GIS Ostrava 2010 24. - 27. 1. 2010, Ostrava ___________________________________________________________________ A Multi-Agent strategy for automatic 3D object recognition based on the fusion of Lidar range and intensity data Farhad Samadzadegan, Fatemeh Tabib Mahmoudi Dept. of Geomatics, Faculty of Engineering, University of Tehran, P.O Box: 11365-4563, Tehran, Iran {samadz, fmahmoudi}@ut.ac.ir Abstract. Three dimensional object recognition and extraction from Lidar and other airborne or space borne data have been an area of major interest in photogrammetry for quite a long time. Therefore, many researchers have been trying to study and develop automatic or semi-automatic approaches for object extraction based on sensory data in urban areas. Lidar data have proved to be a promising data source for various mapping and 3D modeling of objects. But, according to the complicated relationships between objects in urban areas, especially buildings and trees, the performance of obtained results from ordinary object recognition algorithms based on Lidar data, is still dependent on several assumptions and simplifications. In this paper a multi-agent strategy has been proposed for automatic building and tree recognition based on the fusion of textural and spatial information extracted from Lidar range and intensity data. In this multi-agent methodology, two different groups of object recognition agents are defined for building and tree recognition in parallel and the algorithm has two different operational levels based on the types of contextual information. According to the difficulties in the field of building and tree detection based on the textural descriptors or spatial context, using the communicational behaviors and other capabilities of the multi-agent systems can be so useful in the field of 3D object recognition in dense urban areas. The evaluation of obtained results of the proposed methodology confirms the high capabilities of Lidar data and this multi-agent algorithm to decrease the conflicts in the field of building and tree recognition in parallel. Keywords: Multi-Agent, Lidar Data, Information Fusion, Spatial Descriptors, Textural Descriptors 1 Introduction The increasing demand for generation of three dimensional city models and their update, led to research efforts which aim to set up automatic tools for the extraction and recognition of 3D man-made and natural objects. The spectrum of application areas dealing with 3D city models is huge: environmental planning and monitoring, telecommunication, location based services, navigation, virtual reality, cadastre etc. Therefore, many researchers have been trying to study and develop automatic or semi-automatic approaches for object extraction in urban areas [1, 5, 13, 15]. Due to similarities between contextual information and complex relationships between objects in an urban area, the most complicated problems in the field of object recognition are related to the 3D objects such as buildings and trees. Light Detection and Ranging (Lidar) is an active remote sensing technology that directly provides 3D coordinates of objects on the surface. These coordinates can be converted to surface and terrain models by algorithms with a high degree of automation [11]. Lidar data has proved to be a promising data source for 3D object recognition and modeling. But, due to the difficulties in the case of detecting buildings and trees, the performance of obtained results from ordinary object recognition algorithms based on Lidar data is still dependent on several assumptions and simplifications. So, intelligent techniques such as multi-agent systems may be regarded as a powerful tool in order to facilitate 3D object recognition based on Lidar data. Intelligent Agent is a valuable concept in distributed artificial intelligence with strong abilities for solving complicated problems in different applications. Therefore, some of the researchers have proposed proper agent-based methodologies to solve the difficulties in the field of automatic object recognition. On the other side, agents cannot have complete information on their environment. Thus, to solve complicated problems or reach goals, agents must work with other agents. In a multi-agent model, several agents with specific goals and tasks are deployed, and they are trying to reach the main goal together [21]. Therefore, definition of proper social abilities and communication, coordination and negotiation between agents may improve overall results in solving complicated problems by multi- agent applications. In this paper, a proper multi-agent methodology has been proposed for building and tree recognition based on the fusion of textural and spatial context extracted from Lidar range and intensity data.