Building Map Updating Based on Active Contour Models

The rapid growth and development of urban environments has been created a lot of motivation for researchers in Geomatics engineering in order to provide optimal methods for monitoring urban changes and updating maps. Nowadays, using aerial/satellite imagery for updating old maps is one of the important topics in Photogrammetry and Remote Sensing. In this case, the available information in the digital cartographic data can be used as training data for classification, creating conceptual model, reducing the search space and also to estimate the unknown parameters like segmentation parameters. Applying available information of cartographic data leads to contribution this type of information during the feature extraction in order to improve the efficiency and decrease the defects of this progress. Therefore, this paper proposes a novel approach for building extraction in order to building map updating from aerial images with help of old digital cartographic data. In this study, the geometric information of polygons existing in old cartographic data is used as an auxiliary data to improve the process of building extraction and change detection based on active contour models in a hierarchical approach. The building extraction process is done in two step using two types of active contour models which runs upon the height and spectral data. The active contour models in the face with large dimensions, high level of details and also images with weak gradient information have not acceptable performance. Therefore, the focus of this paper is to present a novel approach to compensate the above mentioned defect. So, the building extraction process is done in a hierarchical approach based on a combination of two geometric active contour models which causes the elimination of the defects of these models in the extraction of buildings with different geometric and spectral behaviors. In the proposed method, each of the polygons are considered as an initial curve to a region-based geometric active contour model. This model is run upon a part of the DSM, commensurate with the position and dimension of the old polygon. After primary extraction of the building boundaries from the DSM, geometric change detection is done and then, the change map is produced. The change map gives a comprehensive intuition about the occurred changes. Due to various errors in DSM, the primary extracted boundaries did not have sufficient accuracy. So, to improve the accuracy of these boundaries, the results are introduced to a constrained edge-based geometric active contour model which is one of the innovations of this study. The common edge-based active contour models cannot recognize the object boundary in images with weak gradient information and so, the level set function evolution well be unstable and therefore will not be getting correct results. To solve this problem, a novel approach as constrained level set formulation is proposed. This constraint is derived from output of the region-based model and is caused to solve the deficiency of this model in the face with weak gradient images. After the extraction of the precise boundaries, the MBR-based approach as an approximation or generalization technique is applied for irregular generated boundaries of changed buildings which are obtained from the proposed constrained edge-based model. Finally, the generalized polygons are record in geodatabase. The dataset used in this study concern the part of vaihingen city of Germany. The shape accuracy of extracted buildings and overall accuracy of change detection process were 92% and 78%, respectively. These results clearly demonstrated the success of the proposed method in building extraction using active contour models and change detection in order to automatic building map updating.