Source-linked AI summary
On the Diagnostic of Road Pathway Visibility
Pierre Charbonnier, Jean-Philippe Tarel, Francois Goulette
TL;DR
Existing-road visibility assessment is limited by systems that generally use two vehicles to test fixed distances rather than estimate the maximum available distance. The paper develops single-vehicle stereovision and LIDAR approaches, finding that the 3D-based method agrees well with the reference while stereovision tends to underestimate visibility. Accurate 3D models support ray-tracing or z-buffering, but their acquisition and processing are costly.
Problem
Existing road networks have few systems for estimating maximum visibility distance; most vehicle-following systems only test whether a fixed prescribed distance is available.
Method
The paper estimates visibility using either paired onboard camera images with stereovision or a LIDAR Mobile Mapping System and 3D road models.
Results
The 3D-based method agrees well with the reference, whereas stereovision underestimates visibility over most of the tested section and is sensitive to vehicles and road-color conditions.
Takeaways & Limitations
With an accurate 3D road-and-surroundings model, ray-tracing or z-buffering suffices to estimate available visibility distance.
Takeaways & Limitations
LIDAR acquisition is costly and its point-cloud model must be simplified enough for long road sections without losing too much useful information.
Abstract
from arXiv · showhide
Visibility distance on the road pathway plays a significant role in road safety and in particular, has a clear impact on the choice of speed limits. Visibility distance is thus of importance for road engineers and authorities. While visibility distance criteria are routinely taken into account in road design, only a few systems exist for estimating it on existing road networks. Most existing systems comprise a target vehicle followed at a constant distance by an observer vehicle, which only allows to check if a given, fixed visibility distance is available. We propose two new approaches that allow estimating the maximum available visibility distance, involving only one vehicle and based on different sensor technologies, namely binocular stereovision and 3D range sensing (LIDAR). The first approach is based on the processing of two views taken by digital cameras onboard the diagnostic vehicle. The main stages of the process are: road segmentation, edge registration between the two views, road profile 3D reconstruction and finally, maximal road visibility distance estimation. The second approach involves the use of a Terrestrial LIDAR Mobile Mapping System. The triangulated 3D model of the road and its surroundings provided by the system is used to simulate targets at different distances, which allows estimating the maximum geometric visibility distance along the pathway. These approaches were developed in the context of the SARI-VIZIR PREDIT project. Both approaches are described, evaluated and compared. Their pros and cons with respect to vehicle following systems are also discussed.
1. Introduction
The paper assesses stop-on-obstacle visibility distance on existing roads by comparing required visibility with the maximum distance available from road geometry. It proposes single-vehicle stereovision and LIDAR-based approaches, trading lower cost against more complex but accurate 3D acquisition.
- Visibility-distance problem: Required visibility distance is the distance needed for a driver to react to a roadway obstacle and stop, using factors such as reaction time, grip, and speed.The paper uses conventional worst-case values, including a 2-second reaction time and wet-road grip.
- Visibility-distance problem: Available visibility distance is the maximum distance at which an object can be seen on the road, and it is compared with the required distance.The conventional speed value may be V85, the 85th centile of the speed distribution, modulated by road curvature and slope.
- Proposed approaches: The stereovision approach uses two onboard color cameras to account for the road’s 3D shape while keeping the acquisition system inexpensive.Image pairs can be acquired every 5 meters at normal vehicle speed, but uncontrolled illumination makes 3D reconstruction challenging.
- Proposed approaches: The LIDAR approach adapts the LARA-3D terrestrial Mobile Mapping System to create registered 3D road models using point sampling, GPS/INS, model reduction, and artefact suppression.The system provides directly acquired 3D data with high accuracy.
- Proposed approaches: Compared with stereovision, LIDAR requires a more complex and expensive acquisition system but directly captures accurate 3D data.This is the principal acquisition trade-off identified for the two approaches.
2. Visibility distance using binocular stereovision
The binocular stereovision method estimates available road visibility from paired camera images by segmenting the roadway, registering its edges, reconstructing the road surface, and measuring the farthest visible points.
- Visibility-distance definition: The available visibility distance is defined as the maximum distance of points belonging to the image of the road.This definition is used for the stereovision estimate.
- Road segmentation: Road segmentation classifies pixels in both stereo images as road or non-road using iterative learning of road and non-road colors.The process operates along the image sequence.
- Road segmentation: Batch processing can collect road colors from later images at different distances ahead, improving segmentation when illumination conditions vary.Real-time processing instead learns from past images.
- Edge registration and reconstruction: Differences in the pose of corresponding road edges between stereo views encode the shape of the road surface.The viewpoint change creates the geometric information needed for reconstruction.
- Edge registration and reconstruction: An iterative registration algorithm alternates between matching edge pixels and estimating a polynomial longitudinal road-profile model.The global registration estimates the parameters of the road model.
- Visibility-distance estimation: The estimated visibility distance is the maximum distance of registered road edges after reconstructing the road’s 3D surface.The method also computes the estimator’s standard deviation to qualify the result.
3. Visibility distance from a 3D model of the road and its surroundings
The 3D-model approach uses LIDAR-derived road data and simulated targets to estimate geometric visibility distance along the trajectory. Qt-Ballad supports both fixed-distance availability checks and maximum-distance estimation.
- LIDAR 3D data and the recorded GPS/INS trajectory provide the basis for estimating visibility distances.
- Visibility is defined geometrically by moving a target away from the observer until it becomes invisible, yielding the maximal visibility distance.
- A 1 m-high viewpoint and vehicle-shaped targets are evaluated using line-of-sight visibility, with a 5% visible-surface threshold for parallelepiped targets.
- Qt-Ballad: Qt-Ballad synchronizes point-cloud or mesh views, the trajectory, road images, and targets placed at different distances.
- Qt-Ballad: The interface displays required and estimated available visibility against curvilinear abscissa, revealing insufficient available visibility in the illustrated situation.
- Qt-Ballad: The software uses ray tracing for point targets and GPU-based Z-buffering with occlusion queries for volumetric targets.
- A fixed target distance produces a binary availability function, whereas moving the target outward records the maximum available distance at each observation position.
4. Discussion
The comparative study evaluates stereovision, a LIDAR-based 3D model, and vehicle following on the same road section. The 3D-based estimates agree with the reference, while stereovision generally underestimates visibility and vehicle following is limited in range and requires repeated runs.
- Three techniques—stereovision, a 3D model, and vehicle following—were compared on selected road sections, with VISULINE serving as the reference.
- Stereovision underestimates visibility across most of the tested section and is sensitive to vehicles and difficult long-range road segmentation.
- The 3D-based method agrees well with the vehicle-following reference method.
- 280 m is the vehicle-following system's limit, whereas maximal available visibility reaches about 400 m near abscissa 1150 m.
- Accurate 3D models enable standard ray-tracing or Z-buffering algorithms to estimate available visibility, but producing and processing those models is costly.
- Triangulated models must be simplified for long road sections without discarding too much useful information, while stereovision remains inexpensive and potentially accurate nearby.