A selective vision algorithm based on aggregation of fuzzy evidence

摘要

A task-oriented algorithm that selectively processes image data is developed. The task of the algorithm is detecting/locating circular shaped road signs. Structural characteristics of the road scenes allow us to extract high level (contextual) information directly from the scene. This information is used as evidence for a road sign and aggregated with more evidence obtained by low level (pixel based) processing. The result is interpreted as an image of fuzzy singletons whose membership values quantify the likelihood for a road sign. This image is obtained by simple, computationally inexpensive operations and allows us to allocate our computational resources for the areas of the image that deserve most attention.