Multimedia Analysis on User-Generated Content for Safety-Oriented Applications

Abstract

An important factor that boosts the rapid penetration of smartphone devices is the increasing incorporation of sensors, which have stimulated a new type of content, the so-called user-generated content. The huge amount of media information accumulating every day presents an opportunity to incorporate image analysis methods and applications for a safer and more secure environment for citizens. This chapter proposed an anomaly detection mechanism for video streams, especially from social media. The methodology employs low-level feature extraction over non-overlapping frame patches and density-based clustering. The core idea consists of two steps: cluster the image patches and observe the difference in the number of clusters for successive images. A threshold-based approach triggers the detection mechanism by investigating the change in the number of clusters. The proposed unsupervised approach runs smoothly on ordinary desktop computers and operates in real time. This chapter outlines the approach and underlying methodology together with an evaluation based on YouTube videos depicting car explosions.