Multicore Computer Vision Library

Click here to apply for this internship.

Affiliated to

Movidia

Location

Belfast, UK

Timing

The proposed project duration is around 3 months and timing is flexible depending on student availability

Description

Movidius has developed a large body of custom video codecs for the Myriad multicore architecture. These codecs are highly optimised for performance and power at high frame-rates, and resolutions. Many interesting applications do not require this level of optimisation and Movidius is interested in having an intern develop a rapid prototyping environment based on PC hardware which could be used to evaluate computer-vision algorithms for subsequent porting to the Myriad architecture. A possible starting point would be an existing computer-vision library such as OpenCV (http://opencv.willowgarage.com/wiki/) for which a multicore port to the Cell processor already exists (http://cell.fixstars.com/opencv/index.php/OpenCV_on_the_Cell). An alternative would be a stream oriented library such as https://picoforge.int-evry.fr/cgi-bin/twiki/view/Gpucv/Web/WebHome, http://openvidia.sourceforge.net/index.php/OpenVIDIA or http://www.cs.ucf.edu/vision/public_html/MinGPU/
The work would involve a survey of existing libraries and implementation or porting of a subset of the library to use multiple cores which could later be ported to the Myriad architecture. Benchmarking would be performed on the x86 and Myriad ported code in order to predict the likely performance of x86 based codes when ported to Myriad. The programming language to be used would depend on the library but would be most likely C/C++
Knowledge of multicore processor architectures or computer vision algorithms is highly desireable

<< Go back to Industrial Internship Home Page