An Implementation of AdaBoost on the RavenHX:
In May of 2005 I was presented with the RavenHX
DSP board made by MangoDSP. This project has focused around learning effective
optimization techniques to be used in DSP implementations. For this project
I had to train an AdaBoost cascade using pre-existing software, and then
write the face detection code to utilize that cascade. The initial implementation
with no optimizations ran at 0.5 Frames Per Second (FPS). The system currently
operates at 30 FPS. This increase in speed has been due to the optimizations
implemented.
At the top left is a picture of the board I am using, and to the right is a resulting frame from the face detection code: