From 07d3f4462f44beefb392941566c2e507d54d1726 Mon Sep 17 00:00:00 2001
From: idelpino <telecosg@gmail.com>
Date: Thu, 18 Apr 2024 12:53:35 +0200
Subject: [PATCH] Update Readme.md
---
.gitmodules | 1 +
README.md | 33 +++++++++++++++++++++--------
kf_based_terrain_analysis | 2 +-
params/iri_ground_segmentation.yaml | 2 +-
4 files changed, 27 insertions(+), 11 deletions(-)
diff --git a/.gitmodules b/.gitmodules
index 40da8b2..d001770 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -1,3 +1,4 @@
[submodule "kf_based_terrain_analysis"]
path = kf_based_terrain_analysis
url = https://gitlab.iri.upc.edu/idelpino/kf_based_terrain_analysis
+ branch = main
diff --git a/README.md b/README.md
index 6f5abc7..223ec8f 100644
--- a/README.md
+++ b/README.md
@@ -1,21 +1,36 @@
-We are pleased to share the code repository associated with our paper, “Probabilistic Graph-based Real-Time Ground Segmentation for Urban Robotics.†This paper has been accepted for publication in the IEEE Transactions on Intelligent Vehicles. You can access the Early Access Version of the paper here: https://ieeexplore.ieee.org/document/10487036
+We are pleased to share the code repository associated with our paper, “Probabilistic Graph-based Real-Time Ground Segmentation for Urban Robotics.â€
+This paper has been accepted for publication in the IEEE Transactions on Intelligent Vehicles.
+
+You can access the Early Access Version of the paper here: https://ieeexplore.ieee.org/document/10487036
Currently, our team is diligently working on cleaning and documenting the repository. In just a few days, it will be fully prepared and ready for exploration!
---
-This package implements a ROS node for traversable ground / non-traversable ground / obstacle and overhanging obstacle segmentation. It provides the interface to the library kf_based_terrain_analysis, which angularly exploring a poincloud from the origin to the limits, generates a graph of ground references containing their 2D coordinates (XY) and three Gaussian variables (Z, slope in X (tan(pitch)) and slope in Y (tan(roll))) with means and variances. These three variables are estimated by means of a Kalman Filter.
+ROS Node for Ground Segmentation and Obstacle Detection
+
+This package contains a ROS node designed for ground segmentation and obstacle detection. It interfaces with the kf_based_terrain_analysis library, which generates a probabilistic ground model. Let’s dive into the details:
+Features:
-Dependences:
+ Ground Model Generation:
+ The ground model consists of nodes with 2D coordinates (XY) and three Gaussian variables: Z (height), slope in X (tan(pitch)), and slope in Y (tan(roll)).
+ These Gaussian variables are described by means and variances.
+ The model predicts a Gaussian distribution in the Z-axis for any desired XY coordinate, allowing us to treat obstacle detection as an outlier rejection problem.
-1) gnss_lidar_navigation_preprocessing (https://gitlab.iri.upc.edu/idelpino/gnss_lidar_navigation_preprocessing):
+ Traversable vs. Non-Traversable Ground:
+ To distinguish between traversable and non-traversable ground, we employ a shallow neural network.
+ The neural network operates on features extracted from the ground model.
+ We provide two pre-trained neural networks:
+ ROAD Detection: This network identifies ROAD areas (which are considered traversable).
+ TERRAIN and VEGETATION Classification: This network classifies TERRAIN and VEGETATION as non-traversable classes.
+ The neural networks were trained in Matlab, and their weights were exported in .csv files.
-Regarding the ground segmentation algorithm it provides:
-- A node to adapt the input point cloud to the format internally used by the kf_based_terrain_analysis.
-- And a node for fusing two pointclouds into one (for the ADD robot).
+Usage:
-2) the kf_based_terrain_analysis library (https://gitlab.iri.upc.edu/idelpino/kf_based_terrain_analysis)
+ Input Point Clouds:
+ The ROS node accepts two types of input point clouds:
+ XYZI Point Cloud (Topic: “lidar_pointsâ€): Use this for interfacing with LiDAR sensors.
+ XYZIRGBL Point Cloud: Use this for training. To generate this input, follow the instructions in the semantickitti2bag repository.
-To launch the whole ground segmentation pipeline for the ADD robot just roslaunch the iri_ground_segmentation add_ground_segmentation_pipeline.launch
diff --git a/kf_based_terrain_analysis b/kf_based_terrain_analysis
index c1be17b..61c1886 160000
--- a/kf_based_terrain_analysis
+++ b/kf_based_terrain_analysis
@@ -1 +1 @@
-Subproject commit c1be17b68e5a1af56cdb1a050b45374b1bce963c
+Subproject commit 61c1886b8227e98ec5315dfb2f8f05d011f982ff
diff --git a/params/iri_ground_segmentation.yaml b/params/iri_ground_segmentation.yaml
index 6403f9c..613c87a 100644
--- a/params/iri_ground_segmentation.yaml
+++ b/params/iri_ground_segmentation.yaml
@@ -37,7 +37,7 @@ iri_ground_segmentation: {
# Neural Network related parameters
use_neural_network: true,
- extract_data_to_external_training_of_the_network: false,
+ extract_data_to_external_training_of_the_network: true,
dataset_filename_with_global_path: '/home/idelpino/Documentos/dataset.csv', # This is the filename with path where the data for neural net training is dumped
neural_net_filename_with_global_path: '/home/idelpino/iri-lab/ground_segmentation_ws/src/iri_ground_segmentation/shallow_neural_nets/veg_terrain_and_obs_13_features_39_neurons.csv',
neural_net_number_of_features: 13,
--
GitLab