Automated Tree and Building Detection in Residential Areas from Stereo Ikonos Images

Stereoscopic analysis of target images is the primary and conventional method for photogrammetric 3D range data construction. For object detection in high resolution satellite images, it is advantageous if the processing method can exploit a stereo derived DEM so that the technique can be applied anywhere in the world. Central to this is the need for a very reliable stereo matching algorithm and an improved camera model to perform such an analysis. Object detection using a stereo DEM is described in this article, including optical image processing and data fusion techniques and how these were applied to the IKONOS image. Results are here presented for stereo IKONOS images over several residential areas which are occupied by dense woodland and small housing building unis at a rural location called Barton-Bendish in central England. Derived bare earth DTMs from DEM show < 3m rms height difference compared with OS® PANORAMA. Also, a quality assessment by visual inspection of object detection results shows reasonable accuracy. The tree detection process gives a higher detection ratio than 85% consistently and the building detection process attains up to a 80% detection ratio even with relatively small scale targets.