This repository serves as a tutorial for training and deploying a 3D object detector using MATLAB® R2024b. It guides you through the processes of data processing, labeling, training, and deploying an object detection model. In this demonstration, simulated LiDAR data is collected from a vehicle navigating a parking lot.
To get started, clone this repo to your local machine and open MATLAB or use this button to open the repository in MATLAB Online:
The repository follows the workflow shown below:
This section covers data exporation and different ways one can preprocess LiDAR data. It gives an introduction to the Lidar Viewer App and explains the different exploration and processing option offered in the app. The file also contains scripts to do the same using MATLAB functions in a programmatic way.
This section introduces the Lidar Labeler App and walks through the steps to interactively label objects of interest using the app. For this demo, the focus is on labeling cars, construction barrels and cones. The labeled data is then exported to be used for training a deep learning network.
The third section goes into detail on training a point pillar network to detect the labeled objects in point clouds. This includes data augmentation, configuring the point pillar network, training and evaluating the results.
The final section demonstrates how the trained network can be deployed to a target system using MATLAB® Coder™. Two types of deployment are discussed:
- Static library
- ROS Nodes
This project requires the following products:
- MATLAB®
- Lidar Toolbox™
- Image Processing Toolbox™
- Computer Vision Toolbox™
- Deep Learning Toolbox™
- MATLAB® Coder™
- ROS Toolbox
For questions or clarifications on the code, please contact roboticsarena@mathworks.com
The license is available in the license.txt file in this GitHub repository.
Copyright 2024 The MathWorks, Inc.