Visual Navigation

Jean-Yves Bouguet and Pietro Perona

Department of Electrical Engineering
California Institute of Technology 136-93, Pasadena, CA, 91125


We are developing techniques for estimating the full three dimensional motion of a vehicle from a image sequence acquired with an onboard video camera. The process is done without making any assumption about neither the type of motion nor the shape of the environment. The outcome is the complete vehicle trajectory and the three-dimensional structure of the scene. [ICCV95]


The purpose of this project is to build a system for estimating the full three-dimensional trajectory of a moving vehicle from the image sequence acquired by an onboard video camera. The main applications are assisted and autonomous navigation. There are situations (outdoors navigation, navigation in towns) where a model is either outright impossible or impractical to build, therefore techniques for navigation have to be developed that can work well in unstructured environments. We present here experiments on a long sequence that represents well typical indoor and city navigation.


We are experimenting different algorithms for motion and structure estimation from visual input (known as Structure-From-Motion algorithms in the vision community). This page regroups experimental results achieved on a long navigation sequence (4000 images). It includes: feature tracking results, 3D motion and trajectory estimations, and 3D structure reconstructions .

Initial sequence (3985 images)
mpeg movie (250 frames)
Feature tracking
mpeg movie (250 frames)
Tracked features
mpeg movie (250 frames)

Reconstructed 3D structure and camera trajectory
mpeg movie (250 frames) - Top view
mpeg movie (250 frames) - Side view


We have demonstrated that automatically reconstructing the three-dimensional trajectory as well as the structure of the scene is feasible from purely visual input (monocular) without making any prior information about the type of motion or the structure of the envirnoment. However, further constraints may be enforced (for example planarity of the motion). The next step is to use the computed vehicle motion and scene structure for autonomous navigation.

Online Reports:

Back to main page

Number of visits since Nov. 3rd, 1998: