Monodepth

Mono Depth ROS

• ROS node used to estimated depth from monocular RGB data.
• Should be used with Python 2.X and ROS
• The original code is at the repository Dense Depth Original Code
• High Quality Monocular Depth Estimation via Transfer Learning by Ibraheem Alhashim and Peter Wonka

<img src="https://raw.githubusercontent.com/tentone/monodepth/master/readme/c.png" width="370"><img src="https://raw.githubusercontent.com/tentone/monodepth/master/readme/d.png" width="370">

Configuration

• Topics subscribed by the ROS node
  • /image/camera_raw - Input image from camera (can be changed on the parameter topic_color)
• Topics published by the ROS node, containing depth and point cloud data generated.
  • /image/depth - Image message containing the depth image estimated (can be changed on the parameter topic_depth).
  • /pointcloud - Pointcloud2 message containing a estimated point cloud (can be changed on the parameter topic_pointcloud).
• Parameters that can be configurated
  • frame_id - TF Frame id to be published in the output messages.
  • debug - If set true a window with the output result if displayed.
  • min_depth, max_depth - Min and max depth values considered for scaling.
  • batch_size - Batch size used when predicting the depth image using the model provided.
  • model_file - Keras model file used, relative to the monodepth package.

Setup

• Install Python 2 and ROS dependencies

apt-get install python python-pip curl
pip install rosdep rospkg rosinstall_generator rosinstall wstool vcstools catkin_tools catkin_pkg

• Install project dependencies

pip install tensorflow keras pillow matplotlib scikit-learn scikit-image opencv-python pydot GraphViz tk

• Clone the project into your ROS workspace and download pretrained models

git clone https://github.com/tentone/monodepth.git
cd monodepth/models
curl –o nyu.h5 https://s3-eu-west-1.amazonaws.com/densedepth/nyu.h5

Launch

• Example ROS launch entry provided bellow, for easier integration into your already existing ROS launch pipeline.

<node pkg="monodepth" type="monodepth.py" name="monodepth" output="screen" respawn="true">
    <param name="topic_color" value="/camera/image_raw"/>
    <param name="topic_depth" value="/camera/depth"/>
</node>

Pretrained models

• Pre-trained keras models can be downloaded and placed in the /models folder from the following links:
  • NYU Depth V2 (165MB)
  • KITTI (165MB)

Datasets for training

• NYU Depth V2 (50K)
  • The NYU-Depth V2 data set is comprised of video sequences from a variety of indoor scenes as recorded by both the RGB and Depth cameras from the Microsoft Kinect.
  • Download dataset (4.1 GB)
• KITTI Dataset (80K)
  • Datasets captured by driving around the mid-size city of Karlsruhe, in rural areas and on highways. Up to 15 cars and 30 pedestrians are visible per image.