Video traffic analysis for abnormal event detection using frequent item set mining

As powerful computers and cameras have become wide spread, the number of applications using vision techniques has increased enormously. One such application that has received significant attention from the computer vision community is traffic surveillance. We propose a new event detection technique for detecting abnormal events in traffic video surveillance. The main objective of this work is to detect the abnormal events which normally occur at junction, in video surveillance. Our work consists of two phases 1) Training Phase 2) Testing Phase. Our main novelty in this work is modified lossy counting algorithm based on set approach. Initially, the frames are divided into grid regions at the junction and labels are assigned. The proposed work consist of blob detection and tracking, conversion of object location to data streams, frequent item set mining and pattern matching. In the training phase, blob detection is carried out by separating the modelled static background frame using Gaussian mixture models (GMM) and this will be carried out for every frame for tracking purpose. The blobs location is determined by assigning to the corresponding grid label and numbered moving object direction to form data streams. A modified lossy counting algorithm is performed over temporal data steams for discovering regular spatial video patterns. In testing phase, the same process is repeated except frequent item set mining, for finding the spatial pattern in each frame and it is compared with stored regular video patterns for abnormal event detection. The proposed system has shown significant improvement in performance over to the existing techniques.