B. Hints for Better Localization
Paying attention to the details when using tags for localization can make a big difference for the quality of the results!
Camera and Lighting
Some important points when selecting a camera:
- Lighting: illuminate the tags as brightly as possible. The more light, the shorter the shutter times, the less motion blur. Avoid backlight when possible.
- Check that your camera is focused correctly for the distance you need. When in doubt, set focus to a distances a little longer than you expect (focus is most important for far away tags)
- Use a camera with global shutter, not rolling shutter. Most consumer cameras (webcams, GoPros etc) have rolling shutters that can distort objects that are in motion.
- Short shutter times are crucial for avoiding motion blur. To this end, you want:
- a large sensor with high quantum efficiency. The FLIR machine vision website (formerly PointGrey) has excellent sensor comparisons.
- a large aperture lens (but beware of focus!)
- Auto focus cameras are to be avoided because their calibration could change while recording.
- If using multiple cameras, they must be synchronized (use hardware sync!), because TagSLAM cannot handle unsynchronized camera streams!
- Higher resolution is better. If the resolution is too low, you may not be able to detect any tags.
- Generally wide-angle lenses (up to 120deg) are preferred, but they must be used in combination with a high resolution sensor.
Here some tips on tag placement:
- always have multiple tags in view. Localizing from a single tag is very fragile.
- make sure tags are visible (unoccluded) from all points on the camera path.
- avoid placing tags where there is a lot of backlight
- space tags as far from each other as possible, but in such a way that always multiple tags are still visible.
- resist the temptation to arrange your tags in a linear fashion, for example leaning them against the wall on the floor. TagSLAM uses a triangulation based approach, which works best with … triangles. The following video shows that better results can be obtained with 3 tags in a triangular arrangement than from 5 tags that are collinear.