Camera Stabilization.

[cmmguy]

What is the goal for camera stabilization?

Is it to simply maintain a fixed vector relative to space (independent of the changing orientation of the platform)?

or

Is it to maintain a changing vector that tracks a fixed point in space (also independent of the platform)?

I guess it seems that both would be reasonable expectations. The first used to stay on a moving target such as following a racing object and the second used to stay on a fixed target while orbiting or moving about it.

What is the current state of art for camera control?

 

[llbr22]

Here's how I use it: for my Hex to fly forward, it needs to lean forward. For my hex to fly sideways, it needs to lean sideways. But my camera needs to keep the horizon level. Some people think it should neutralize all the movement caused from the multirotor's movement. I would like that too, but it's easier said than done.
Here's a basic example: forward flight followed by a hard right then backwards cyclic. My camera stays relatively level to the horizon, while my multi is banking about 30 degrees either direction:
<a href="http://youtu.be/iIw0qPwcj2Y" target="_blank">

 

[DennyR]

What is the goal for camera stabilization?

Is it to simply maintain a fixed vector relative to space (independent of the changing orientation of the platform)?

or

Is it to maintain a changing vector that tracks a fixed point in space (also independent of the platform)?

I guess it seems that both would be reasonable expectations. The first used to stay on a moving target such as following a racing object and the second used to stay on a fixed target while orbiting or moving about it.

What is the current state of art for camera control?

You can only track a fixed point or a moving point with additional components, GPS, optic flow etc. imu stabilisation works by correcting the disturbance relative to it's own fixed point in space.

 

[cmmguy]

You can only track a fixed point or a moving point with additional components, GPS, optic flow etc. imu stabilisation works by correcting the disturbance relative to it's own fixed point in space.

Thank you for your reply.

So most camera stabilization maintain camera orientation only, ie, correcting for only pitch, yaw and roll?

Are there systems that correct for target, ie, keep the camera pointed at a fixed point regardless of platform pitch, yaw, roll and location.

Would this be a desirable capability?

 

[RTRyder]

Thank you for your reply.

So most camera stabilization maintain camera orientation only, ie, correcting for only pitch, yaw and roll?

Are there systems that correct for target, ie, keep the camera pointed at a fixed point regardless of platform pitch, yaw, roll and location.

Would this be a desirable capability?

Yes, MikroKopter can do it a couple of different ways. One is done through software and they call it point of interest. Designating the spot as a GPS coordinate and setting up other parameters similar to waypoint flight will keep the camera orientation fixed on that spot as the MK flys. The other option is called follow me and it relys on a remote GPS unit, same as the one the MK flight controller uses, to be attached to the object you want to track. Both work fairly well once you get past the inital setup and understand how to make them work, no other hobby grade flight control system that I know of has capability to match it at the moment. Part of the problem is that it requires a solid GPS navigation system to make it work and only a couple of controllers have good enough GPS capability to make it feasible, most have a hard enough time just trying to figure out where the multi is and keep it within 10 or 15 meters of that spot.

Ken

 

[DennyR]

Thank you for your reply.

So most camera stabilization maintain camera orientation only, ie, correcting for only pitch, yaw and roll?

Are there systems that correct for target, ie, keep the camera pointed at a fixed point regardless of platform pitch, yaw, roll and location.

Would this be a desirable capability?

The vast majority of camera imu's that are sold for model use are compromised by the fact that almost all of the commercially available camera mounts were created without any thought as to how they would be stabilized or how they would manage the inertial loads. That aspect was simply left to the buyer to fathom out. The result of this scenario is that we have a few lame attempts to create a jack of all trades device that must try and cope with many different methods to power the movements along the desired axis. Most use gyros that have a full scale range that is no better than that which the model itself uses. Some mounts have high reduction gears and some use servos in direct drive. Some have very high inertial loads and others less so. What is even more ridiculous is that some try to use the imu on different parts of the camera mount ie. one axis out on the base plate etc. Some will use the servos in feed forward and also in feedback. Proportional and slew. The object of the exercise is to measure the disturbance between what the model does and where level is, so that should tell you where it needs to be mounted. And that point is not where the opposite reaction force of the servo can be seen. Compounded by soft mounting.
About two years ago I had this problem resolved and have since that time played around with various low cost ways to replicate the system. I am pleased to announce that I have joined forces with a team of people in the US who have both the manufacturing capacity to reliably produce such a device and have an extraordinary group of dedicated dev. guys who will now offer this device hopefully in the next few months.
Camera stability actually starts with the model itself as that controls the level at which the camera imu operates. It is rather like a self noise canceling headset where the reverse of the motion wave is fed back into the system to create a straight line. Direct drive high speed motors form the essential part of keeping the latency as minimal as possible.
On the subject of the model stability here is a preview of the ROWLAND F1-535 showing the one piece autoclaved carbon base frame. each motor can lift around 2.5 kilos but will probably never need to do that as weight saving is the key to every part of this dedicated system. Camera mount not show for obvious reasons.
View attachment 2018 Shown alongside a flame wheel to compare the size.

