Anomaly Detection and Localization in Crowded Scenes

Abstract

Abstract-The detection and localization of anomalous behaviors in crowded scenes is considered, and a joint detector of temporal and spatial anomalies is proposed. The proposed detector is based on a video representation that accounts for both appearance and dynamics, using a set of mixture of dynamic textures models. These models are used to implement 1) a center-surround discriminant saliency detector that produces spatial saliency scores, and 2) a model of normal behavior that is learned from training data and produces temporal saliency scores. Spatial and temporal anomaly maps are then defined at multiple spatial scales, by considering the scores of these operators at progressively larger regions of support. The multiscale scores act as potentials of a conditional random field that guarantees global consistency of the anomaly judgments. A data set of densely crowded pedestrian walkways is introduced and used to evaluate the proposed anomaly detector. Experiments on this and other data sets show that the latter achieves state-ofthe-art anomaly detection results.

Detection and Localization of Anomalies from Videos based on Optical flow Magnitude and Direction

“…The experiment on both datasets is performed on MATLAB, the optical flow code is mixed, with system configuration of Intel i7 processor 3.40 GHz and 8 GB RAM. As mentioned in [7], the detection can be of two types, frame level and pixel level. The proposed method at frame level has achieved for Ped1 dataset AUC of 67.78%, EER of 34.85% and computation time is 0.9 fps.…”

“…The benchmark dataset [14,15] contain training videos with only normal behaviors and testing videos with abnormalities, which help to clearly distinguish between normal and abnormal situations so that an automated system can take a proper decision. For learning normal behaviors, recent approaches are built on motion and appearance [2,3], spatiotemporal [6][7][8] and trajectories [9,10] as prime features. The input frame is subdivided into small blocks called cell and the features are computed on this small part so that detail analysis can be done.…”

“…Temporal relations are captured by Hidden Markov Model (HMM) and spatial by coupled HMM. Li et al [7], proposed a combine detection and localization of anomalous behaviors in the crowded scenes for temporal and spatial anomalies based on Mixture of Dynamic Textures (MDT). Authors have introduced two normalcy models that jointly account for the appearance and dynamics of complex crowd scenes, temporal normalcy and spatial normalcy.…”

Abnormal crowd behavior detection using novel optical flow-based features

“…To achieve this, two key issues need to be addressed: event representation and anomaly measurement. For abnormal event representation, some methods consider the spatial-temporal information, such as Histogram of Optical Flow (HOF) [9], Histogram of Motion Direction (HMD) [10], spatial-temporal gradient [11], social force model [12], chaotic invariant [13], mixtures of dynamic textures [14], force field [15] and sparse representation [16]. On the other hand, for anomaly measurement, generally a one-class learning method is used to learn normal samples.…”

Pedestrian detection based on hierarchical co-occurrence model for occlusion handling

“…Pedestrian detection is an important topic for practical applications, such as video surveillances [9,29], intelligent vehicles [10], and robot sensing. State-of-the-art algorithms have been used for achieving progress on pedestrian detection.…”