Curb Detection

[Esam Rosado]

Thesize of the data set is 109 MB, and it contains 50 preprocessed organized point clouds. Each point cloud is specified as a 64-by-1856-by-3 array. The ground truth data contains the semantic segmentation labels for 13 classes. The point clouds are stored in PCD format, and the ground truth data is stored in PNG format.

Based on the installed position of the lidar sensor, the point cloud data is sparse beyond a certain distance. To ensure the point cloud you extract is dense enough for further processing, specify an ROI within a limited distance of the sensor.

If semantic labels are not available with your data set, you can compute them using a deep learning network. For more information, see the Lidar Point Cloud Semantic Segmentation Using SqueezeSegV2 Deep Learning Network example.

The feature curb points may contain false positives. To remove the false positives, the function further processes the feature curb points using RANSAC based quadratic polynomial fitting to extract the candidate curb points.

Loop through and process the lidar data to extract and track the candidate curb points. Tracking curb points improves the robustness of curb detection. You can halt the curb tracking at the segmented roads, and restart it when the ego vehicle leaves the segmented roads. Curb tracking involves polynomial fitting on the XY-points using a 2-degree polynomial represented as y=ax2+bx+c, where a, b, and c are the polynomial parameters. Curb tracking is a two-step process:

To analyze the curb detection results, compare them against the tracked curb polynomials by plotting them in figures. Each plot compares the parameters with and without the Kalman filter. The first figure compares the drift of curbs along the y-axis, and the second figure compares the smoothness of the curb polynomials. Smoothness is the rate of change of the curb polynomial slope.

All articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer to

Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.

Abstract: Curbs are used as physical markers to delimit roads and to redirect traffic into multiple directions (e.g., islands and roundabouts). Detection of road curbs is a fundamental task for autonomous vehicle navigation in urban environments. Since almost two decades, solutions that use various types of sensors, including vision, Light Detection and Ranging (LiDAR) sensors, among others, have emerged to address the curb detection problem. This survey elaborates on the advances of road curb detection problems, a research field that has grown over the last two decades and continues to be the ground for new theoretical and applied developments. We identify the tasks involved in the road curb detection methods and their applications on autonomous vehicle navigation and advanced driver assistance system (ADAS). Finally, we present an analysis on the similarities and differences of the wide variety of contributions. Keywords: road curb detection; feature-based curb detection; classification-based curb detection

Reliable knowledge of road boundaries is critical for autonomous vehicle navigation. We propose a robust curb detection and filtering technique based on the fusion of camera semantics and dense lidar point clouds. The lidar point clouds are collected by fusing multiple lidars for robust feature detection. The camera semantics are based on a modified EfficientNet architecture which is trained with labeled data collected from onboard fisheye cameras. The point clouds are associated with the closest curb segment with L2-norm analysis after projecting into the image space with the fisheye model projection. Next, the selected points are clustered using unsupervised density-based spatial clustering to detect different curb regions. As new curb points are detected in consecutive frames they are associated with the existing curb clusters using temporal reachability constraints. If no reachability constraints are found a new curb cluster is formed from these new points. This ensures we can detect multiple curbs present in road segments consisting of multiple lanes if they are in the sensors' field of view. Finally, Delaunay filtering is applied for outlier removal and its performance is compared to traditional RANSAC-based filtering. An objective evaluation of the proposed solution is done using a high-definition map containing ground truth curb points obtained from a commercial map supplier. The proposed system has proven capable of detecting curbs of any orientation in complex urban road scenarios comprising straight roads, curved roads, and intersections with traffic isles.

Since precise calibration is needed for the localization and mapping tasks, in this thesis, methods for real-time calibration of the sensor setup is proposed. First, a method is proposed to calibrate sensors with non-overlapping field-of-view. The calibration quality is verified by mapping known features in the environment. Nevertheless, the verification process was not real-time and no observability analysis was performed which could give us an indicator of the analytical traceability of the trajectory required for motion-based online calibration. Hence, a new method is proposed where calibration and verification were performed in real-time by matching estimated sensor poses in real-time with observability analysis. Both of these methods relied on estimating the sensor poses using the state estimator developed in our earlier works. However, state estimators have inherent drifts and they are computationally intensive as well. Thus, another novel method is developed where the sensors could be calibrated in real-time without the need for any state estimation.

The datasets below can be used to train fine-tuned models for curb detection. You can explore each dataset in your browser using Roboflow and export the dataset into one of many formats.

At the bottom of this page, we have guides on how to train a model using the curb datasets below.

I want to draw a line on the curb. The purpose of drawing a straight line is to caculate the position of the line in the image, because when I do self-driving, I want my ROS car to drive along the curb.My country is right side driving.

Using classic Hough transform (HoughLines) gives a better result other than the the probabilistic Hough transform (HoughLinesP). I ran this code on your exact image and got these results. The threshold value for the first image is 100 and for the next image is 200

Curbs are one of the essential elements of urban and highway traffic environments. Robust curb detection provides road structure information for motion planning in an autonomous driving system. Commonly, video cameras and 3D LiDARs are mounted on autonomous vehicles for curb detection. However, camera-based methods suffer from challenging illumination conditions. During the long period of time before wide application of Deep Neural Network (DNN) with point clouds, LiDAR-based curb detection methods are based on hand-crafted features, which suffer from poor detection in some complex scenes. Recently, DNN-based dynamic object detection using LiDAR data has become prevalent, while few works pay attention to curb detection with a DNN approach due to lack of labeled data. A dataset with curb annotations or an efficient curb labeling approach, hence, is of high demand...

Abstract: For reducing the false detection and miss detection in the process of curb detection a novel curb detection and tracking method is proposed with 3D-LIDAR as sensor. Firstly the point cloud is preprocessed and a distance-based filter is used to filter the interference points in the original point cloud that affect feature extraction are filtered by a distance-based filter to enhance the extraction accuracy of curb points. For the filtered point cloud the ground segmentation method with ground plane segment-wise fitting is used to extract the ground point cloud. Then an adaptive multi-feature fusion algorithm for curb point extraction is designed by using the spatial features of curbs i.e. height difference smoothness and angle threshold. Next aiming at the problem of partial curb loss caused by obstacles the Rao-Blackwellized particle filter tracker is used to track and predict the curb points. Finally the method is applied to the multi-condition experiments of the unmanned sanitation vehicle and the results show that the method can accurately detect the road boundary information and effectively reduce the false detection and missing detection of curb points.

Mobile lidar is a remote sensing technique adopted by several transportation agencies to collect dense 3D point clouds with high accuracy and efficiency on a regular basis for a wide variety of applications (e.g., asset management, civil design, etc.). Leveraging mobile lidar technology can substantially improve the conventional procedure of asset management. However, there is still a need for automatic tools to process these 3D point clouds effectively and efficiently. In this study, the authors developed an automatic workflow to extract curbs and localize curb ramps in large point cloud datasets. The proposed method consists of three steps: Vo-SmoG ground filtering followed by refinement for preserving curbs, curb detection based on the sudden elevation change and linearity of the curbs, and curb ramp localization leveraging the context provided by the curb lines. It was evaluated both qualitatively and quantitatively with a representative mobile lidar dataset, resulting in recall, precision, and F-1 scores of 72.4%. Beyond extracting curbs and curb ramps, the proposed approach can be potentially used for further analyses such as feature characterization and point cloud classification for other assets and objects of interest.

3a8082e126