Cone Detection Using Deep Learning for Formula Student Driverless

This project presents a complete workflow for cone detection in Formula Student Driverless scenarios using deep learning. It demonstrates how to use MATLAB® and Simulink® for data preparation and labeling, YOLOX neural network design and training, and deployment to a GPU for real-time inference. We gratefully acknowledge the DIAN Racing Team at Tongji University, China, for providing the video datasets used in this demonstration.

Installation

This project has been tested on MATLAB® release R2024b and R2025a. Before getting started, ensure that the below MathWorks Products and Support Packages are installed and configured correctly in MATLAB®:

MathWorks Products

• Image Processing Toolbox™
• Computer Vision Toolbox™
• Deep Learning Toolbox™
• Parallel Computing Toolbox™
• MATLAB Coder™
• GPU Coder™

Support Packages

• The Computer Vision Toolbox™ Automated Visual Inspection Library
• MATLAB Coder Support Package for NVIDIA Jetson and NVIDIA DRIVE Platforms

Please see the Setup and Configuration for detailed setup and configuration instructions of the MATLAB Coder Support Package for NVIDIA Jetson and NVIDIA DRIVE Platforms .

Getting Started

Information about Getting Started

• Open the coneDetectionWithYOLOX.mlx live script in MATLAB®, and run the code section-by-section to understand the workflow. Please note: In Step 7, do not forget to change the NVIDIA Jetson™ GPU settings to your own, and also change the video name vidName in the coneDetection.m to the file location of your own test video.
• Open the coneDetectionWithObjectDetectorBlock.slx in Simulink® and update the File name of the "From Multimedia File" block and the File path of the "Deep learning Object Detector" block to your own file name or path.

License

The license is available in the license.txt file in this GitHub repository.

Community Support

MATLAB Central

Copyright 2025 The MathWorks, Inc.