Back
to indexKoller, Daniilidis, Nagel
Summary
This paper presented an algorithm to track vehicles in cluttered, outdoor environments using static cameras. The techniques require a database (albeit small) of vehicle templates, and an illumination model to reduce the noise and increase the accuracy in the expected translational velocities of the vehicles in the scene. An interesting question was posed: the agent in the scene must remain in the field of view, so you can either use a stationary camera with a large field of view, or a movable camera with gaze control. The authors selected the stationary model to be able to find vehicles around occulsions with a larger field of view.
Methods
The algorithm matches a simple circular motion pattern with constant magnitude of velocity and constant angular velocity to a normal plane around which a vehicle is moving. The example images in the appendix display vehicles moving around a circular drive in a parking lot, and circular backwards of a car parking in a spot. An external model of the vehicle is depicted in a database of vehicle templates, which may be parameterized to the current vehicle. The database contains the 3D images of the vehicles, to facilitate monitoring the model as it turns. Since the vehicle parameters do not change, as the vehicle moves its motion may be modeled by calculating the difference between the principal axis through the center of the vehicle model, contained in a simple derivative on the circular rotation. The matching between the model and the database is done on the edge segments, and a certain uncertainty variable is introduced.
An illumination model was used to capture the errors that occurred when shadows were misintreped as edges of the vehicle. The illumination model assumed a parallel series of incoming light.
Keywords
illumination model, vehicle templates, velocity control, pattern matching
Rating
7
Bibtex Entry
@article{ koller93model,
author = "D Koller and K Daniilidis and HH Nagel",
title = "Model-Based Object Tracking in Monocular Image Sequences of Road Traffic Scenes",
journal = "International Journal of Computer Vision",
volume = "10",
number = "3",
pages = "257--281",
year = "1993",
url = ""
}