Method

TopNet-HighRes [la] [TopNet-HighRes]


Submitted on 7 Sep. 2018 13:17 by
Sascha Wirges (Karlsruhe Institute of Technology)

Running time:101ms
Environment:NVIDIA GeForce 1080 Ti (tensorflow-gpu)

Method Description:
Object detection by deep convolutional
networks
consistently adapted to the multi-layer
occupancy
grid domain. Only Velodyne used.
Parameters:
Meta architecture: Faster R-CNN;
Feature extractor: Resnet101;
Grid cell features: Intensity, min./max. z
coordinate, detections, observations;
Grid cell size: 10cm;
Box encoding: position, length, width,
sin(2*angle)/cos(2*angle);
Latex Bibtex:
@article{Wirges2018,
abstract = {A detailed environment perception
is
a crucial component of automated vehicles.
However, to deal with the amount of perceived
information, we also require segmentation
strategies. Based on a grid map environment
representation, well-suited for sensor fusion,
free-space estimation and machine learning, we
detect and classify objects using deep
convolutional neural networks. As input for
our
networks we use a multi-layer grid map
efficiently encoding 3D range sensor
information.
The inference output consists of a list of
rotated bounding boxes with associated
semantic
classes. We conduct extensive ablation
studies,
highlight important design considerations when
using grid maps and evaluate our models on the
KITTI Bird's Eye View benchmark. Qualitative
and
quantitative benchmark results show that we
achieve robust detection and state of the art
accuracy solely using top-view grid maps from
range sensor data.},
archivePrefix = {arXiv},
arxivId = {1805.08689},
author = {Wirges, Sascha and Fischer, Tom and
Frias, Jesus Balado and Stiller, Christoph},
eprint = {1805.08689},
month = {may},
title = {{Object Detection and Classification
in
Occupancy Grid Maps using Deep Convolutional
Networks}},
url = {http://arxiv.org/abs/1805.08689},
year = {2018}
}

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) 59.77 % 48.87 % 43.15 %
Car (3D Detection) 15.29 % 12.58 % 12.25 %
Car (Bird's Eye View) 67.53 % 53.71 % 46.54 %
Pedestrian (Detection) 22.98 % 17.57 % 17.35 %
Pedestrian (3D Detection) 13.45 % 9.66 % 9.64 %
Pedestrian (Bird's Eye View) 24.30 % 19.08 % 18.46 %
Cyclist (Detection) 29.34 % 19.15 % 19.69 %
Cyclist (3D Detection) 4.48 % 5.98 % 6.18 %
Cyclist (Bird's Eye View) 15.70 % 12.45 % 12.76 %
This table as LaTeX


2D object detection 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



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



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