On October 29th I started training an hand detector using the Viola/Jones method and the openCV tools. I used 12k positive and 20k negative samples.
Today the training completed and the results were a bit off. I had a lot of false negatives. I collected lots of texture samples usually found in 3d modelling sites and started the training again with 32k positive and 32k negative samples. Hopefully the results will be better.
In the meantime I will keep using the old detector and concentrate on developing the tracker.
Today the training completed and the results were a bit off. I had a lot of false negatives. I collected lots of texture samples usually found in 3d modelling sites and started the training again with 32k positive and 32k negative samples. Hopefully the results will be better.
In the meantime I will keep using the old detector and concentrate on developing the tracker.
