The KITTI Vision Benchmark Suite

Method

Map-Supervised Road Detection [MAP]

Submitted on 15 Jan. 2016 06:54 by
Ankit Laddha (Carnegie Mellon University)

Running time:		0.28s
Environment:		GPU

Method Description:

Generate training labels for pixels using OpenStreetMap. We project
the OpenStreet map data onto the image plane and the refine these
projected labels. We use these automatically generated labels to
train the Fully Convolutional Neural Network (FCN).

Parameters:

-

Latex Bibtex:

@INPROCEEDINGS{Laddha2016IV,
author={Ankit Laddha and Mehmet Kemal Kocamaz and Luis E. Navarro-Serment and Martial Hebert},
booktitle={IEEE Intelligent Vehicles Symposium Proceedings},
title={Map-Supervised Road Detection},
year={2016},
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	87.33 %	89.62 %	85.77 %	88.95 %	6.73 %	11.05 %
UMM_ROAD	89.97 %	92.14 %	87.47 %	92.62 %	14.58 %	7.38 %
UU_ROAD	84.44 %	87.17 %	83.66 %	85.23 %	5.42 %	14.77 %
URBAN_ROAD	87.80 %	89.96 %	86.01 %	89.66 %	8.04 %	10.34 %

This table as LaTeX

Behavior Evaluation

Benchmark	PRE-20	F1-20	HR-20	PRE-30	F1-30	HR-30	PRE-40	F1-40	HR-40

This table as LaTeX

Road/Lane Detection

The following plots show precision/recall curves for the bird's eye view evaluation.

This figure as: png eps pdf

Distance-dependent Behavior Evaluation

The following plots show the F1 score/Precision/Hitrate with respect to the longitudinal distance which has been used for evaluation.

Visualization of Results

The following images illustrate the performance of the method qualitatively on a couple of test images. We first show results in the perspective image, followed by evaluation in bird's eye view. Here, red denotes false negatives, blue areas correspond to false positives and green represents true positives.

This figure as: png