Anomalous Event Detection in Videos Using Supervised Classifier

Abstract

Observing and modeling human behavior and activity patterns for detecting anomalous events has gained more attention in recent years, especially in the video surveillance system. An anomalous event is an event that differs from the normal or usual, but not necessarily in an undesirable manner. The major challenge in detecting such events is the difficulty in creating models due to their unpredictability. Most digital video surveillance systems rely on human observation, which are naturally error prone. Hence, this work validates the rising demand of analysis of video surveillance system. The system being proposed here is of minimum requirements with a competitive computational power when compared to the existing ones.

The main objective of this research work is to build up a framework that recognizes small group of human and to detect the event in the video. A combination of feature extraction using Histogram of Oriented Gradient (HOG) and feature reduction with Principle Component Analysis (PCA) is proposed in this work. The knowledge base and video feed for test cases are classified using the Support Vector Machine (SVM) to categorize the event as either anomalous or not based on various parameters.

The experimental result demonstrates that this approach is able to detect anomalous events with a competitive success rate. The framework can be used to identify various events such as anomalous detection of events, counting people, fall detection, person identification, gender classification, human gait characterization etc.