find_best_camera_settings [Digitale Elektronik und Technische Informatik]

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Find best camera settings ======

As reflections are one of our major problems when taking images from the drone, its important to conduct some tests to find the best camera setting to reduce the reflections to a minimum. One major problem which bring reflections into the detection algorithms and human vision is that there is no contrast at these areas anymore. Its like a nonlinearity which limits the information around these areas.

Even if the image is darker afterwards, we can increase the brightness / black white adjust the image afterwards. The decrease in the colordynamics range is not so important for us as nonlinearitys. 

The informationdensity can be interpreted as entropy of a image. The higher the entropy of a image, the more information is in the image. As we are also interested in the informationdensity in the area of reflections or bright spots, we should also have a look at the entropy per area.

Tests including polarizationfilters will also be analyzed here.

===== Methodology =====

  * use pi camera v2
  * use pistill programm: https://www.raspberrypi.org/documentation/raspbian/applications/camera.md
  * test different settings regarding exposure : 
      * ex: exposure mode (auto, sports, beach)
      * ev: exposure compensation (-10 ... 10)
  * measurments
      * Shannon entropy: To check overall entropy
         * https://github.com/scikit-image/scikit-image/blob/master/skimage/measure/entropy.py 
      * Local Entropy: Area based entropy 
          * https://scikit-image.org/docs/dev/auto_examples/filters/plot_entropy.html
   * Testscripts: https://gitlab.com/searchwing/development/searchwing-pi/tree/master/tests/cameraSettingsTest

===== Tests =====

Setup:
{{ ::setup.jpg?600 |}}


==== Direct light ====
{{ ::0_img_ev-10_exauto.jpg.fig.jpg }} 
{{ ::1_img_ev-10_exsports.jpg.fig.jpg }} 
{{ ::2_img_ev-10_exbeach.jpg.fig.jpg }} 
{{ ::3_img_ev0_exauto.jpg.fig.jpg }} 
{{ ::4_img_ev0_exsports.jpg.fig.jpg }} 
{{ ::5_img_ev0_exbeach.jpg.fig.jpg }} 
{{ ::6_img_ev10_exauto.jpg.fig.jpg }} 
{{ ::7_img_ev10_exsports.jpg.fig.jpg }} 
{{ ::8_img_ev10_exbeach.jpg.fig.jpg }} 

==== Reflected light ====

{{ ::reflect_0_img_ev-10_exauto.jpg.fig.jpg }} 
{{ ::reflect_1_img_ev-10_exsports.jpg.fig.jpg }} 
{{ ::reflect_2_img_ev-10_exbeach.jpg.fig.jpg }} 
{{ ::reflect_3_img_ev0_exauto.jpg.fig.jpg }} 
{{ ::reflect_4_img_ev0_exsports.jpg.fig.jpg }} 
{{ ::reflect_5_img_ev0_exbeach.jpg.fig.jpg }} 
{{ ::reflect_6_img_ev10_exauto.jpg.fig.jpg }} 
{{ ::reflect_7_img_ev10_exsports.jpg.fig.jpg }} 
{{ ::reflect_8_img_ev10_exbeach.jpg.fig.jpg }} 



==== Polarization filtered reflected light ====

===== Results =====

Looking at the results, we cannot use the shannon entropy as a measure to get the value of information in the bright areas. Its more important to look at the local entropy to get a idea about the information in the bright areas.

Based on the current results, i would recommend to choose the EV -10 setting to reduce the exposure for areas with high brightness to get some information in these areas. Additionally i would recommend to choose either the beach or sports setting for the exposure control, as they resemble our usecase the best. Beach is good to tell the camera to not focus its brightness to the bright parts (https://www.adorama.com/alc/0008541/article/100-in-100-Part-II-Beach-Mode-or-Snow-Mode-or-BeachSnow-Mode). Sports might be usable due it forces short exposures due to huge movement of the target, or the other way around: our drone.