When running run_TLD.m, it should look like http://youtu.be/r6lNNPl-2Oc
… subtraction. Now the object does not need to be present in the first frame. However, a background image is required.
…of the debug flags.
…led in run_TLD. - Turn off the save object flag in run_TLD.
…f the first frame. The only requirement is a picture of the background inside the folder [opt.source.input 'background/'].
Basically, it saves the part of each frame that is surrounded by the bounding box. The fps rate seems to drop with about 10% when you turn saving on (for the motorbike example).
…lation time analysis.
Bug when acquiring frame from stream server solved by reshaping the array passed by sockets. Hardcoded the dimentions and position of the initial object selection box. Now the octave version can use webcam stream.
Replaced ocorrencies of exist('videoinput') by exist('videoinput','file').
… package. This works well under matlab 2010a, OpenCV 2.0+ and Ubuntu 10.04. The only thing the user has to do is execute ./server_stream before running TLD and set the in run_TLD.m the flag for camera capture. This code is very close to working with Octave. Octave can succesfully grab the frames, but has problemas with the mouse input to draw the rectagle over the first object. This version should have the same behavior as before for people using the Image Acquisition package. For this keep executing as before.
This completes the clustering functionality that OpenTLD needs. Note that this is a standard matlab .m file so unlike linkagemex.cpp its therefore not optimized. However this may not such be a concern as linkagemex is probably doing the most work by first building the cluster tree. The missing cluster() function then splits the cluster built be linkagemex. See: http://www.mathworks.com/help/toolbox/stats/linkage.html http://www.mathworks.com/help/toolbox/stats/cluster.html