Tabs

Calibration

Introduction

The calibration stage is essentially linking the cameras to the real environment (including each other). The main purpose of the calibration stage is to create a relative 3D reference for each camera (so that each camera knows where it is in 3D space relative to each other camera; in addition we reference the relative distances of the camera with the world origin as it is in the actual room

The system needs to know where the cameras are relative to each other in order to calculate the triangulation of the markers, as described in the introductory section. The system needs to know where the origin of the world is in order that the real and virtual spaces match up (i.e. it is not for triangulation).

The other part of the calibration is to determine the distortion of the lens (image distortion is shown below). Image distortion is a divergence from rectilinear projection caused by a change in magnification with increasing distance from the optical axis of an optical system and means that the further the marker is from the centre of the image, the more it will be distorted. Obviously this distortion is unwanted, but is generally present even in the most expensive lenses (to lesser degrees). By using this calibration stage, the software can determine the distortion and create what is called a "Distortion Map", this is used to correct for the distortion so that a marker on the edge of the Field of View, is represented as closely as it would if it were in the centre of the of the Field of View.

Distortion

Wand

Calibration is performed using a calibration wand; the wand is setup with 3 markers setup to known distances (the software has knowledge of these distances). By recording the movement of the wand throughout a set of frames, the software can use this information to make a correspondance and set the positions in 3D space.

This movement is generally termed the "Wand Wave" and our objective is to provide the cameras with as much information as possible in order to calculate the 3D position, and the distortion maps. In this respect each camera should have the wand information presented to fill its viewing frame over the period of the "Wand Wave".

The "Wand Wave" is tricky, too little data and the 3D space calculation and the distortion will be too approximate, too much data and errors in recording will add up into bigger errors in the final calculation (in a kind of noise to signal ratio). In general, for the space provided, we would want to have approximately 6000 frames in total (+/- 10%) in order to have sufficient data, whilst not going over the recommendation.

For all body tracking we use the 240mm wand (which has 3 markers at 0mm, 80mm and 240mm along the top edge), we need to make sure this is specified in Vicon before proceeding with the actual calibration.

It is also important to remember that whilst the wand is 240mm or 120mm depending on what we are doing, the most important aspect is the relative distance of the markers (0, 1/3rd, 1) rather than the actual distance itself. We can imagine a wand far from the camera will look the same as a smaller wand close to the camera (with the same distances) - it is assumed that the software ignores the actual size of the marker.

240mm Wand

Wand Wave

The Wand Wave can be done in many ways, the ultimate goal is to fill the frame of each camera, but the easiest way to do the entire process is by "Painting the Room", in this we use the wand like a paint brush and imagine a triangular room created by the boundaries on the floor. We need to "paint" both the inside and outside of the room (because when we paint one side we are blocking certain areas of the view by our body, by painting the inside we manage to avoid this occlusion later on.

A video of the wand wave is shown below:

Wand Wave

Wand Wave, Tips and Tricks

  • Result: Firstly, the goal is either "Excellent" or "Awesome"; anything else really requires the process to be repeated; this is because the amount of distortion for "Good" or less is too high and will result in poor accuracy later on.
  • Wand Movement: The wand should actually be slightly loose in your hand, do not hold it tight, but let it rotate slighty so that whilst you are going up and down, the want is actually rotating around in your hand (say between 180 and 360 degrees every couple of strokes).
  • Speed: Try and maintain a steady speed, 6000 frames, at 120fps, is about 50 seconds (or 25 seconds per room side - inside then outside), with 4 walls, this should take about 6 seconds per wall; try not to rush it, too fast and the wand becomes a blur for the frame.
  • Distance: As discussed in the introduction section, the near focus for the cameras is approximately 1m away, less than that and the marker will be blurred, with the space you should be at least 2m away anyway, but less than 1m will cause problems.
  • High/Low: In order to obtain the entire virtual space, it is a good idea to "nearly" (not actually) touch the ground with the wand, and stretch as a high as you can with the wand. The system is setup to capture about 2.5m above the ground, so extending to that space is essential to make sure all markers are seen properly.
  • Clothes: At any stage when the cameras are capturing data, whether it is for calibration or subject capture, we need to minimise the data. Therefore, even though we are not capturing a person, the markers are not on the person doing the calibration, and the wand is set to a specific distance it is advised to make sure the person doing the capture is not reflective (white t-shirts, sweaty brow, etc) or wearing anything reflective. This will avoid the cameras mistaking a random reflection for the wand (as described above, the main aspect is the relative distances of the markers, not the actual distance from end-to-end).