Segmentation and Free Space Detection Using Gabor Filters Tadayoshi Shioyama 1 , Haiyuan Wu 2 , Masaya Takebe 1 , and Naoya Shimaoka 1 1 Department of Mechanical and System Engineering Kyoto Institute of Technology Matsugasaki, Sakyo-ku, Kyoto 606-8585, Japan shioyama@ipc.kit.ac.jp 2 Department of Computer and Commucation Sciences Wakayama University Abstract. This paper proposes a new segmentation method and applies it for detecting free space, which is defined as a free road area without any obstacle. The problem is important for an intelligent transportation control or collision avoidance in vehicles, and also important for assisting the blind pedestrian. In this paper, the texture analysis is carried out us- ing Gabor filters. Input images are segmented with not only the outputs of Gabor filters, but also both edge and color information. A free space is extracted from the segmented image. Then, the three-dimensional (3-D) information such as a free path at arbitrary direction from the observer is computed by stereo method. In order to evaluate the effectiveness of this proposed method, experimental results are shown for real images consisting of outdoor and indoor scenes. 1 Introduction This paper proposes a new method for segmentation and applies it to free space detection. A free space is defined as a road area without any obstacle. A detection of free space is important for an intelligent transportation control [1] or collision avoidance in vehicles, and also important for assisting the blind pedestrian. For the purpose of free space detection, the first important task is to segment input images. In previous works on segmentation, methods using optical flow vectors [2]–[8] have been proposed to detect moving vehicles, and a method using Ga- bor filters [10] has been proposed to segment an image into an object and a background. These methods have not aimed at detecting a free space. A method based on a texture analysis using a decay autocorrelation function [9] has been proposed to detect a free space. However, this has not obtained the 3-D infor- mation such as a free path from the observer to an obstacle, which is defined as the length without obstacle. In this paper, we propose a new method for segmentation using Gabor filters [10][11] and using a new devised criterion. The goal of this research is to develop a method for detecting the free space. Hence the segmentation method is applied to detecting the free space from stereo images. The area in an image corresponding J. Bigun and T. Gustavsson (Eds.): SCIA 2003, LNCS 2749, pp. 311–319, 2003. c  Springer-Verlag Berlin Heidelberg 2003