Project for Udacity's Sensor Fusion Engineer Nanodegree Program

Project Goals

• Implement Obstacle detection on real PCD from a lidar
• use pcl-library for general data handling and initial testing
• implement following modules:
  • PCD filtering, for reducing computational cost, without loss of detail
  • Segment the filtered cloud into two parts, road and obstacles, using RANSAC based 3D-plane extraction
  • Cluster the obstacle cloud, using K-D Tree for 3D space.
  • Find bounding boxes for the clusters

The configuration I used

• Ubuntu 16.04 OS
• cmake >= 3.14
• gcc/g++ >= 8.0
• PCL >= 1.2 : The code extensively utilizes the Point Cloud Library (PCL).

• README.md: this file.
• images - folder with images for the readme-file
• ./src/
  • environment.cpp - main function
  • ransac.cpp - function for RANSAC-based segmentation implementation
  • cluster_kdtree.cpp & kdtree.h - functions for KD-Tree based clustering implementation
  • processPointClouds.cpp & processPointClouds.h - functions for point-cloud processing. functions that use segmentation and clustering based on PCL-library are also present, but commented
  • /quiz/... - contains quiz functions for testing ransac and clustering implementation
  • /render/... - contains rendering functions for display
  • /sensors/.. - contains point-cloud-data files and functions for use with synthetic data.

clone this repository, enter the cloned directory/folder and build:

mkdir build && cd build
cmake ..
make

to run, use following from within the build folder:

./environment

To test performance of 2D and 3D RANSAC implementation, build:

cd src/quiz/ransac/
mkdir build && cd build
cmake ..
make

to view result of 3D RANSAC implementation, launch

./qizRansac3d

a sample result is shown below, where road-plane in highlighted in green and objects on road in red.

3D RANSAC sample image:

To test performance of KD-Tree implementation on 2D sample points, build:

cd src/quiz/cluster/
mkdir build && cd build
cmake ..
make

to view result of 3D RANSAC implementation, launch

./qizCluster

a sample result is shown below.

2D KD-Tree implementation sample:

once ./environment is launch pcd data is read from files at /sensors/data/pcd/data_1/ and plotted after filtering-segmentation-clustering, as shown in sample image below.

sample lidar-obstacle-detection image:

• To install PCL, C++ https://larrylisky.com/2014/03/03/installing-pcl-on-ubuntu/