CNN-Based Object Segmentation in Urban LIDAR with Missing Points

Zelener, Allan; Stamos, Ioannis

doi:10.1109/3dv.2016.51

Cited by 17 publications

(21 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The study results performed well on the segmentation of cars, trees, pedestrians, and so forth [16]. CNN-based object segmentation for LiDAR data was conducted on the large-area LiDAR data of New York city and indicated good results even with missing points in vehicle object detection and segmentation [17]. In 2019, Shaoshuai Shi et al proposed the PointRCNN model that generate and detect objects used 3D point cloud data, This work conclude two stages of bottom-up 3D generation from the point cloud data and refines the 3D proposals in the canonical coordinate.…”

Section: Related Workmentioning

confidence: 81%

Multiple Object Detection Based on Clustering and Deep Learning Methods

Nguyen

Lee

Bae

et al. 2020

Sensors

View full text Add to dashboard Cite

Multiple object detection is challenging yet crucial in computer vision. In This study, owing to the negative effect of noise on multiple object detection, two clustering algorithms are used on both underwater sonar images and three-dimensional point cloud LiDAR data to study and improve the performance result. The outputs from using deep learning methods on both types of data are treated with K-Means clustering and density-based spatial clustering of applications with noise (DBSCAN) algorithms to remove outliers, detect and cluster meaningful data, and improve the result of multiple object detections. Results indicate the potential application of the proposed method in the fields of object detection, autonomous driving system, and so forth.

show abstract

Section: Related Workmentioning

confidence: 81%

Multiple Object Detection Based on Clustering and Deep Learning Methods

Nguyen

Lee

Bae

et al. 2020

Sensors

View full text Add to dashboard Cite

show abstract

“…The signed angle feature described in [ 24 ] measures the elevation of the vector formed by two consecutive points and indicates the convexity or concavity of three consecutive points. Input features converted from depth images of normalized depth (D), normalized relative height (H), angle with up-axis (A), signed angle (S), and missing mask (M) were used in [ 11 ]. We are using DHS in this work to project 3D depth image to 2D since as shown in [ 11 ] adding more channels did not affect classification accuracy significantly.…”

Section: Related Workmentioning

confidence: 99%

“…Input features converted from depth images of normalized depth (D), normalized relative height (H), angle with up-axis (A), signed angle (S), and missing mask (M) were used in [ 11 ]. We are using DHS in this work to project 3D depth image to 2D since as shown in [ 11 ] adding more channels did not affect classification accuracy significantly. Keeping the number of total channels to three, allow us to use networks with pre-trained weights for starting our training.…”

Section: Related Workmentioning

confidence: 99%

“…In the upper left we see a 2D RGB image, along with the 2D detected bounded boxes (a chair and a desk). On the upper right we see a 2D pseudo-RGB image that was generated from the associated 3D range image (see [ 11 ]), along with similarly detected 2D bounded boxes. We call this pseudo-RGB image a DHS image, where D stands for depth, H for height, and S for signed angle (see Section 1.2 ).…”

Section: 3d Object Detectionmentioning

confidence: 99%

“…We call this pseudo-RGB image a DHS image, where D stands for depth, H for height, and S for signed angle. The depth is a normalized distance of the associated 3D point, height is a normalized height of the 3D point, and the signed angle measures the elevation of the vector formed by two consecutive points on a scanline indicating the convexity or concavity of three consecutive points (see [ 11 ]). We can apply traditional 2D detectors on this pseudo-RGB image, making our method applicable even when no RGB information is available.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

3D Object Detection and Instance Segmentation from 3D Range and 2D Color Images

Shen

Stamos

2021

Sensors

Self Cite

View full text Add to dashboard Cite

Instance segmentation and object detection are significant problems in the fields of computer vision and robotics. We address those problems by proposing a novel object segmentation and detection system. First, we detect 2D objects based on RGB, depth only, or RGB-D images. A 3D convolutional-based system, named Frustum VoxNet, is proposed. This system generates frustums from 2D detection results, proposes 3D candidate voxelized images for each frustum, and uses a 3D convolutional neural network (CNN) based on these candidates voxelized images to perform the 3D instance segmentation and object detection. Results on the SUN RGB-D dataset show that our RGB-D-based system’s 3D inference is much faster than state-of-the-art methods, without a significant loss of accuracy. At the same time, we can provide segmentation and detection results using depth only images, with accuracy comparable to RGB-D-based systems. This is important since our methods can also work well in low lighting conditions, or with sensors that do not acquire RGB images. Finally, the use of segmentation as part of our pipeline increases detection accuracy, while providing at the same time 3D instance segmentation.

show abstract