Estimating road networks using archived GMTI data

It is increasingly accepted that accurate maps of road networks can make a critical difference in enabling accurate tracking of ground movers using GMTI radar data, especially when sensor resources are limited. However, road maps are often incomplete and inaccurate to such an extent that their utility is eliminated or greatly reduced. At the same time, users of GMTI data have noted in heavily trafficked areas that the road networks are readily apparent on positional displays of GMTI data. This has lead to the notion of estimating the road networks using GMTI data, an idea, which is operationally appealing given that the data, can be collected over a time period of several days to several months. This paper addresses one of the fundamental issues of estimating road networks from GMTI data. We derive a methodology for estimating a road network that views the road in a fundamentally different way than has been the case in previous approaches to this problem. The methodology is motivated by the stochastic models typically employed to model target trajectories as indexed by time, which we modify to come up with a stochastic model for the road trajectory which is indexed by arc-length. We apply this new method and compare it to a recently presented method that views the road as fundamentally composed of segments and vertices, and show using a limited data set that the stochastic estimation approach seems to offer much better performance.