Awesome

Road Boundary Detection Using Deep Learning

This repository provides a pretrained RESA [1] network for MATLAB®, trained on the Comma10k [2] dataset for road boundary detection. The RESA network was originally proposed for lane detection, but this version has been trained to detect road boundaries extracted from the Comma10k dataset.

Creator: MathWorks Development

Includes un-trained model: :x:

Includes transfer learning script: :x:

Requirements

MATLAB® R2022b or later
Deep Learning Toolbox™
Computer Vision Toolbox™
Automated Driving Toolbox™

Getting Started

Download or clone this repository to your machine and open it in MATLAB®.

Setup

Add path to the source directory.

addpath(genpath('src'));

Download the pretrained network

Use the code below to download the pretrained network.

model = helper.downloadPretrainedRoadBoundaryDetection;
net = model.net;

Detect Road Boundaries Using Pretrained RESA model

Use to code below to perform detection on an example image using the pretrained model.

% Read test image.
img = imread(fullfile("images", "testImage.png"));

% Specify parameters for road boundary detection
params = helper.createDetectionParameters;

% Specify the executionEnvironment as either "cpu", "gpu", or "auto".
executionEnvironment = "auto";

% Call detectRoadBoundaries to detect the road boundaries.
roadBoundaries = detectRoadBoundaries(net, img, params, executionEnvironment);

% Visualize the detected boundaries.
fig = figure;
helper.plotBoundaries(fig, img, roadBoundaries);

Alternatively, you can also run the model on sample videos. Use the script roadBoundaryVideoDetection.m to run the inference on a driving scene.

Network Details

Recurrent Feature-Shift Aggregator (RESA) is one of the best performing lane detection architectures. It takes advantage of strong shape priors of lanes and captures spatial relationships of pixels across rows and columns. This repository leverages the RESA architecture for road boundary detection. The road boundary labels used for training were generated by labeling the Comma10k dataset.

References

[1] Zheng, Tu et al. "RESA: Recurrent Feature-Shift Aggregator for Lane Detection." AAAI (2021).

[2] https://github.com/commaai/comma10k