AN IMPROVED MIXTURE OF GAUSSIAN MODEL FOR REAL TIME VEHICLE DETECTION

Abstract

This paper proposes a novel method to segment video sequences which undergoes gradual changes into foreground and background layers. The background layer contains all objects which have been stationary since the beginning of the video sequence. The foreground layer contains objects which have entered into or move within the video scene and these objects can be moving or stationary. An improved and adaptive Mixture of Gaussian (MoG) model with a feedback mechanism algorithm has been formulated. The MoG model will classify every pixel in the image as belonging either the foreground or the background layer. Every object in the foreground layer will be tracked and updated in the MoG via the feedback mechanism. This feedback avoids stationary foreground objects being updated into the MoG and thus affecting the approximation done by the MoG. This algorithm has been implemented into an Intelligent Transportation System (ITS) to detect vehicles on the road in an outdoor environment. A promising result is obtained in extracting vehicles on the road.