[en] Image and video processing algorithms represent an efficient tool for features extraction, event detection and motion tracking methods. However, these algorithms are seriously hampered by their high consumption of both computing power and memory when processing large sets of high resolution images or videos. Therefore, we present an heterogeneous implementation of several image features extraction algorithms (edge and corner detection) which enabled to boost firstly performance of a medical application of vertebra segmentation, and secondly performance of an application of image and video indexation. We propose also an efficient GPU implementation of a motion tracking method, based on computing optical flow vectors of detected corners. Experimentations have been conducted using different sets of high reslution images and videos, showing a global speedup ranging from 10 to 50 by comparison with CPU implementations.
