Visual road following using intrinsic images

Abstract

Abstract-We present a real-time visual-based road following method for mobile robots in outdoor environments. The approach combines an image processing method, that allows to retrieve illumination invariant images, with an efficient path following algorithm. The method allows a mobile robot to autonomously navigate along pathways of different types in adverse lighting conditions using monocular vision.To validate the proposed method, we have evaluated its ability to correctly determine boundaries of pathways in a challenging outdoor dataset. Moreover, the method's performance was tested on a mobile robotic platform that autonomously navigated long paths in urban parks. The experiments demonstrated that the mobile robot was able to identify outdoor pathways of different types and navigate through them despite the presence of shadows that significantly influenced the paths' appearance.

On the Impact of Illumination-Invariant Image Pre-transformation for Contemporary Automotive Semantic Scene Understanding

“…The use of illumination invariant approaches in most of the literature is predominantly for shadow removal [5], [12], [13], [8], [14], and to improve scene classification and segmentation [7], [6], [8]. Figure 2 shows an example of an RGB image (A) from [15] followed by four different illumination invariant images (B), (C), (D) and (E) generated using the approaches of [7], [6], [9], and [5]. In Section A we review recent approaches for illumination invariant image representation.…”

“…In Section A we review recent approaches for illumination invariant image representation. Section B introduces the illumination- RGB image [Krajník, Blažíček, and Santos, 2015] Fig. 2: An example of an RGB image (A) from KITTI dataset [15] followed by four different illumination invariant images, (B) [7], (C) [6], (D) [9], and (E) [5] where all the illumination variations such as shadows are significantly reduced within the scenes.…”

“…In order to improve mobile robot localization in outdoor environments, [9] calculates the illumination-invariant image I (see Fig. 2C) by projecting 3D log-chromaticity to a 2D log-chromaticity space ψ 1,2 using a 3D to 2D space transformation matrix U .…”

“…By contrast, here we evaluate the relative performance of each of these available set of invariant transforms, (IÁ lvarez [7], I Maddern [6], I Krajník [9], I Finlayson [5]), as a pretransformation prior to segmentation and semantic labeling via a contemporary scene understanding methodology [1]. Fig.…”

“…Within this context, despite the success of recent deep convolutional neural network (CNN) techniques to achieve state-of-the-art performance for generalized automotive scene understanding [1]- [4], extreme illumination variation can lead to inaccurate scene classification and segmentation [5]- [10]. To attempt to overcome the impact of illumination conditions and isolate the outdoor scenes from these effects, we evaluate the use of illumination invariant image pre-processing as proposed in [5], [6], [7], [9] to reduce illumination variations and improve object classification and segmentation for automotive scenes.…”