CSAIL logo

Body-Relative Navigation Using Uncalibrated Cameras

Authors: Olivier Koch, Seth Teller


Our goal is to design a system and set of algorithms that help humans find their way within GPS-denied environments (indoor, or dense urban areas). Our work has two main applications: soldiers in the field, who tend to get lost easily during stressful missions; the visually impaired, who need navigation assistance in absence of blind dog or human help.

We propose a purely vision-based solution based on an uncalibrated set of cameras. Neither the intrinsic or extrinsic parameters of the cameras are required. Instead, the algorithms learns automatically the correspondence between user motion and feature matches between cameras.

The method is entirely non metric and builds a topological map of the environment during exploration. The output of our method is a live guidance provided to the user in their body frame (e.g. turn left, go straight). The method maintains the position of the user in the graph at all times.

Our method requires no camera calibration and provides rotation guidance with a rotation accuracy of 8 to 10 degrees. The method is particularly robust to off-path trajectories and includes a loop closure detection algorithm. Because our method does not rely on frame to frame tracking, it is also robust to typical high-frequency human motion (e.g. falling, ducking, etc.)


Portable multi-camera system composed of four Pointgrey Firefly MV cameras. Horizontal FOV: 360 deg. Vertical FOV: 90 deg.

Place graph built online during exploration. The arrows show the guidance provided to the user in a replay scenario.

Rotation guidance is provided in the user's body frame (red arrow) all along the exploration path. The method relies on matching points of interests (e.g. SIFT, shown in yellow) upon revisit of a location.

References

O. Koch and S. Teller, Body-Relative Navigation using Uncalibrated Cameras, IEEE International Conference on Computer Vision, Kyoto, Japan, 2009 (to appear)

Here are the slides [PDF] for a presentation of my work I have been giving to ECCV 2008, IROS 2008, and within several labs in Europe.