engleski

Towards automatic assessment and mapping of traffic infrastructure by adding vision capabilities to a geoinformation inventory

We consider geoinformation inventory systems containing locations and types of traffic signs and road surface markings in a given geographical region. Such systems allow a human operator to assess the current state of the traffic infrastructure by comparing a recent georeferenced video with the prescribed state stored in the inventory. We are concerned with automating this assessment procedure (as well as the initial creation of the inventory) by designing appropriate computer vision modules. In this paper, we briefly present our research project, provide a review of suitable computer vision techniques, and present in more details two techniques which proved most promising in preliminary experiments. The two presented techniques are traffic sign detection by boosted Haar wavelets, and enhancement of surface markings by applying steerable filters to suitably preprocessed images. The experiments have been performed on individual real images acquired from a moving vehicle by a single perspective camera. The obtained results are provided and discussed.

geoinformation inventory; computer vision; mapping and assessment of traffic infrastructure

nije evidentirano