Fast color/texture segmentation for outdoor robots

Abstract

Abstract-We present a fast integrated approach for online segmentation of images for outdoor robots. A compact color and texture descriptor has been developed to describe local color and texture variations in an image. This descriptor is then used in a two-stage fast clustering framework using K-means to perform online segmentation of natural images. We present results of applying our descriptor for segmenting a synthetic image and compare it against other state-of-the-art descriptors. We also apply our segmentation algorithm to the task of detecting natural paths in outdoor images. The whole system has been demonstrated to work online alongside localization, 3D obstacle detection, and planning.

Detection of Nearby Obstacles with Monocular Vision for Earthmoving Operations

“…For this purpose, a rapid online segmentation method is proposed. Using the compact color and texture descriptor proposed by Blas et al [17], this study integrates an intensity feature to reduce the effects of lighting changes during the segmentation process. In addition, a two-stage unsupervised online learning process is proposed.…”

Integrating stereo structure for omnidirectional trail following

“…It is standard to first fit a ground plane to stereo data for obstacle detection [8], [9], but some of our data makes this step complicated. In areas with considerable height variation due to foliage, the ground region often occupies a minority of the image, breaking common techniques like RANSAC.…”

“…Along the lines of [3], a method to learn long-range obstacle appearance from short-range stereo labels was given in [8]. Among LAGRderived work, [9] and [10] stand out for explicitly looking for path-like corridors of homogeneous color or texture along the ground. The European ELROB competitions have also required path-following skills; one robot effectively followed paths by finding "passages" among scattered trees in ladar data [11].…”

“…This method does not assume that the trail color distribution is known a priori, and thus works on a wide range of trail types without training, and during tracking is quite robust to sudden changes in trail material, ambient illumination, or camera exposure. Briefly, following [9] we compute a small set of exemplar colors for each image using k-means clustering in CIE-Lab space and assign every pixel one of these k labels (this labeling is illustrated in Figure 2(b)). A label histogram is computed for each candidate region and its neighbors, and the likelihood is obtained as a weighted combination of contrast (as measured by the χ 2 distance between the region and its neighbors) and homogeneity (the entropy of the region color distribution).…”

Mobile robot vision navigation & localization using Gist and Saliency

“…Some systems rely on recognizing road edges (based on the intensity of images, for example) that separate the road from its surroundings [20]. Others do so by performing color and texture segmentation [21], [22]. In addition, there are systems that combine both techniques [23], [24], [25] to extract the road region in the image.…”