Anomaly detection using DBSCAN clustering technique for traffic video surveillance

Detecting anomalies such as rule violations, accidents, unusual driving and other suspicious action increase the need for automatic analysis in Traffic Video Surveillance (TVS). Most of the works in Traffic rule violation systems are based on probabilistic methods of classification for detecting the events as normal and abnormal. This paper proposes an un-supervised clustering technique namely Novel Anomaly Detection-Density Based Spatial Clustering of Applications with Noise (NAD-DBSCAN) which clusters the trajectories of moving objects of varying sizes and shapes. A trajectory is said to be abnormal if the event that never fit with the trained model. Epsilon (Eps) and Minimum Points (MinPts) are essential parameters for dynamically calculating the sum of clusters for a data point. The proposed system is validated using benchmark traffic dataset and found to perform accurately in detecting anomalies.