wrwheeler's profile - activity

I suppose this is more of a request than a question. I think OpenCV might find a new market of usefulness if it enabled an API to support operation within a Hadoop environment. For those of you who may not know, Hadoop is a distributed processing environment enabling massive computational parallelism to be supported within its framework. While is has been typically been used for data mining and data analytics applications (and growing rapidly in this area), it has recently been envisioned to employ its massive computational throughput in the area of video analytics. For example, performing complex video analysis techniques on hundreds of video sources simultaneously (which is the space I've been exploring).

It seems to me that OpenCV could easily support this technique by the addition of a simple API call: CvCapture * cvCaptureFromStream( std::istream input_stream, const std::string video_format_ext ) or VideoCapture::VideoCapture( std::istream input_stream, const std::string video_format_ext )

Instead of passing in a filename to read from, passing in an input stream which, in this case, would be derived from the Hadoop Distributed File System, would enable specialized Hadoop files to be processed instead of files defined on the local file system. This seems like a simple enhancement which would add considerable value and enable OpenCV to be very relevant in the area of massively parallel video analytics. Just a thought...