Code & Dataset for mesh-based scene estimation

This work is described in the paper "Real-Time Mesh-based Scene Estimation for Aerial Inspection", by Lucas Teixeira and Margarita Chli, published in the Proceedings of the IEEE/RSJ Conference on Intelligent Robots and Systems (IROS) 2016 [paper].

Video:

If you use this Code or Dataset, please cite the following publication:

 @inproceedings{Teixeira:etal:IROS2016,
title	    = {{Real-Time Mesh-based Scene Estimation for Aerial Inspection}},
author	    = {Lucas Teixeira and Margarita Chli},
booktitle	= {Proceedings of the {IEEE/RSJ} Conference on Intelligent Robots and Systems({IROS})},
year	    = {2016}
}

Code

The original code was developed inside OKVIS VI-SLAM system and can be found Here. This code is a standone version that can be integradeted to other SLAM system. We already sucessfully tested it with the ORB-SLAM output. A new topic has been added to publish a point cloud that is sampled over the mesh that is being constructed.

License

This code can be used for research purposes for free under a GPLv3 license, but we use two libraries that are not: Fade2D is a commercial software that can be used for research proposes for free (please check their website http://www.geom.at/), while Scratchapixel.com requests GPLv3 license. We will be working to remove this latter dependence in the future.

Environment

This code is tested for Ubuntu 14.04 and 16.06. If you want to run it on a different configuration you will have to change the pointer to the Fade2D library in the main CMakeLists. Note that the Fade2D binaries are included in this code.

ETHZ_V4RL-CAB Dataset

The dataset used in the paper can be found below

Examples

Data

Aerial 1 - Bagfile - Youtube

Aerial 2 - Bagfile - Youtube

Aerial 3 - Bagfile - Youtube

Ground - Bagfile - Youtube

Calibration

The images were captured using a VI-Sensor and calibrated using ETHZ ASL Kalibr. Below are the calibration parameters. Note that T_SC is the transformation from the Camera to the Sensor (IMU). The two sets of values correspond to camera0's intrinsics and camera1's intrinsics, respectively.

cameras:
    - {T_SC:     
        [ 0.9999921569165363, 0.003945890103835121, 0.0003406709575200133, -0.030976405894694664,        
         -0.003948017768440125, 0.9999711543561547, 0.0064887295612456805, 0.003944069243840622,         
         -0.00031505731688472255, -0.0064900236445916415, 0.9999788899431723, -0.016723945219020563,
         0.0, 0.0, 0.0, 1.0],
        image_dimension: [752, 480],
        distortion_coefficients: [0.0038403216668672986, 0.025065957244781098, -0.05227986912373674, 0.03635919730588422],
        distortion_type: equidistant,
        focal_length: [464.2604856754006, 463.0164764480498],
        principal_point: [372.2582270417875, 235.05442086962864]}
    - {T_SC:
        [ 0.9999722760516375, 0.007329193771005421, 0.0013153124248847282, 0.0790982900835488,
          -0.007335059837348439, 0.9999629194070246, 0.004511840884063492, 0.003549628903031918,
          -0.0012821954962168199, -0.00452136369336114, 0.9999889565615523, -0.01713313929463862,
          0.0, 0.0, 0.0, 1.0],
        image_dimension: [752, 480],
        distortion_coefficients: [0.00823121215582322, -0.015270152487108836, 0.03085334360639285, -0.017760720995454376],
        distortion_type: equidistant,
        focal_length: [456.3683366282091, 455.03924786357857],
        principal_point: [375.1783411236692, 238.22971133267725]}

Qualitative Results

You can reproduce the qualitative results of our paper using this bagfile and this ground-truth laser scan. You can also visualize the point-cloud inside the RViz software together with the laser scan of the facade. You only need to use rosbag play to run the bagfile. You can use the ros node read from the package ethz-asl:point_cloud_io.

Contact

For any questions or bug reports, please create an issue if you have questions or bug reports. Alternatively, you can also contact me at lteixeira@mavt.ethz.ch.