Auto exposure solves flickering lights issue
Hello,
I was conducting an experiment with the D435i camera on using a really low exposure time (2-10) but with a bright light on. I recognized that there were dark bands on the image, most likely 50/100Hz flickers due to the electrical grid frequency and confirmed this by choosing different frame per second counts (6, 15, 30, 60) and got fewer bands for higher fps:s.
What confuses me is the connection between exposure and fps. When I select an exposure of, say, 5, I would expect this to be an exposure time of 500us. But if we only expose the sensor to light for 500us, then the shutter speed must surely be less than 500us in order to capture a complete image in this time? But the fact that we can see (what I presume to be) 100 Hz flickering would imply that it takes 1/fps (~10ms) time to capture an image. I want to be able to capture an image of a moving object without motion blur, so it is imperative that I can take the image in 1 ms or less.
Furthermore, if I turn on auto-exposure for 6 fps color footage, the dark bands disappear! I could not find any exposure setting for which the bands disappeared, and the only determining factor appeared to be the fps rate.
To summarize:
- What is the connection between exposure, fps and the rolling shutter speed?
- Why did auto-exposure remove the flickering artefacts? Did it find the perfect exposure setting or did it do anything beyond that?
-
Hi Oscar, the physical shutter speed on RealSense cameras is fixed, and so shutter speed is controlled indirectly with the exposure time value. There is a direct relationship between exposure and FPS. The mathematics of the relationship when using auto-exposure and manual exposure are discussed at the link below.
-
On camera models with a slower rolling shutter like D415 and the D435 types (D435, D435i, D435f, D435if), the RGB image can lag or blur when capturing motion faster than a waving hand compared to the faster global shutter on the depth sensors. Depth and RGB performance can be equalised by having auto-exposure enabled and running depth at 30 FPS and RGB at 60 FPS.
-
Whilst the exposure and FPS values can be used to compensate for interference caused by fluorescent lights such as overhead ceiling strip-lights, it is easier to negate that interference by configuring an RGB option called Power Line Frequency to 30, 50 or 60 depending on the operating frequency of the lights in the camera's location.
Fluorescent lights in particular cause interference because they contain heated gases that flicker at frequencies difficult to see with the human eye. Other types of lighting such as LED should cause less interference.
-
The frequency of the lights (and hence the Power Line Frequency that should be matched with it) typically depends on the country that the lights are in. North American regions use 60 Hz and European regions use 50 Hz. There is no harm though in trying 30, 50 and 60 for the Power Line Frequency to see which setting provides the best interference negation (if any).
If your lights' frequency is 50 Hz then that makes it impractical to attempt to use the FPS speed to reduce interference, since 50 is not a supported FPS speed for RealSense cameras and so a close match with the operating frequency of 50 Hz lights cannot be achieved.
You could test whether the High Speed Mode can negate the noise by selecting the resolution of your D435i camera to 848x100 and setting the FPs to 300. At 300 FPS the camera performs as though it is only set at 90 FPS because the 848x100 resolution limits the vertical height of the depth image to a 100 pixel high strip at the center of the screen and so there is less of a processing burden on the computer's CPU.
If you need a full-size depth image then another option that you could explore is using a Temporal post-processing filter and setting its 'Filter Smooth Alpha' setting to '0.1' instead of the default '0.4'. This can stabilize fluctuating depth values by updating the depth value less frequently.
Please sign in to leave a comment.
Comments
8 comments