Modelling Crowd Scenes for Event Detection

Abstract

This work presents an automatic technique for detection of abnormal events in crowds. Crowd behaviour is difficult to predict and might not be easily semantically translated. Moreover it is difficulty to track individuals in the crowd using state of the art tracking algorithms. Therefore we characterise crowd behaviour by observing the crowd optical flow and use unsupervised feature extraction to encode normal crowd behaviour. The unsupervised feature extraction applies spectral clustering to find the optimal number of models to represent normal motion patterns. The motion models are HMMs to cope with the variable number of motion samples that might be present in each observation window. The results on simulated crowds demonstrate the effectiveness of the approach for detecting crowd emergency scenarios.

Texture-based homogeneity analysis for crowd scene modelling and abnormality detection

“…After composing the average optical flow, 20 STT slices are then composed along each of the 4 directions (80 slices in total). The overall performance of the proposed approach and the prototype system based on the ROC and RP tests are comparable to other works in the field [10,19].…”

Classification of traffic video based on a spatiotemporal orientation analysis

“…The analysis of the movement of pedestrians and crowds is an inter- esting domain for further application of the proposed system (cf. [2]). …”

Crowd Analysis Using Computer Vision Techniques

“…To overcome this problem, some authors [41], [2] have used crowd simulation algorithms to generate controlled situations with known ground truth to test their algorithms. In fact, concepts related to crowd simulation are also being explored to distinguish normal and abnormal behaviors, as in [42].…”

“…Andrade et al [41] characterized a "usual behavior" of a crowd based on the analysis of their optical flow, using hidden Markov models (HMMs). In their work, an unsupervised algorithm is used for feature extraction and spectral clustering.…”

“…While such research area aims to provide group behaviors in a generic point of view, with main applications in games and simulators, this kind of work can also be used to improve crowd analysis. In fact, computer graphics have been used to train or validate computer vision algorithms [9], [10], and in particular, crowd synthesis approaches have been explored to validate crowd analysis algorithms [11], [41], [2]. Conversely, computer vision algorithms designed for crowd analysis can be used to extract information from real life in an automatic manner, which may reduce manual intervention and help to improve the realism of crowd synthesis algorithms.…”