 

[matwelli]

cant wait to see that mount!

 

[jffry7]

View attachment 3212 Shown alongside a flame wheel to compare the size.

Great Job
Thats one quad is what I have been looking for a long time, protected blades and single piece chassis (ok maybe term is wrong). I hope the price will not be as crazy as others but I might still be tempted to buy whatever the price is.

 

[DennyR]

jffry7
I'll try to keep you informed as thing progress. It does have some quite unique ways to get the job done. Hand built unfortunately but a lot less expensive than a big OKTO. Having seen the problems that some have had by releasing stuff before it is full tested and 100% reliable it wont be on offer for a few weeks I guess. BTW never had a broken prop with one yet. My biggest relief was not blowing myself-up with the home made autoclave that pressurizes the molding at 70 psi and 275 degrees.
D.

 

[jffry7]

D, sound scary molding that frame. the protected prop purpose for me is a protection to those around me in case an accident happen oh and of course moving around tight spaces with confidence. Glad you posted this, means I have to hold off my purchase of next frame Thanks for keeping me informed.

I didnt mean to hijack the thread but I think Denny is onto something video stab starts with a solid frame. but going to your question i think both are valid but depends on what application you want. I remember a gimbal stab for heli if not mistaken the CSM that track/lock to what you put it on but most cam stab built in to FC are already doing it. The latter you mentioned was attempted by Arducopter flight mode Circle but never had luck with it. I tried it and while it will always focus and the center location, orbiting around it fails. but ultimately I think it all depends on the needs

 

[cmmguy]

Yes, MikroKopter can do it a couple of different ways. One is done through software and they call it point of interest. Designating the spot as a GPS coordinate and setting up other parameters similar to waypoint flight will keep the camera orientation fixed on that spot as the MK flys. The other option is called follow me and it relys on a remote GPS unit, same as the one the MK flight controller uses, to be attached to the object you want to track. Both work fairly well once you get past the inital setup and understand how to make them work, no other hobby grade flight control system that I know of has capability to match it at the moment. Part of the problem is that it requires a solid GPS navigation system to make it work and only a couple of controllers have good enough GPS capability to make it feasible, most have a hard enough time just trying to figure out where the multi is and keep it within 10 or 15 meters of that spot.

Ken

Thanks for that explanation. Is this a capability of the OpenPilot products?

The vast majority of camera imu's that are sold for model use are compromised by the fact that almost all of the commercially available camera mounts were created without any thought as to how they would be stabilized or how they would manage the inertial loads. That aspect was simply left to the buyer to fathom out. The result of this scenario is that we have a few lame attempts to create a jack of all trades device that must try and cope with many different methods to power the movements along the desired axis. Most use gyros that have a full scale range that is no better than that which the model itself uses. Some mounts have high reduction gears and some use servos in direct drive. Some have very high inertial loads and others less so. What is even more ridiculous is that some try to use the imu on different parts of the camera mount ie. one axis out on the base plate etc. Some will use the servos in feed forward and also in feedback. Proportional and slew. The object of the exercise is to measure the disturbance between what the model does and where level is, so that should tell you where it needs to be mounted. And that point is not where the opposite reaction force of the servo can be seen. Compounded by soft mounting.
About two years ago I had this problem resolved and have since that time played around with various low cost ways to replicate the system. I am pleased to announce that I have joined forces with a team of people in the US who have both the manufacturing capacity to reliably produce such a device and have an extraordinary group of dedicated dev. guys who will now offer this device hopefully in the next few months.
Camera stability actually starts with the model itself as that controls the level at which the camera imu operates. It is rather like a self noise canceling headset where the reverse of the motion wave is fed back into the system to create a straight line. Direct drive high speed motors form the essential part of keeping the latency as minimal as possible.
On the subject of the model stability here is a preview of the ROWLAND F1-535 showing the one piece autoclaved carbon base frame. each motor can lift around 2.5 kilos but will probably never need to do that as weight saving is the key to every part of this dedicated system. Camera mount not show for obvious reasons.
View attachment 3212 Shown alongside a flame wheel to compare the size.

