Meeting Date: 23.08.1999
What I did last week: I compiled the other training set with more complex gestures like House, Pentagon and Heart. The interesting thing is that these quite complicated gestures are recognized fairly well (with the new recognition system. With the old one, they wouldn't have stood a chance... unfortunately). Now the biggest problem seems to be for people to be able to draw them. For example I choose the House because there are so many ways that you can draw it -- but not all possibilities work (not in one continuous line anyway). Also some people had trouble drawing with the polhemus tracker. The difference in ability is quite striking. Oh, I suppose I should mention, that I ported the new recognition system to CoRgi. It amounted to almost 2000 lines of code, which had to be bundled into a class and provided with the same interface as the first recognizer system. It took me a whole day to do the porting and another to get a memory issue resolved. For future reference:

If allocating memory with:
myvar = new MYTYPE [many]; // we want to delete with
delete [] myvar; // as opposed to just delete myvar;

This may seem trivial to some, but is easily overlooked and can cost as many as plenty of hours if neglected!!! So remember it. Anyway, it is working now. Yeah!

What I'll do this week: I have two courses at the moment and still to finish up advanced architecture, so I'll be doing some course work. Otherwise I will create some gestures that people have less hassles drawing, so I can send them out and test the system. I also want to neaten up the code a bit and the object-hierarchy, so that users of the recognizer will not have to deal with two recognizers seperately. The correct one for them will be choosen, depending on what kind of training data is found in the directory specified in the constructor. Default will be the new recognizer. Otherwise that will be it for the programming part of my project. I will do some collection of data from now on and get people to test my system under different conditions, with different gestures and with the different recognizers.