Method

BirdNet: a 3D Object Detection Framework from LiDAR information [la] [BirdNet]


Submitted on 2 May. 2018 10:48 by
Carlos Guindel (Universidad Carlos III de Madrid)

Running time:0.11 s
Environment:Titan Xp GPU

Method Description:
We present a LiDAR-based 3D object detection pipeline entailing three stages. First, laser information is projected into a novel cell encoding for bird’s eye view projection. Later, both object location on the plane and its heading are estimated through a convolutional neural network originally designed for image processing. Finally, 3D oriented detections are computed in a post-processing phase. A paper has been submitted to the IEEE Intelligent Transportation Systems Conference 2018.
Parameters:
Resolution: 0.05 m
16 angle bins, without pool4, with ground.
Latex Bibtex:
@article{BirdNet2018,
author = {{Beltran}, J. and {Guindel}, C. and {Moreno}, F.~M. and {Cruzado}, D. and {Garcia}, F. and {de la Escalera}, A.},
title = "{BirdNet: a 3D Object Detection Framework from LiDAR information}",
journal = {arXiv preprint arXiv:1805.01195},
year = 2018,
month = may
}

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
Car (Detection) 78.18 % 57.47 % 56.66 %
Car (Orientation) 50.85 % 35.81 % 34.90 %
Car (3D Detection) 14.75 % 13.44 % 12.04 %
Car (Bird's Eye View) 75.52 % 50.81 % 50.00 %
Pedestrian (Detection) 36.83 % 30.90 % 29.93 %
Pedestrian (Orientation) 21.34 % 17.26 % 16.67 %
Pedestrian (3D Detection) 14.31 % 11.80 % 10.55 %
Pedestrian (Bird's Eye View) 26.07 % 21.35 % 19.96 %
Cyclist (Detection) 64.88 % 49.04 % 46.61 %
Cyclist (Orientation) 41.48 % 30.76 % 28.66 %
Cyclist (3D Detection) 18.35 % 12.43 % 11.88 %
Cyclist (Bird's Eye View) 38.93 % 27.18 % 25.51 %
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



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



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