Skip to Content

Computer Vision System for autonomous landing of VTOL RPAS

For the autonomous landing of VTOL RPAS, we use a computer vision system what is cheaper than a LIDAR and allow us to measure the pose of the helipad respect to the RPAS with accuracy enough.

We assume that the heliport is a common and most extended helipad with a white H surrounded by a white circle, painted on the gray heliport surface.

Fig 1: Heliport Marks

To detect the heliport, and to measure its pose with respect to the RPAS's pose, we propose to use a mono-camera computer vision system. We selected a mono-camera system instead of a stereo pair because we assume that the size of the square landing platform is known. As such, the 3-D reconstruction could be calculated using the platform model and the camera calibration parameters. This on-board color camera is pointing down.

As the helipad has no image descriptors enough (like SURF features), the detection and the tracking cannot be based on matching them with a previously known template. We have to use other features of the helipad, like the color or the marks (an H surrounded by a circle).

The computer vision algorithm has the following four main steps:

  • Image Acquisition and Preprocessing.

 In this step, the image is acquired and preprocessed both in color and in gray-scale.

Fig. 2: Example of acquired image 

 

  • Heliport Zone Extraction.

The heliport zone is exttracted using the color image. Then, this information is moved to the intensity image.

Fig. 3: Example of an Intensity image after heliport extraction

 

  • Helipad Marks Extraction.

Thank to this step, the marks of the helipad are segmented.

Fig. 4: Example of Circle selected blob Fig. 5: Example of a H selected blob

 

  • Heliport 3D Reconstruction.

This final step achieve the 3D reconstruction of the heliport using only visual information.

   
 Fig. 6: Example of Output Image after the computer vision algorithm. In green, the heliport. In red, the H corners (maximum of signature). In purple, the minimum of the H signature Fig. 7: Example of 3D reconstruction after the computer vision algorithm. The camera is fixed in the point (0, 0, 0), looking downwards

 

The computer vision system performance depends on different parameters:

  • The camera selected: the resolution is a key factor in the detection step, and also affect on the 3D reconstruction algorithm.
  • The lens selected: with the resolution of the camera, is the other key factor in the detection step and also in the 3D reconstruction.
  • The computer used affects to the framerate of the computer vision algorithm.

 

For our testing helipad and our selected camera-lens-computer, we achieve a very good performance in the detection:

 
Fig. 8: Example of the detection with the heliport tilted
 
Fig. 9: Example of the detection with the heliport far away
 
Fig. 10: Example of contamination on the helipad

 

Also we obtained a good performance in the 3D reconstruction:

Example 1: Heliport at a distance of around 6 times the size of the helipad. The heliport plane is parallel to the camera plane. This is one of the cases in which the 3D reconstruction works worst (very noisy).

 
Fig. 11: acquired image with the dectection and 3D reconstruction.

 

   
Fig. 12: Position x, y and z of the centre of the heliport respect to the camera when is stationary. Mean values: 0.008 m, -0.0229 m, and 1.2415 m. Standard deviations: 0.000460 m, 0.000425 m, and 0.0069 m. Fig. 13: Euler Angles Yaw, Pitch and Roll of the heliport respect to the camera when is stationary. Mean values: 161.6º, 162.9º, and 108.6º. Standard deviations: 11.60º, 3.90º, and 12.47º.

 

Example 2: Heliport at a distance of around 3 times the size of the helipad. The heliport plane is tilted respect to the camera plane. In this case the 3D reconstruction works better (less noisy).

 
Fig. 14: acquired image with the dectection and 3D reconstruction.

 

   
Fig. 15: Position x, y and z of the centre of the heliport respect to the camera when is stationary. Mean values: -0.0433 m, -0.0142 m, and 0.6857 m. Standard deviations: 0.000425 m, 0.000526 m, and 0.0016 m. Fig. 16: Euler Angles Yaw, Pitch and Roll of the heliport respect to the camera when is stationary. Mean values: 7.9º, 133.0º, and 72.8º. Standard deviations: 0.98º, 0.38º, and 1.22º.

 In this video, you can see the system working in real time: