2020
DOI: 10.1109/tiv.2020.2966074
|View full text |Cite
|
Sign up to set email alerts
|

Ground Plane Polling for 6DoF Pose Estimation of Objects on the Road

Abstract: This paper introduces an approach to produce accurate 3D detection boxes for objects on the ground using single monocular images. We do so by merging 2D visual cues, 3D object dimensions, and ground plane constraints to produce boxes that are robust against small errors and incorrect predictions. First, we train a single-stage convolutional neural network (CNN) that produces multiple visual and geometric cues of interest: 2D bounding boxes, 2D keypoints of interest, coarse object orientations and object dimens… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
9
0

Year Published

2021
2021
2024
2024

Publication Types

Select...
4
3

Relationship

2
5

Authors

Journals

citations
Cited by 15 publications
(9 citation statements)
references
References 40 publications
0
9
0
Order By: Relevance
“…This procedure is illustrated in Figure 3a. Alternatively, this could be replaced by a purely vision based 3D detector like the one proposed in [37], where both the global pose and 2D bounding box of an object are obtained from the same algorithm. In cases where LiDAR sensors are available, an alternative is considered (shown in Figure 3b).…”
Section: Fusion Of Object Proposalsmentioning
confidence: 99%
“…This procedure is illustrated in Figure 3a. Alternatively, this could be replaced by a purely vision based 3D detector like the one proposed in [37], where both the global pose and 2D bounding box of an object are obtained from the same algorithm. In cases where LiDAR sensors are available, an alternative is considered (shown in Figure 3b).…”
Section: Fusion Of Object Proposalsmentioning
confidence: 99%
“…3 Distance via 2D/2.5D/3D constraint [158] , 3D-RCNN [127], Mono3D [31], ROI-10D [160], MonoGRNet [196], ApolloCar3D [236], 6D-VNet [268], GPP [199], RTM3D [141], [48] , [78], Deep3dBox [171], Shift R-CNN [174], GS3D [139], [277], [168], [102].…”
Section: Keypoints and Shapementioning
confidence: 99%
“…Krull [124] presented a model for 6D pose estimation, which 1180 applied a CNN to map images and revealed that training on a single object was sufficient and that CNN successfully generalized all the different objects and backgrounds of an image. Rangesh et al, [199] applied an exclusive idea for a 3D identification box suitable for the object on the ground 1185 to combine 2D visual context, 3D dimension and ground plane. Eppner et al, [59] presented and evaluated the winning system for the 2015 Amazon Picking Challenge, where they created four key aspects of system building: integration, manipulation, manipulation planning, and estimation.…”
Section: H Cnn/ Deep Learning -Based Approachesmentioning
confidence: 99%
See 1 more Smart Citation
“…Recently, applications related to autonomous and partially automated cars have attracted significant attention. For research in these applications, a full-surround multi-modal dataset with 2D and 3D annotations, as well as assigned track IDs would be very useful [1]- [3]. Motivated needs, our work focuses on the challenging task of 2D image and 3D point cloud annotation of street scenes.…”
Section: Introductionmentioning
confidence: 99%