Thank you for a description of the mechanical issues. I was modeling a camera mount(I like to kill time in SolidWorks at night) and was thinking about the mechanics of the camera mount. I see that many of them use somewhat random locations for the pivot points(most appear to be in the center of the camera) and was thinking that it would be better(more stable and less obvious motion correction to the camera) to control the camera by pivoting about the platform's CG point. Figuring that platform instability pivots on the CG point. Is this wrong thinking?

What do you think the dynamic range, in degrees of rotation, is needed for a successful Camera Stabilization/tracking system?

Btw, very nice frame design!

 

[RTRyder]

Thanks for that explanation. Is this a capability of the OpenPilot products?



Thank you for a description of the mechanical issues. I was modeling a camera mount(I like to kill time in SolidWorks at night) and was thinking about the mechanics of the camera mount. I see that many of them use somewhat random locations for the pivot points(most appear to be in the center of the camera) and was thinking that it would be better(more stable and less obvious motion correction to the camera) to control the camera by pivoting about the platform's CG point. Figuring that platform instability pivots on the CG point. Is this wrong thinking?

What do you think the dynamic range, in degrees of rotation, is needed for a successful Camera Stabilization/tracking system?

Btw, very nice frame design!

At the moment OpenPilot has nothing similar to what Mk can do. They have not yet released their equivalent flight controller with GPS to the public, only the CC board which is just a basic flight controller with no GPS capability.

DJI has just released multiple waypoint flight for the Wookong-M controller, I expect they will have similar capability by the end of the year if not sooner.

Ken

 

[DennyR]

Thanks for that explanation. Is this a capability of the OpenPilot products?



Thank you for a description of the mechanical issues. I was modeling a camera mount(I like to kill time in SolidWorks at night) and was thinking about the mechanics of the camera mount. I see that many of them use somewhat random locations for the pivot points(most appear to be in the center of the camera) and was thinking that it would be better(more stable and less obvious motion correction to the camera) to control the camera by pivoting about the platform's CG point. Figuring that platform instability pivots on the CG point. Is this wrong thinking?

What do you think the dynamic range, in degrees of rotation, is needed for a successful Camera Stabilization/tracking system?

Btw, very nice frame design!

The Full Scale Range of my IMU gyros and the accels. sensitivity is a trade secret but it is significantly more sensitive than the typical 500 deg/sec used in most FC. imu's. You can with a little patience get quite good result with a CC board. About the same as a Naza.

 

[DennyR]

Thanks for that explanation. Is this a capability of the OpenPilot products?



Thank you for a description of the mechanical issues. I was modeling a camera mount(I like to kill time in SolidWorks at night) and was thinking about the mechanics of the camera mount. I see that many of them use somewhat random locations for the pivot points(most appear to be in the center of the camera) and was thinking that it would be better(more stable and less obvious motion correction to the camera) to control the camera by pivoting about the platform's CG point. Figuring that platform instability pivots on the CG point. Is this wrong thinking?

What do you think the dynamic range, in degrees of rotation, is needed for a successful Camera Stabilization/tracking system?

Btw, very nice frame design!

I think solidworks is a great way to create some ideas for camera mount designs. What I have done with this latest creation is to have a different mount for each of my cameras. GoPro, Panasonic DMC TZ10 and Nex have a different style to the one for the 550D and the Pannasonic HDC-TM900 and different one again for the 5D. It is all about weight distribution. The heavy cameras need to be underslung to keep the center of mass where it can be balanced and the lighter cameras can be at the front. The idea is to keep the moments of inertia close to the center of gravity, whilst having an unobstructed view. Direct drive servos and mass balancing on each axis should be the aim. Latency is the most common fault with most of the stuff out there. The gear driven axis was born out of a desire to create a smoother movement as direct drive systems were sometimes thought to be too notchy. What was was actually happening was the Tx outputs from the aux. channels was not the same rez. as the 4 main sticks.

 

[jes1111]

Direct drive servos and mass balancing on each axis should be the aim.

I, too, thought direct drive would be the ideal, but backlash (within the servo itself) kills the results.

 

[DennyR]

Jess have you not seen a HKS-HV787P. Even if it did have any backlash I think a man of your capability could solve that issue with a spring?

 

[jes1111]

Hadn't seen that particular servo but "no backlash" - impossible, surely? Any gear system will, by definition have backlash. The dual servo scheme (as you've used yourself) is the only effective control scheme (short of a robotics-style dual pinion compensated system, etc. - too complex for this application). Springs and/or dampers can "smooth" or "disguise" the backlash, but it's still there.

Certainly I agree with you that the problems/solutions for a small camera are very different to those required for a large, heavy camera (which is where my interest lies).