The KITTI Vision Benchmark Suite

Method

RoadNet3 [RoadNet3]

Submitted on 14 Feb. 2019 17:03 by
Yecheng Lyu (Worcester Polytechnic Institute)

Running time:		300 ms
Environment:		GPU @ GTX950M (Python +Tensorflow)

Method Description:

CNN+Distributed LSTM

Parameters:

Training Epoch=60, Optimizer=Adam, Learning
Rate=1e-4

Latex Bibtex:

@inproceedings{lyu2019road,
title={Road Segmentation using CNN and
Distributed LSTM},
author={Lyu, Yecheng and Bai, Lin and Huang,
Xinming},
booktitle={2019 IEEE International Symposium on
Circuits and Systems (ISCAS)},
pages={1--5},
year={2019},
organization={IEEE}
}

Evaluation in Bird's Eye View

Benchmark	MaxF	AP	PRE	REC	FPR	FNR
UM_ROAD	93.54 %	92.64 %	93.65 %	93.44 %	2.89 %	6.56 %
UMM_ROAD	95.88 %	95.46 %	96.37 %	95.40 %	3.95 %	4.60 %
UU_ROAD	92.95 %	91.93 %	93.32 %	92.58 %	2.16 %	7.42 %
URBAN_ROAD	94.44 %	93.45 %	94.69 %	94.18 %	2.91 %	5.82 %
UM_LANE	91.47 %	91.01 %	91.78 %	91.17 %	1.44 %	8.83 %

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
UM_LANE	99.18 %	99.21 %	99.07 %	98.39 %	97.23 %	95.57 %	94.57 %	83.72 %	80.26 %

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.

This figure as: png eps pdf

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