Method

Frustum-PointPillars: A Multi-Stage Approach for 3D Object Detection using RGB Camera and LiDAR [Frustum-PointPillars]


Submitted on 29 Sep. 2021 15:42 by
Anshul Paigwar (INRIA)

Running time:0.06 s
Environment:4 cores @ 3.0 Ghz (Python)

Method Description:
Instead of solely relying on point cloud features, we leverage the mature field of 2D object detection to reduce the search space in the 3D space. Then, we use the Pillar Feature Encoding network for object localization in the reduced point cloud. We also propose a novel approach for masking point clouds to further improve the localization of objects. On the KITTI test set our method outperforms other multi-sensor SOTA approaches for 3D pedestrian localization (Bird's Eye View) while achieving a significantly faster runtime of 14 Hz.
Parameters:
\alpha=0.2
Latex Bibtex:
@inproceedings{paigwar:hal-03354114,
TITLE = {{Frustum-PointPillars: A Multi-Stage Approach for 3D Object Detection using RGB Camera and LiDAR}},
AUTHOR = {Paigwar, Anshul and Sierra-Gonzalez, David and Erkent, {\"O}zg{\"u}r and Laugier, Christian},
URL = {https://hal.archives-ouvertes.fr/hal-03354114},
BOOKTITLE = {{International Conference on Computer Vision, ICCV, Workshop on Autonomous Vehicle Vision}},
ADDRESS = {California, United States},
YEAR = {2021},
MONTH = Oct,
PDF = {https://hal.archives-ouvertes.fr/hal-03354114/file/Frustum_Pointpillars_ICCV.pdf},
HAL_ID = {hal-03354114},
HAL_VERSION = {v1},
}

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).


Benchmark Easy Moderate Hard
Pedestrian (Detection) 76.80 % 67.51 % 63.81 %
Pedestrian (Orientation) 49.04 % 42.97 % 40.69 %
Pedestrian (3D Detection) 51.22 % 42.89 % 39.28 %
Pedestrian (Bird's Eye View) 60.98 % 52.23 % 48.30 %
This table as LaTeX


2D object detection results.
This figure as: png eps pdf txt gnuplot



Orientation estimation results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker