Detecting change in road environment via analysis of marked point processes associated with traffic signs

In the paper, logs of traffic signs - possibly recognized and recorded by an automatic traffic sign recognition system are analyzed to detect change in some aspect of the road environment along a route and to locate the change point also along the route between the different environments. The logs considered here keep a record of the locations and types of the traffic signs installed and detected along a route. The traffic sign logs are seen as realizations of marked binomial processes and the minimal description length (MDL) approach is used for detecting change in the road environment along a route. In particular, the change detection problem associated with driving from one topographical area to another is addressed here as a simple illustrative example. In order to cater for an efficient solution of this task and also for that of other road-environment change detection tasks the on-the-fly minimization method used in the Page-Hinkley change detector has been adopted. Simulation results in respect of traffic sign data generated for test purposes corroborate the expected behavior of the detector. In respect of real traffic sign data, a good qualitative agreement was found between the GPS-based altitude-profile of the data collection trip thresholded at some manually or automatically selected altitude after the trip and the MDL-based topographical segmentation of the route. For this segmentation, the traffic sign locations, more precisely the path-lengths corresponding to these locations measured from the starting point of the route along the route, and the corresponding traffic sign types were used as input. (C) 2017 The Authors. Published by Elsevier B.V.