Road detection is one of the most important issues within the field of intelligent vehicles and driver assistance systems. This area has been an important research topic in the past two decades. Nowadays, many people lose their lives on road accidents; therefore, the use of intelligent systems to assist the drivers can significantly reduce the risks of driving. In addition, the successful navigation of robots in urban environments is strongly dependent on the identification of the road. In fact, for an autonomous robot moving from a starting point to a destination successfully, it is necessary to identify the road ahead and try not to deviate from the road. In this thesis, we are trying to detect roads using monocular vision technique. Using this method compared to other famous methods such as stereo imaging, LIDAR and Radar is much more economical. In this research, we try to present and evaluate five different methods to detect roads. In the first method, based on the assumption of differences over the color distribution of roads and background, weapply image segmentation using watershed algorithm improved by using the vanishing point. Therefore, shadow detection and removal from image in pre-processing stage are achieved. Also in second method, we cluster image pixels according to their color information and by defining the cluster which contains the road, a primary model for road will be achieved. To make the road detection method robust in different scenarios, we need to use differences in road and background textures in addition to color features. For this reason, in the third method, by using haralick texture features like contrast, entropy, etc and applying them to a learning system, we attempt to method, we use SFTA texture descriptor and justify; MARGIN: 0in 0in 0pt; unicode-bidi: embed; DIRECTION: ltr" As mentioned, each proposed method has some advantages and drawbacks, which distinguishes them from each other. Combining the outputs of these methods shows that appearance-based characteristics of image such as color and texture can successfully detect the road and increase accuracy. Another advantage is that this method can extract the road surface in different conditions, including straight or curved roads, suburban or urban environments and theexistence of shadows in the image. Keywords: Road detection, Texture descriptor, Hough transform, Vanishing point, Clustering
