Back to Tao Zhao's Research Page

Car Detection in Low Resolution Aerial Images

Tao Zhao and Ram Nevatia
{taozhao|nevatia} (at) iris (o) usc (o) edu
ICCV 2001 (International Conference on Computer Vision), Vancouver, British Columbia, Canada, July, 2001.
Image and Vision Computing Journal, to appear.

Abstract

We present a system to detect passenger cars in aerial images where cars appear as small objects. We pose this as a 3D object recognition problem to account for the variation in viewpoint and the shadow. We started from psychological tests to nd important features for human detection of cars. Based on these observations, we selected the boundary of the car body, the boundary of the front windshield, and the shadow as the features. Some of these features are a ected by the intensity of the car and whether or not there is a shadow along it. This information is represented in the structure of the Bayesian network that we use to integrate all features. Experiments show very promising results even on some very challenging images.

Car Model and Features

A wireframe car model is used and the features used are: 4 boundary sides (in red), 4 front windshield sides (in blue), (at most) two shadow boundary sides (in green) and intensity of the shadow region.

Feature Integration

The features are collected from the image and integrated using a Bayesian network with the following structure. Car intensity and the shadow beside the feature are two enviornmental variables measured from the image or predicted by sun angle. The parameters of the Bayesian network is learnt from examples.

Selected Results

Washington DC Dataset

Fort Hood Dataset

Download PDF file of the paper (conference version).
Download PDF file of the paper (journal version).