Method

Automatic Road Scene Labeling using Approximate Marginal Inference [ARSL-AMI]


Submitted on 15 Jul. 2014 21:43 by
Mario Passani (University of Alcala)

Running time:0.05 s
Environment:4 cores @ 2.5 Ghz (C/C++)

Method Description:
Approach based in a probabilistic graphical
model.
The method presents two stages: learning, based
on Conditional Random Fields and inference that
relies on Tree-Reweighted Belief Propagation.








Parameters:
rho = 0.5;
rez = 0.15
N_interations = 5
Latex Bibtex:
@inproceedings{Passani2014ITSC,
author = {Mario Passani and J. Javier Yebes
and Luis M. Bergasa},
title = {CRF-based semantic labeling in
miniaturized road scenes
},
booktitle = {Proc. {IEEE} Intelligent
Transportation Systems},
year = {2014},

}

Evaluation in Bird's Eye View


Benchmark MaxF AP PRE REC FPR FNR
UM_ROAD 71.97 % 61.04 % 78.03 % 66.79 % 8.57 % 33.21 %
UMM_ROAD 89.56 % 82.82 % 85.87 % 93.59 % 16.93 % 6.41 %
UU_ROAD 70.33 % 61.97 % 83.33 % 60.84 % 3.97 % 39.16 %
URBAN_ROAD 80.36 % 70.23 % 83.24 % 77.67 % 8.61 % 22.33 %
This table as LaTeX

Behavior Evaluation


This table as LaTeX

Road/Lane Detection

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


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

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png

This figure as: png


eXTReMe Tracker