Best-fit calibration with multiple points

Description

We can improve the calibration of Lecturesight for the pan-only case by using multiple calibration points (5 or more), and then using a best-fit algorithm to map overview co-ordinates to camera co-ordinates.

This should be more accurate than the linear mapping currently used which uses only the left and right points of the overview image, especially where the camera is not positioned centrally, or when the camera is relatively close to the front of the venue.

The elements of this are:

1. Extend the Scene Profile editor to (A) save the overview dimensions with the scene profile, (B) add a Calibration zone type, so multiple calibration points can be identified on the overview image and saved with the scene profile.

2. Extend the Calibration UI so it reads the scene profile, and allows the camera to be positioned at each of the calibration points in turn, and then saves the co-ordinates for each calibration point (probably in a new calibration.properties file).

3. Update the code which maps overview co-ordinates to camera co-ordinates to use an interpolation technique such as:

http://commons.apache.org/proper/commons-math/apidocs/org/apache/commons/math3/analysis/interpolation/AkimaSplineInterpolator.html

(note this requires 5 or more points). We can extrapolate linearly from the left-most two calibration points to get the left scene limit, and the right-most two calibration points to the get the right scene limit.

Activity

Show:
Stephen Marquard
April 28, 2016, 11:40 AM

Image shows the calibration markers in the Scene Profile UI. There is a new "Add Calibration Marker" icon on the right underneath the "measure" icon, and 4 calibration markers are shown placed on the overview image, as black circles with red cross-hairs.

Assignee

Stephen Marquard

Reporter

Stephen Marquard

Components

Affects versions

Priority

Major
Configure