The Electronic Camera Gimbal

[CamNerd]

Short flight test video compilation testing a electronic pan tilt gimbal without moving parts. The data was transmitted using single frame JPEG compression and an IEEE802.11n Wi-Fi connection:
The Electronic Camera Gimbal

[jwatte]

So, basically, cropping out a sub-picture of a high-resolution fish-eye input?
Eliminating the cost and power draw of two servos might be worth it, and the cropped image may require less bandwidth to transmit than the full fish-eye view.
But don't you get much less light sensitivity when you use a high-resolution sensor, where each pixel is small?

[CamNerd]

"So, basically, cropping out a sub-picture of a high-resolution fish-eye input?"
Basically yes, the fisheye cam is calibrated to get out undistorted images.

"Eliminating the cost and power draw of two servos might be worth it, and the cropped image may require less bandwidth to transmit than the full fish-eye view"
Yes, for this use case you need the undistortion onboard. The other use case is a high capacity data-link (e.g. 801.11n) to transmit the whole fisheye image. You can do the undistortion on ground and in addition you will have the multiuser feature.

"But don't you get much less light sensitivity when you use a high-resolution sensor, where each pixel is small?"
You are right, we compensate it using a cam with a large sensor.

[bdeuell]

Cool idea, I imagine the improvement of camera technologies in recent years makes this a more practice approach.

Any idea how the weight and volume of this system, using the larger camera and fish-eye lens, compares to a system with a smaller camera having the equivalent resolution and light sensitivity of the cropped portion placed on a gimbal?

[jwatte]

Thanks for the answers!

What's the onboard computer? If it's something with a GPU, I imagine you can do anti-distortion using that pretty efficiently.

And, the $1M question: Is the additional cost of the larger sensor smaller than the saved cost of the servos you eliminate? Or is this mostly about weight savings?

[CamNerd]

Cool idea, I imagine the improvement of camera technologies in recent years makes this a more practice approach.

Absolutely right!

Any idea how the weight and volume of this system, using the larger camera and fish-eye lens, compares to a system with a smaller camera having the equivalent resolution and light sensitivity of the cropped portion placed on a gimbal?

The calculation for such a comparison is possible, but not easy and depends of the sensor pixel size, sensor noise and the aperture of your lens. Furthermore, the difference depends on the actual electronic gimbal position.
With operation during good light conditions, there is no quality difference between the 30gramms 1080p system and an analog PAL/NTSC camera with 45deg horizontal opening angle lens.  For smaller opening angles, the quality will suffer.

[CamNerd]

What's the onboard computer? If it's something with a GPU, I imagine you can do anti-distortion using that pretty efficiently.
And, the $1M question: Is the additional cost of the larger sensor smaller than the saved cost of the servos you eliminate? Or is this mostly about weight savings?

The undistortion can be done on a 1.6Ghz dual core atom board without the need for the GPU. We didn't use GPU calculations.
On the glider we used such an atom board, on the last plane we have a Core I7. The undistortion can be done there using one core with an utilization of about 50%.

The costc of the digital camera system, the computer board and the WiFi link we used is much higher than two additional servos and an analog link, but the multiuser feature, the low weight and the capability of forensic pan/tilt compensate for it.

[jwatte]

The undistortion can be done on a 1.6Ghz dual core atom board without the need for the GPU

That sounds like an order of magnitude more cost and power than necessary to me :-)

FWIW: The integrated mobile GPUs on devices such as the Raspberry Pi should be able to do un-warping at 60 Hz, at lower power, lower weight, and lower cost.

[CamNerd]

That sounds like an order of magnitude more cost and power than necessary to me :-)

FWIW: The integrated mobile GPUs on devices such as the Raspberry Pi should be able to do un-warping at 60 Hz, at lower power, lower weight, and lower cost.

For our use the time to demo was more important, but i think you are right, with enough implementation time and GPU acceleration it should be able to run on embedded platforms like the Raspberry. However the interface speed (USB...) could be a problem, the 3. prototype with the Ximea cam is able to produce more than 300MByte/sec.

[jwatte]

Clearly, what we need is a mobile phone chip with USB 3 support!
:-)

[mhobbies]

 You can do the undistortion on ground and in addition you will have the multiuser feature.

[jwatte]

You would  have to have a very high bandwidth link to be able to do undistortion on the ground.
Unfortunately, "wireless" and "high bandwidth" don't mix very well :-(