Experimenting with Routes of Different Geometric Complexity in the Context of Urban Road Environment Detection from Traffic Sign Data

Abstract

Traffic sign data is used as input for detecting change in the type of urban road environment in which an ego-car is driven. The automatic urban road environment type detection is seen as a useful advanced driver assistance systems (ADAS) function that could be implemented in a straightforward manner relying on an existing ADAS function. Concretely, the traffic signs encountered along the route could be detected and logged by an on-board camera-based traffic sign recognition (TSR) system; and in order to perform the required function, it could be augmented with the change detection method described in the paper. Based on the analysis of the traffic sign and car trajectory data gathered for this preliminary study, the empirical distributions of the traffic signs along routes taken depend also on the geometrical complexity of these routes. A convenient model for describing traffic sign data along a route is a marked Poisson process. A traffic sign log is seen as a realization of such a process, and the minimum description length principle is harnessed to detect change in the road environment type. To test the applicability of this approach for urban routes of different complexity, a number of road environment transitions are looked at, while taking the route complexity into consideration. Finally, the detected changes are compared to ground truth.