A Novel Visual Attention Framework using Unsupervised Feature Learning for Road Scene Understanding

摘要

Road scene understanding plays a key role in autonomous driving for intelligent vehicle. For the problem making semantic labeling with equivalent priority results in confliction between huge amounts of data and limited computation resource, this paper proposes a novel framework that efficiently fuses selective visual attention mechanism into the solution to scene perception task. Incorporating top-down and bottom-up two kinds of attention effect into an integrated Bayesian framework, total saliency map can be obtained taking use of implicit feature representation by unsupervised feature learning from natural images.