Recognition of Anomalous Motion Patterns in Urban Surveillance

Abstract

Abstract-We investigate the unsupervised K-means clustering and the semi-supervised hidden Markov model (HMM) to automatically detect anomalous motion patterns in groups of people (crowds). Anomalous motion patterns are typically people merging into a dense group, followed by disturbances or threatening situations within the group. The application of Kmeans clustering and HMM are illustrated with datasets from four surveillance scenarios. The results indicate that by investigating the group of people in a systematic way with different K values, analyze cluster density, cluster quality and changes in cluster shape we can automatically detect anomalous motion patterns. The results correspond well with the events in the datasets. The results also indicate that very accurate detections of the people in the dense group would not be necessary. The clustering and HMM results will be very much the same also with some increased uncertainty in the detections.

Dynamic Phase and Group Detection in Pedestrian Crowd Data Using Multiplex Visibility Graphs

“…They identifies two groups, containing agents (0,1) and agents (2,3,4) respectively merging together, then splitting to groups with agents (0,1,2) and agents (3,4). Cross-checking these groups with the trajectories shown in Figure 1 confirms that the correct clusters have been identified, indicating that agent 2 has moved across from one group to another during Scene 0.…”

“…Our method does this efficiently by computing correlations between the sparse adjacency matrices describing the layers, and thereby avoids using computationally expensive approaches such as K-means clustering conditioned on density estimation [2].…”

“…These data are represented visually as graphs of layers in Figure 4 with the caveat that more chaotic behaviour, such as that in the fight scene, loses any strong mutual correlations between the layers of the DMVG. [13,2,3] investigate anomalous movements in crowd scenes using topology [13] and hidden Markov models [3,2]. On the other hand [9,15,17] model pedestrian group formation using techniques such as social force models and optical flow.…”

Detection and Localization of Anomalous Motion in Video Sequences from Local Histograms of Labeled Affine Flows

“…Determined urban groups dynamics can also be viewed as a special case of anomaly detection in crowded videos. With this goal, the authors in Andersson et al (2013) proposed an algorithm to detect disturbances caused by individuals merging groups. Other works are able to detect anomalies locally in videos and without an explicit definition of what the abnormality is.…”

Activity recognition and localization on a truck parking lot

“…The approach is evaluated with real tracking data as input data to investigate its performance when there are uncertainties present. A similar approach has earlier been evaluated with annotated data as input data [8].…”