179 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B2. Amsterdam 2000. AUTOMATIC CARTOGRAPHY FROM AERIAL IMAGES (SITE OF SALE, MOROCCO) O. El Kharki * , M. Sadgal * , A. Ait Ouahman * , A. El Himdy ** , M. Ait Belaid*** *Laboratory of Electronic and Instrumentation, Faculty of Science Semlalia, BOX 2390 Marrakech, Morocco. elkharki@yahoo.fr **Administration de la Conservation Foncière du Cadastre et de la Cartographie, Rabat, Morocco. ***Royal Centre for Remote Sensing, Rabat, Morocco. KEY WORDS: Maps, Aerial Photos, Neural Network, Image Processing, Pattern Recognition, Interpretation, Geographic Information System, Remote Sensing. ABSTRACT The maps or database of geographic objects (DBGO) are generally - except for some works [1,2,4,5] - create and update by topographers from measures executed on the lot. This manual update and creation of maps are today a laborious task, long, no exempt of errors and several years separate map versions. The goal of this paper is to present a new automatic process of creation and update maps from aerial images based on the utilisation of the image processing techniques (low and high levels) and neural networks. The comparison of the results with the Photopolis project[4] shows the performance of our approach. This work has two objects: - Create new maps. - Update the ancient maps. RESUME Les cartes ou base de données d'objets géographiques (BDOG) sont généralement (à part quelques travaux[1,2,4,5]) crées et mis à jour par des géomètres topographes à partir de mesures effectuées sur le terrain. Cette mise à jour manuelle est aujourd'hui très longue, plusieurs années séparent les différentes versions de cartes. Le but de cet article est de présenter un nouveau processus automatique d'élaboration de cartes à partir des photographies aériennes. Notre approche est basée sur l'utilisation des techniques de traitement d'images bas et haut niveaux et réseaux de neurones. La comparaison des résultats avec ceux du projet PHOTOPOLIS [4] montre la performance de notre approche. Ce travail a un double intérêt: - La création des nouveaux cartes pour des nouvelles villes. - La mise à jour des anciennes cartes. 1 INTRODUCTION The central theme of our project is the development of powerful tools for cartographic applications. In particular, image processing methods will be developed that delineate and detect man-made structures mainly buildings in aerial imagery. The main emphasis will be on digital urban mapping. The tools will allow greatly reducing the requirement for expensive manual labour. Our approach consists of comparing two sets of information in order to detect the divergences. The first information is a real representation (an aerial image, Figure.2) ,the second is a symbolic representation (maps, city plan). These informations are differents. In the purpose to remedy to this problem, we segment the aerial images by a split and merge segmentation method in order to obtain comparable data. The matching step between objects in the segment image and objets in map is realised by a neural network. We use the following criterion: the geometric shape, the gray level and textural parameters. The map is used in training phase. The interpretation results gives information on buildings, roads and so on, built or removed. Our project is organized in the following steps (Figure 1.): - Aerial images acquisition. - geometric and relief corrections - Aerial images filtering and segmentation - Matching objects by neural network - Updating ancient map. Omar El Kharki