Automatic extraction of information for catenary scene analysis

This paper presents an original system for the automatic recognition of catenary elements. Based on a bottom-up approach, our analysis is composed of two stages: the first stage provides recognition hypothesis regarding previously segmented elements, with a precision rate of 91.7 %. The second stage allows to correct some hypothesis using a Markovian model which analyzes the catenary element sequence. It allows to improve precision up to 98.1%.