This project is modified from the lanenet-lane-detection project(see: https://github.com/MaybeShewill-CV/lanenet-lane-detection)
Here I use lanenet-lane-detection project to implement a Ros node to do lane detection. Input and output parameters can be config at launch file.
Install the dependencies as lanenet-lane-detection metioned:
pip install -r requirements.txt
Then you need to install the ROS kinetic.
That's all.
In you catkin workspace src dir, clone the project:
git clone https://github.com/AbangLZU/LaneNetRos.git
download the pretrained weight(trained by @MaybeShewill-CV) in: https://www.dropbox.com/sh/tnsf0lw6psszvy4/AAA81r53jpUI3wLsRW6TiPCya?dl=0
copy these checkpoints files to the foler:
cd model
mkdir tusimple_lanenet
cd tusimple_lanenet
cp -r YOU_DOWNLOAD_FILES .
build and source (in the workspace)
catkin_make
sorce devel/setup.bash
you may need to change the mode of the python script, as follow:
cd lane_detector/scripts/
sudo chmod +x lanenet_node.py
Then you can launch the lane detector node with:
roslaunch lane_detector lanenet.launch
play one of you rosbag which contains the images, I use the KITTI dataset, which can be download at: http://www.cvlibs.net/datasets/kitti/raw_data.php?type=road
Download any drive data as you want (the synced+rectified data amd calibration), use tomas789's project kitti2bag (see: https://github.com/tomas789/kitti2bag) to convert it to a rosbag.
Use the following command to play the bag:
rosbag play kitti_2011_??????????.bag
Open your RQT to visualize the output, assign the image topic as you set in the launch file, and you should get this:
See more in this video:
It's almost realtime in my GTX1070. If you think this work is useful to you, please both star this repository and the lanenet-lane-detection repository. THX!
- Implement the Curve Fitting of the lane with a C++ node.
- Retrain the LaneNet, improve its performance
- TensorRT optimization