Vehicle image classification via expectation-maximization algorithm

Abstract

In this paper, we present a statistical method to extract images of passenger cars from highway traffic scenes. The expectation-maximization (EM) algorithm is used to classify the vehicles in the images as either being passenger cars or some other bigger vehicles, cars versus non-cars. The vehicle classification algorithm uses training sets of 100-frame video sequences. The car group is comprised of passenger cars and light trucks. The non-car group is comprised of heavy single trucks as well as 3-axle and more combination trucks. We use the properties of their dimensional distribution and the probability of their appearance to identify the vehicle group. We present a validation of our algorithm using real-world traffic scenes.