The downselection of measurements used for free space determination in ADAS

Abstract

A parametric curve can be a compact representation of the free space boundary in automotive perception systems. In such an application it is usually obtained by an approximation of a boundary between free and occupied space on the occupancy grid. Existing algorithms which approximate such a boundary deal with a huge number of measurement points (grid cells) which have to be processed. Actually, many of those points are redundant and do not add any information. They can be rejected from further processing with no deterioration of the quality of the approximation, but decreasing the demand for processing power. In the paper we present several downselection algorithms which can reach such goal. All algorithms are compared by using common statistical metrics. The comparison of algorithms’ performance is done on the basis of dozens of logs presenting different road scenarios.