Re: ragarding simon development

Hi Mahesh,

Am Freitag, 23. März 2012, 15:22:19 schrieb Mahesh Hegde:
> *secondly for face recognition I prefer to use opentld(track learn detect)
> algorithm,which was basically implemented for realtime object tracking.Its
> pretty amazing*
> *and accurate because of following reasons:*
> *Input to this is single template from first frame *
> *It is robust to intensity(invariance to illuminations) variations and
> rotation and size*
> *Feature is combination of haar wavelets and local binary patterns.*
> *Feasibility in terms of number of features*
> *It provides proper parameters to estimate accuracy in recognition*
> *these are some of the filtered advantages*
TLD looks like a great tracking algorithm (I had already seen a predator video 
once and was amazed about how good it actually works) but isn't it "just" a 
tracking algorithm? How will that help with the actual recognition? Or am I 
misunderstanding you here?

> *Since future ideas of simon like acoustic modelling(which i liked the
> most) are based on depth recognition,the above algorithm needs less or no
> modification in *
> *providing necessary feature for modelling.*
If you are referring to the distance to the microphone: As this will mostly be 
applicable in controlled environments, tracking the location will likely use 
different techniques (like ZigBee).

Best regards,
Peter