RESOURCES OPTIMIZATION: ALGORITHM SWITCHING AND TEMPORAL ACTIVATION MAPS

摘要

The computer vision community is recently paying much attention on how to combine all the variety of image processing methods produced over the years in a working system. Consequently, architectural issues and system control are attracting growing interest. In this paper, we discuss two distinct improvements. The first is a contextual algorithm switching policy to increase operation robustness in a changing environment. We present the general framework and two sample applications to video surveillance and crossing pedestrian detection. The second contribution exploits the temporal correlation in image sequences acquired by a static camera to optimize computational resources management. We introduce a temporal activation map as a strategy to pursue an economy of means.