## Method

3-Dimentional Deep-learning based on Elevation Patterns [la] [3D-DEEP]

Submitted on 5 Dec. 2019 11:58 by

 Running time: 0.03 s Environment: GPU @ 1.5 Ghz (Python)

 Method Description: Road segmentation through image and lidar data. LiDAR data are processed to obtain images with elevation patterns. Paper accepted for the Intelligent Vehicles Symposium 2020 Parameters: \lr=0.03125 \scheduler=polynomial \patience=3 \optimizer=ASGD Latex Bibtex:

## Evaluation in Bird's Eye View

 Benchmark MaxF AP PRE REC FPR FNR UM_ROAD 95.35 % 93.50 % 95.20 % 95.51 % 2.20 % 4.49 % UMM_ROAD 97.27 % 95.76 % 97.01 % 97.54 % 3.31 % 2.46 % UU_ROAD 94.67 % 93.04 % 94.23 % 95.12 % 1.90 % 4.88 % URBAN_ROAD 96.02 % 94.00 % 95.68 % 96.35 % 2.39 % 3.65 %
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

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

 This figure as: png eps pdf This figure as: png eps pdf This figure as: png eps pdf 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 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