

Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning

Video anomaly detection based on attention and efficient spatio-temporal feature extraction