A new approach to outdoor scene description based on learning and top-down segmentation

摘要

This paper presents a new approach to outdoor scene description, by providing a methodology for labelling image objects in a top-down way. The system's strategy is based on a cooperative set of distributed tasks, composed of several segmentation processes devoted to the recognition of the objects of interest, and a coordinator process, used to control the segmentation tasks. The approach includes a learning process to generate flexible models that can fit with the relevant objects of outdoor scenes, which can vary significantly under different environment conditions. The results of a series of tests carried out under different weather and seasonal conditions demonstrate the feasibility of the approach.