Outdoor scene description systems entail many additional problems due to the scenes variability. Until now, most existing systems have not taken into account such variations, and are restricted to the descriptions of some predetermined images in specific outdoor conditions only (e.g., sunny images without shadows, only green trees, etc.). The main reason for this inability, is that such description systems are based on bottom-up strategies, which implies the use of general-purpose methods that are not able to deal with the large number of specific cases that outdoor scenes can exhibit.

To deal with this, we have proposed a top-down system specialized in recognising natural objects in outdoor scenes. Our system's strategy is based on a co-operative set of distributed tasks, composed of several segmentation processes devoted to the recognition of the objects of interest, and a coordinator process, used to control the segmentation tasks. The approach includes a learning process to generate flexible models that can fit with the relevant objects of outdoor scenes, which can vary significantly under different environment conditions. The results of a series of tests carried out under different weather and seasonal conditions demonstrate the feasibility of the approach.

Our current research focuses on the development of an hybrid system that takes into account the geometric features of the objects. Such challenge triggers new sub-problems that we would like to address in the near future:

• First, an hybrid strategy imply general purpose segmentation.
• We have to review the process of modelling incorporating shape features to the learning approach. Until now, we only take into account colour, texture and spatial features, but it is obvious that geometric features are indispensable to recognize many objects. Current feature selection and evaluation methods must be revised.
• We have to review the strategy itself. On the one hand, re-design a method to combine the information avalaible (color, texture, shape, and spatial disposition of models and regions) and on the other hand,  thinking in how to design methods to automatically recognize the objects of interest using the most appropiated features.

Related papers

• J.Batlle, A.Casals, J. Freixenet, and J.Martí. A Review On Strategies For Recognizing Natural Objects In Colour Images Of Outdoor Scenes. International Journal Image And Vision Computing. Elsevier Science. 18(6-7):515-530, May 2000.
• J.Freixenet. PhD. Thesis. Descripció d'Escenes Exteriors a partir d'un Aprenentatge Supervisat de les Característiques més Significatives dels Objectes. Supervisor: Dr. J. Martí. Presented in Girona, July 2000.
• J. Freixenet, J.Martí, X.Cufí, and X.Lladó. Use of Decision Trees in Colour Feature Selection. Application to Object Recognition in Outdoor Scenes. Proceedings of the IEEE International Conference On Image Processing (ICIP), vol. 3 pp. 496-499. Vancouver, Sept. 2000.
• J.Freixenet, J.Martí, X.Cufí, J.Batlle. Describing Outdoor Scenes By A Distributed Set Of Special Purpose Processes. Spanish Symposium On Pattern Recognition And Image Analysis (II), Bilbao, 1999, Pp: 113-114.
• X.Lladó, J.Freixenet, J.Martí, J.Forest, J.Salvi. Reconeixement D'objectes Naturals En Escenes D'exteriors Basat En Una Estratègia Bottom-Up. 2on Congrés Català d'Intel.Ligència Artificial, Girona, 1999, Pp: 299-307.
• M.Grabulosa, J.Freixenet, J.Martí, J.Batlle, X.Cufí. Goal-Directed Segmentation Of Outdoor Scenes Based On A Previous Learning Task. 2on Congrés Català d'Intel.Ligència Artificial, Girona, 1999, Pp: 308-315
• J.Freixenet, A.Casals, J.Batlle, and J.Martí. Recognising Natural Objects in Outdoor Scenes: A Survey. In Proc. of the 1st Congres Català d’intel.ligència Artificial, pp 290-297. Tarragona, Catalunya. Oct. 1998.
• J.Freixenet, J.Martí, J.Batlle. An Approach to the Urban Scene Understanding Problem Useful for Autonomous Navigation. In Proc. of the 3rd IFAC Symposium on Intelligent Autonomous Vehicles (IAV'98), pp. 718-723. Madrid, Spain. March 1998.
• J. Martí, A. Casals, J. Batlle. Model-Based Approach for Urban Scenes Description. Research reportIIiA 98-01RR. Institut d'Informàtica i Aplicacions. Universitat de Girona. January 1998.
• J. Martí, J. Batlle, A. Casals. Model-based Objects Recognition in Industrial Environments for Autonomous Vehicles Control. In proc. of the IEEE International Conference on Robotics and Automation (ICRA'97). Pp. 1632-1637. Albuquerque, NM, EUA. April 1997.
• J. Martí, A. Casals. Object Recognition using Multiple Cues. In proc. of the VII Simposium Nacional en Reconocimineto de Formas y Análisis de Imagen (SNRFAI). Pp. 97-102. Bellaterra, Barcelona. Abril 1997.
•  J. Martí, A. Casals. Model-based objects recognition in man-made environments. In proceedings of the 5th IEEE International Workshop on Robot and Human Communication (RO-MAN'96). Pp. 358-363. Tsukuba, Japan. Nov. 1996.
• J. Martí. Aportació a la descripció d'escenes urbanes mitjançant models aproximats. In proceedings of the 1er Seminari de Treball en Automàtica, Robòtica i Percepció. Pp. 253-260. Barcelona. Feb. 1996.