Ricardo Mejía - Iñigo , María E . Barilla - Pérez and Héctor A . Montes - Venegas Universidad Autónoma del Estado de México México

Traffic flow is the study of interactions between vehicles, drivers and related infrastructure with the goal of understanding and developing an efficient road network. A number of devices have been implemented for monitoring traffic flow such as magnetic strips and radar detectors (Haidarian-Shahri et al. (2007)). All these devices have served for a very specific purpose. However, they do not provide as much information of the scene as cameras may do. The resulting information acquired from cameras should be properly processed to monitor vehicles’ behavior. Vision-based techniques have been increasingly used to extract useful traffic information from image sequences such as vehicle count, vehicle flow, vehicle speed, vehicle classification, and lane changes. All these tasks require an accurate vehicle tracking (Zhu et al. (2009)). Vehicle tracking is a technique that generates the trajectory of the vehicle over time by locating its position in every frame of an image sequence (Yilmaz (2006)). Usually, vehicle tracking requires three initial steps, namely, (a) choosing a suitable object representation, (b) selection of image features used as an input for the tracker and (c) vehicle detection. Vehicle detection is considered a difficult task, as an appropriate feature selection is needed to deal with the visual complexity of the road scene that may include congestions, shadows, weather incidences, headlights, road conditions and other traffic participants (Coifman et al. (1998)). Different approaches can be used to carry out the vehicle detection task, such as point detectors, background subtraction, segmentation or supervised learning methods. The selection of discriminative features plays a critical role because they should be so unique as to be easily distinguished in a feature space (Yilmaz (2006)). Texture is considered to be a rich source of information about the nature and the three-dimensional shape of objects. It is defined by Petrou (2006) and Materka & Strzelecki (1998) as complex visual patterns and subpatterns with singular properties such as brightness, color and size. There exist different approaches for texture modeling, such as statistical models, structural models and models based on transforms. Nevertheless, texture by itself is not enough to differentiate among different objects because contrasting surfaces cause false positives detections (Sun et al. (2006)). Color is considered a relevant feature when dealing with the perception of static and moving images. Visual contrast is useful to filter information present in each color component (Trémeau et al. (2008)) and to distinguish among similar gray-scale intensities (Barilla-Pérez & Spann (2008)). Color-based Texture Image Segmentation for Vehicle Detection 14