3D blob finder to be used with ofxKinect. Avoids occlusion issues when using a 2D contourfinder
See github.com/dasaki/ofxKinectBlobTracker for a working example code::blocks project/code.
Watch it in action here: https://www.youtube.com/watch?v=uwvYgiwrTwU
Comments:
Part of the cvCinema project: www.cvcinema.com
The technique used is Euclidean Cluster Extraction, as explained in: http://www.pointclouds.org/documentation/tutorials/cluster_extraction.php
To imporve performance the neighbor points are searched in a box (X/Y/Z distances) instead of a sphere (radius/distance). As the kinect cloud is ordered, the adjacency is first checked in the XY plane, within a specified pixel area around each point.
See example of usage in the ofxKinectBlobTracker addon.
The finder function is called findBlobs, and it's parameters are:
ofImage * maskImage a binary grayscale image where black pixels indicates non wanted / background pixels.
const ofVec3f boundingBoxMin first/minimum vertex defining the crop box that delimites the space where points shoud be to be considered in the search for blobs.
const ofVec3f boundingBoxMax, secnd/maximum vertex defining the crop box that delimites the space where points shoud be to be considered in the search for blobs.
const ofVec3f thresh3D X/Y/Z threshold ranges for two points to be considered as adjacent
const int thresh2D xy threshold ranges for two pixels to be considered as adjacent
const float minVol minimum volume that a blob shold have to be included in the result
const float maxVol maximum volume that a blob shold have to be included in the result
const int minPoints minimum number of volume that a blob shold have to be included in the result
const unsigned int maxBlobs maximum number of blobs to return as result