Real-time monocular image-based path detection

Hybrid vision-based navigation for mobile robots in mixed indoor/outdoor environments

“…The path-following method used as motion primitive of the proposed approach is based on a previous work by De Cristóforis et al [15]. Here, a mobile robot is autonomously steered in order to remain inside a semi-structured path by means of image processing alone.…”

“…An example of such a method is a real-time monocular path following algorithm that allows a mobile robot navigate through semistructured or unstructured paths [15]. This method uses a probabilistic approach similar to the proposed by Dahlkamp et al [16], learning a visual model of the nearby road using an area in front of the vehicle as a reference.…”

“…The main contribution of this work is to propose a hybrid navigation method that can work successfully both in indoor and outdoor environments while overcoming the aforementioned drawbacks of existing methods, which are generally oriented towards a specific scenario. The proposed method combines a teach-and-replay feature-based method [7] and a segmentation-based approach for following paths [15]. Both methods have the same principle behind: to correct the heading of the robot to guide the autonomous navigation.…”

Visual road following using intrinsic images

“…Cristóforis et al [7] proposed a novel real-time imagebased monocular path detection that allows to detect delimited or semi-structured outdoor paths. In their method, the images are segmented into super-pixels and each super-pixel is classified in order to detect the navigable space and thereby to calculate the path contour, which is used to guide the robot.…”

“…Rather that that, we apply the intrinsic image for the problem of reactive visual-based road following, where the robot's steering and velocity is determined solely from its on-board camera image. Contrary to [7] and [8], which deal with the problem of shadows cast over the pathways by using separate color models for shadowed and illuminated path, our method uses the intrinsic image method to suppress the effect of shadows prior to the path detection / pixel classification step.…”