ELECINF344/381

Partie interactive du site pédagogique ELECINF344/ELECINF381 de Télécom ParisTech (occurrence 2011).

Catégories

[Casper] Face tracking

We now have our face tracking algorithm running on BeagleBoard. Since a tracking only based on face detection was to slow and not very intuitive (you always have to look the camera), we decided to use a blob tracking algorithm, which we initialize with the face detection algorithm.

First, Casper looks for a face. When it finds one, it learns the associated color histogram. After what it tracks the histogram (with the OpenCV Camshift function), for a few tens of frames. If it does not find a face again during the blob tracking, it stops at the end of the frames. Otherwise, it keeps tracking the « blob » face.

We adopt a multithread program : a thread looks for a face every second, and a thread is responsible for blob tracking when a face is found. The first thread is used to set a counter which is decremented by the second thread.

 

[CASPER] Today’s news

We progressed in different fields today.

Alain designed a small extension PCB for the beagleboard. This board will include the necessary elements for audio amplification in and out, and for level shifting between the beagleboard’s output and the motors’ input.

At the same time, we worked with Thomas to build a first tracking system demo, by placing the webcam on top of casper’s body, connecting it to the beagleboard and then connecting the serial link to drive the motors. This demo gave some first results tonight, and will be kept under improvement.

Finally, we managed to create a custom language model and dictionary, which combined with the pocketsphinx engine’s data now allow the beagleboard to understand french vocal commands.