Analysis and Noise Modeling of the Intel RealSense D435 for Mobile Robots

Ahn, Min Sung; Chae, Hosik; Noh, Donghun; Nam, Hyunwoo; Hong, Dennis

doi:10.1109/urai.2019.8768489

Cited by 49 publications

(32 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is known from other research that Intel RealSense depth sensor noise grows quadratically with distance [ 43 ], meaning that the farther the sensor is positioned from the curb, the lower the accuracy of the curb detection will be. On the other hand, our findings showed high accuracy up to 1.5 m away from the curb where the user could have the option to activate the CRN system ahead of time.…”

Section: Discussionmentioning

confidence: 99%

Automated Curb Recognition and Negotiation for Robotic Wheelchairs

Sivakanthan

Castagno

Candiotti

et al. 2021

Sensors

View full text Add to dashboard Cite

Common electric powered wheelchairs cannot safely negotiate architectural barriers (i.e., curbs) which could injure the user and damage the wheelchair. Robotic wheelchairs have been developed to address this issue; however, proper alignment performed by the user is needed prior to negotiating curbs. Users with physical and/or sensory impairments may find it challenging to negotiate such barriers. Hence, a Curb Recognition and Negotiation (CRN) system was developed to increase user’s speed and safety when negotiating a curb. This article describes the CRN system which combines an existing curb negotiation application of a mobility enhancement robot (MEBot) and a plane extraction algorithm called Polylidar3D to recognize curb characteristics and automatically approach and negotiate curbs. The accuracy and reliability of the CRN system were evaluated to detect an engineered curb with known height and 15 starting positions in controlled conditions. The CRN system successfully recognized curbs at 14 out of 15 starting positions and correctly determined the height and distance for the MEBot to travel towards the curb. While the MEBot curb alignment was 1.5 ± 4.4°, the curb ascending was executed safely. The findings provide support for the implementation of a robotic wheelchair to increase speed and reduce human error when negotiating curbs and improve accessibility.

show abstract

Section: Discussionmentioning

confidence: 99%

Automated Curb Recognition and Negotiation for Robotic Wheelchairs

Sivakanthan

Castagno

Candiotti

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…The D435i uses stereo infrared cameras to generate a depth map. Depth noise grows quadratically with distance, and empirical evidence indicates as much as four centimeter RMS error at a two meter distance [65]. The Intel RealSense SDK provides denoising post-processing filters including decimation (downsampling), spatial bilateral smoothing, temporal filtering, and depth thresholding [66].…”

Section: Rgbd Camerasmentioning

confidence: 99%

“…Each of the experimental result sections use different sensors resulting in dissimilar point cloud characteristics such as density, spatial distribution, noise, and accuracy to the ground truth surface. These characteristics have been studied both for Airborne Laser Scanning (ALS) technology [70,71], Velodyne LiDAR [72] and Intel RGBD [65] sensors. The airborne LiDAR point clouds used in Section 9.2.1 were captured in swathes with a Nominal Point Spacing (NPS) of 30 cm creating a semi-random distribution with less than 2.5 cm RMSE [60].…”

Section: Point Cloud Characteristics and Parameter Selectionmentioning

confidence: 99%

Polylidar3D-Fast Polygon Extraction from 3D Data

Castagno

Atkins

2020

Sensors

View full text Add to dashboard Cite

Flat surfaces captured by 3D point clouds are often used for localization, mapping, and modeling. Dense point cloud processing has high computation and memory costs making low-dimensional representations of flat surfaces such as polygons desirable. We present Polylidar3D, a non-convex polygon extraction algorithm which takes as input unorganized 3D point clouds (e.g., LiDAR data), organized point clouds (e.g., range images), or user-provided meshes. Non-convex polygons represent flat surfaces in an environment with interior cutouts representing obstacles or holes. The Polylidar3D front-end transforms input data into a half-edge triangular mesh. This representation provides a common level of abstraction for subsequent back-end processing. The Polylidar3D back-end is composed of four core algorithms: mesh smoothing, dominant plane normal estimation, planar segment extraction, and finally polygon extraction. Polylidar3D is shown to be quite fast, making use of CPU multi-threading and GPU acceleration when available. We demonstrate Polylidar3D’s versatility and speed with real-world datasets including aerial LiDAR point clouds for rooftop mapping, autonomous driving LiDAR point clouds for road surface detection, and RGBD cameras for indoor floor/wall detection. We also evaluate Polylidar3D on a challenging planar segmentation benchmark dataset. Results consistently show excellent speed and accuracy.

show abstract

“…The right-most RGB camera is used to collect visible-light signals. An active stereo sensor produces noise owing to the non-overlapping of image areas or lack of texture [8]; the presence of system noise also produces noise in the form of holes in the captured depth images. The quality of captured depth images is poor because of the holes.…”

Section: Introductionmentioning

confidence: 99%

Visual Positioning in Indoor Environments Using RGB-D Images and Improved Vector of Local Aggregated Descriptors

Zhang

Xia

Liu

et al. 2021

IJGI

View full text Add to dashboard Cite

Positioning information has become one of the most important information for processing and displaying on smart mobile devices. In this paper, we propose a visual positioning method using RGB-D image on smart mobile devices. Firstly, the pose of each image in the training set is calculated through feature extraction and description, image registration, and pose map optimization. Then, in the image retrieval stage, the training set and the query set are clustered to generate the vector of local aggregated descriptors (VLAD) description vector. In order to overcome the problem that the description vector loses the image color information and improve the retrieval accuracy under different lighting conditions, the opponent color information and depth information are added to the description vector for retrieval. Finally, using the point cloud corresponding to the retrieval result image and its pose, the pose of the retrieved image is calculated by perspective-n-point (PnP) method. The results of indoor scene positioning under different illumination conditions show that the proposed method not only improves the positioning accuracy compared with the original VLAD and ORB-SLAM2, but also has high computational efficiency.

show abstract

Analysis and Noise Modeling of the Intel RealSense D435 for Mobile Robots

Cited by 49 publications

References 10 publications

Automated Curb Recognition and Negotiation for Robotic Wheelchairs

Automated Curb Recognition and Negotiation for Robotic Wheelchairs

Polylidar3D-Fast Polygon Extraction from 3D Data

Visual Positioning in Indoor Environments Using RGB-D Images and Improved Vector of Local Aggregated Descriptors

Contact Info

Product

Resources

About