The customer, a supplier for surveillance solutions, was working on a product for high-resolution video surveillance for getting high-resolution video on large areas such as, e.g., grandstands in stadiums or large parking lots. To this end, they wanted to stitch together a large array of cameras to a single, very high-resolution, video streams.

I implemented for them the real-time decoding and stitching of these video streams on GPUs, as well as some basic image processing on the resulting stream, in CUDA.