Anomaly detection in crowd scene

Abstract

Anomaly detection in crowd scene is very important because of more concern with people safety in public place. This paper presents an approach to automatically detect abnormal behavior in crowd scene. For this purpose, instead of tracking every person, KLT corners are extracted as feature points to represent moving objects and tracked by optical flow technique to generate motion vectors, which are used to describe motion. We divide whole frame into small blocks, and motion pattern in each block is encoded by the distribution of motion vectors in it. Similar motion patterns are clustered into pattern model in an unsupervised way, and we classify motion pattern into normal or abnormal group according to the deviation between motion pattern and trained model. The results on abnormal events detection in real video demonstrate the effectiveness of the approach.

Motion Influence Map for Unusual Human Activity Detection and Localization in Crowded Scenes

“…Illustration of k-means clustering with frame division. (1,1), (1,9), and (14,1) denote, respectively, the coordinate of the respective mega blocks. blocks at each frame, and finally concatenate the motion influence vectors of the recent t number of frames.…”

“…In our work, we estimate the motion information indirectly from the optical flows [9], [12]. Specifically, after computing the optical flows for every pixel within a frame, we partition the frame into M by N uniform blocks without a loss of generality, where the blocks can be indexed by {B 1 , B 2 , · · · , B M N }, and then compute a representative optical flow for each block by taking the average of the optical flows of the pixels within the block:…”

“…To this end, recent researches have focused on the direct use of motion patterns in an image. Wang et al used Kanade-LucasTomasi (KLT) corners [19] to represent moving objects and clustered similar motion patterns in an unsupervised manner [9]. They detected anomalies in a frame sequence using two types of historical motion descriptors: the self history and the neighboring history [10].…”

“…In the meantime, other groups have used optical flows that directly characterize motion features in a sequence, e.g., a motion heat map [8], clustered motion patterns [9], [10], spatial saliency of the motion feature [11], crowd prediction using a force field model [12], optical flow fields [13], particle trajectory [14], a social force model [15], and a local motion histogram [16]. Although motion flow based approaches have shown their efficacy in previous works, we believe it is still important to consider the information on the size of the objects and their interactions.…”

Detection of Anomalous Crowd Behaviour Using Hyperspherical Clustering

“…Wang and Miao [12] extracted the Kanade-Lucas-Tomasi (KLT) corners indicating the moving objects and optical flow for tracking the feature points. Identical motion patterns from different blocks were clustered to generate a model and was classified as normal or abnormal based on the amount of deviation from the trained model.…”