SWIMMING POOLS LOCALIZATION IN COLOUR HIGH-RESOLUTION SATELLITE IMAGES C. Galindo, P. Moreno, J. Gonz´ alez, and V. Ar´ evalo Dept. of System Engineering and Automation University of M ´ alaga Campus Teatinos, 29071 M´ alaga, Spain ABSTRACT Detecting and localizing objects from space, like roads, rivers, lakes, etc., is a challenging task with multiple applications in remote sensing. In this paper we address the detection of swimming pools in colour high-resolution images of urban ar- eas acquired by the Quickbird satellite. The main motivation of this work is to survey and localize filled swimming pools during drought periods, fact that should be then punished by the local authorities. The proposed algorithm applies colour analysis for water detection and approximate segmentation, as an initial, rough localization, and active contours techniques to refine the pools’ shape. We have tested our algorithm in both satellite and aerial images with satisfactory results. Index Terms— Colour high-resolution satellite images, Quickbird, Swimming-Pools localization 1. INTRODUCTION The exploitation of satellite images is nowadays becoming relevant for ecological reasons, due to the climatic change, and it has being used for monitoring natural features like coastlines, river courses, forests, etc. In this line, one of the main problems that causes for concern in the last years is the control and management of limited and indispensable resources, as for instance potable water. Regions that un- dergo long periods of drought have to become aware about the moderate use of this valuable resource, striving for strictly controlling and punishing its waste. This problem is espe- cially significant in tourist areas where a high demand of water is required to be maintained, for example, for golf courses, swimming pools, waterparks, etc. This is the case of the South of Spain, where tourism is one of the major components of its economy and development, but at the same time the responsible for a great negative effect on the water c DigitalGlobe QuickBird imagery used in this study is distributed by Eurimage, SpA. (www.eurimage.com) and provided by Decasat Ingenieria S.L., M´ alaga, Spain. (www.decasat.com). This work has been partly sup- ported by the Spanish Government under research contract CICYT DPI- 2008-03527. reservoirs. An example of this situation can be observed in figure 4-a) where almost every plot of land entails a private swimming pool. Bearing in mind this problem, our work strives for developing techniques for the automatic detec- tion of open-air swimming pools which are filled up during drought periods. This would permit the local authorities to keep an inventory of the swimming pools located within its area, as well as, to impose fines to those who waste water. Detection and localization of land features from space is a challenging task which is being addressed and exploited in the last years due to the availability of high-resolution images ac- quired by satellites like Quickbird or Orbview. Remote sens- ing applications aim to facilitate (and insofar as it is possible, to automate) monitoring tasks on large areas of terrain. Some examples can be found in the literature for detecting and lo- cating human constructions, such as roads, buildings, sport fields, etc. (see [1] for a survey), plantations, e.g. of olive trees [2], and geographical features, like coastlines [3], lakes [4], or mountains [5]. Some related works to ours is [6, 7] that have proposed the classification of basic terrain classes, like water or green areas in satellite images, but to the best of our knowledge, the relevant issue of the accurately detection of filled swimming pools using high resolution satellite images has not been tackled in the literature. This paper approaches this issue and proposes an algorithm that automatically de- tects open-air swimming pools in Quickbird satellite images of urban areas. More precisely, it considers colour analysis for detecting areas in the image with a high probability of containing water. These areas are then refined through the use of active contours. The proposed method has been tested in both satellite and aerial images with satisfactory results. 2. PROBLEM STATEMENT The problem addressed in this work is to detect filled open-air swimming pools in colour high-resolution satellite images of urban areas. At a first glance, the automation of such a task seems to be almost trivial by searching for small rectangular blue areas in the image. Nevertheless, there are some thorny issues that turn it in a tricky problem: