Dynamic event detection in crowded environment by HOS
Abraham Mathew, Savitha V. Nair, B. Vinoth · International Conference on Electrical, Electronics, and Optimization Techniques · 2016
For traditional computer vision methods the analysis of motion and behaviours in crowded scenes constitutes a challenging task, as barriers like occlusions, varying crowd densities and complex stochastic nature of their motions are difficult to overcome. As it has to be kept within reasonable limits, the one more complicating factor is the computational cost. It is very crucial to analyse crowded scenes in real time, or at least fast as possible, in many practical situations, considering the fact that security personnel should act quickly if something seems to be “unusual”. Anomalous is a problem which is not fitting into a familiar type, classification or pattern. HOS (Histogram of Oriented Swarm) is used for detecting and localizing anomalous events in videos of crowded scenes. HOS together with the HOG (Histogram of Oriented Gradient) are combined to give a descriptor that helps to effectively characterize each scene. The occurrences of gradient orientation in localized portion of images can be count by this technique. The HOS descriptor analyses and localizes the anomalous and normal events separately.