In this project we developed a generic object detector based on covariance matrices as object descriptor and Support Vector Machines as classifier. The system is able to process VGA images at more than 60 frames per second, with an FPGA maximum working clock-frequency of 213 MHz. Details can be found on my ICDSC 2011 paper*.
We are currently extending the architecture to detect objects using not only intensity cue but also stereo-based depth maps and optical flow.
* S. Martelli, D. Tosato, M. Cristani and V. Murino.
FPGA-based pedestrian detection using array of covariance features.
In International Conference on Distributed Smart Cameras (ICDSC), Ghent – Belgium