Incorporating Plane-Sweep in Convolutional Neural Network Stereo Imaging for Road Surface Reconstruction

2020

PFG

Self Cite

This paper covers the problem of road surface reconstruction by stereo vision with cameras placed behind the windshield of a moving vehicle. An algorithm was developed that employs a plane-sweep approach and uses semi-global matching for optimization. Different similarity measures were evaluated for the task of matching pixels, namely mutual information, background subtraction by bilateral filtering, and Census. The chosen sweeping direction is the plane normal of the mean road surface. Since the cameras’ position in relation to the base plane is continuously changing due to the suspension of the vehicle, the search for the base plane was integrated into the stereo algorithm. Experiments were conducted for different types of pavement and different lighting conditions. Results are presented for the target application of road surface reconstruction, and they show high correspondence to laser scan reference measurements. The method handles motion blur well, and elevation maps are reconstructed on a millimeter-scale, while images are captured at driving speed.

show abstract

Section: Similarity Measurementioning

confidence: 99%

Road Surface Reconstruction by Stereo Vision

2020

PFG

Self Cite

show abstract

“…The distance between the position of matching pixels together with the geometric setup corresponds to the position of the corresponding object point in 3D space. In this work, the algorithm that we published in [1] is utilized for this purpose.…”

Section: Depth Extractionmentioning

confidence: 99%

Pavement distress detection by stereo vision / Straßenzustandserkennung durch stereoskopische Bildverarbeitung

Tm - Technisches Messen

2019

Self Cite

For road maintenance up-to-date information about road conditions is important. Such information is currently expensive to obtain. Specially equipped measuring vehicles have to perform surface scans of the road, and it is unclear how to automatically Ąnd faulty sections in these scans. This research solves the problem by stereo vision with cameras mounted behind the windshield of a moving vehicle so that the system can easily be integrated into a large number of vehicles. The stereo images are processed into a depth map of the road surface. In a second step, color images from the cameras are combined with the depth map and are classified by a convolutional neural network. It is shown that the developed system is able to Ąnd defects that require knowledge about surface deformations. These defects could not have been found on monocular images. The images are taken at usual driving speed.

show abstract

Deep learning self-calibration from planes

Twelfth International Conference on Machine Vision (ICMV 2019)

2020