In 2007, I joined Texas Instruments as a founding member of the Embedded Vision R&D team. Over the years, we developed a number of embedded prototypes and DSP-optimized libraries that advanced the state-of-the-art in automotive, video, and gesture interfaces.
Our operations were analogous to a computer vision tune-up, combining algorithmic insights with architectural know-how. We would start with a little "algorithmic surgery" to render applications more embedded-friendly. Then we would optimize the C code through C6x intrinsics to ensure tight SW pipelines so that we got full performance out of DSP's parallelism. Finally, we would craft "data flows" to manage pixel-intensive datatraffic between off-chip (DDR) and on-chip memories (L1, L2 buffers) through the Direct Memory Access. Our end results always yielded speedier, more agile algorithms much to the delight of our customers.
Please note that the content of the pages on this website is subject to Copyright but otherwise approved for public viewing.