Skip to content

Latest commit

 

History

History
62 lines (55 loc) · 3.03 KB

README.md

File metadata and controls

62 lines (55 loc) · 3.03 KB
Raw

LIDAR

Fun with LIDAR and SLAM from scratch

Introduction

This repository contains work done for a Master's Degree capstone project in Robotics Engineering. The sensor used is the YDLIDAR X4. Documentation and details are available on the product webpage.

Dependencies

  • This project was written using ROS Kinetic.
  • The YDLIDAR ROS nodes are necessary to operate the sensor. They can be found at the product download page in the 'ROS.zip' file. There is also a helpful set of instructions for installing RVIZ, which is necessary for viewing LIDAR output, in the User Manual PDF.
  • The Point Cloud Library (PCL) is used in this project. PCL can be downloaded at http://www.pointclouds.org/downloads/.
  • Python dependencies include Scikit Learn for the Mean-Shift algorithm, Numpy, and the TF transformations library. These can be installed with Pip.

Setup

  1. Clone this repository into your existing Catkin workspace using Git
  2. Install the necessary dependencies
  3. Run catkin_make in the root of your Catkin workspace to build the project files

Usage

After plugging in the sensor, the following commands can be used to begin data collection, view the data, extract features using the Mean-Shift algorithm, and match point clouds.

Sensor startup

In a terminal, run:

$ roslaunch ydlidar x4_view.launch

This will open RVIZ and display the points seen by the sensor.

Feature server node

The feature server node re-publishes the latest LIDAR data message whenever the user presses 'Enter'. This is necessary to avoid overloading the Mean-Shift algorithm with too many points, since it takes about 10 seconds to extract features each time the callback is triggered. To start the feature server, run:

$ rosrun lidar_slam feature_serv.py
Mean-Shift algorithm node

To extract Mean-Shift features from incoming point clouds, run the node:

$ rosrun lidar_slam mean_shift_features.py

The features are published to the topic /corners, since Mean-Shift commonly identifies corners to be features. To view these features, subscribe to the /corners topic using RVIZ. The message type is PoseArray and the features will be represented as 2D vectors.

Point cloud matching algorithm node

To match new point clouds to ones that have been seen before, run the node:

$ rosrun lidar_slam lidar_slam

This runs the Iterative Closest Point algorithm on an incoming point cloud to determine if it matches one that has been seen before. The node keeps a database of all unique point clouds that have been seen before and returns the closest match to an incoming point cloud along with the homogeneous transformation matrix between the two point clouds. To view the old point cloud (the one that is the best match for the new point cloud), subscribe to the /pclmatch topic in RVIZ. The message type is PointCloud2 and making the points a different color will help distinguish them from the real-time LIDAR input.