Method

[la] Real-Time 3D Vehicle Detection in LiDAR Point Cloud for Autonomous Driving [RT3D]


Submitted on 11 May. 2018 15:02 by
Qingli Guo (Duke University)

Running time:0.09 s
Environment:GPU @ 1.8Ghz

Method Description:
RoI generated from BEV
and pre-RoI-pooling conv are used to boost speed
Parameters:
lr = 0.001
Latex Bibtex:
@ARTICLE{8403277,
author={Y. Zeng and Y. Hu and S. Liu and J. Ye
and Y. Han and X. Li and N. Sun},
journal={IEEE Robotics and Automation Letters},
title={RT3D: Real-Time 3-D Vehicle Detection in
LiDAR Point Cloud for Autonomous Driving},
year={2018},
volume={3},
number={4},
pages={3434-3440},
keywords={collision
avoidance;convolution;feedforward neural
nets;object detection;optical radar;pose
estimation;road vehicles;RT3D;autonomous
driving;collision avoidance;path planning;LiDAR
sensors;LiDAR point cloud;LiDAR based 3D vehicle
detection;real time three dimensional vehicle
detection method;pre RoIpooling convolution
technique;pose sensitive feature map design;3D
object detection methods;KITTI benchmark
dataset;Three-dimensional
displays;Convolution;Proposals;Laser radar;Two
dimensional displays;Vehicle detection;Feature
extraction;Object detection;segmentation and
categorization;autonomous agents},
doi={10.1109/LRA.2018.2852843},
ISSN={2377-3766},
month={Oct},}

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) 50.33 % 39.69 % 40.04 %
Car (Orientation) 24.41 % 18.96 % 19.85 %
Car (3D Detection) 23.74 % 19.14 % 18.86 %
Car (Bird's Eye View) 56.44 % 44.00 % 42.34 %
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