Amazing 360x180 P3P Panoramic from 500ft above my house

[mmjax]

Both are great, but I think I like the original one better.

 

[nhoover]

Very nice - I'll give it a try soon.

 

[GadgetGuy]

- UPDATE -

Had a break in the weather here so based on suggestions from this thread I went up again and did a "redo" using only stills snapped with the camera instead of taking video and extracting the frames. I went a little crazy taking a couple hundred photos but was easy.

New version using 4,000 x 3,000 pixel camera shots
http://theitalianpalace.com/tipv2/tip360v2.htm

I am not going to give my opinion on which one I think looks better - want to see what everyone thinks. I am convinced however we need an "app for that" via the API to do this automatically.

Original version using 3840 x 2160 extracted movie frames
http://theitalianpalace.com/tip360.htm

Interesting observation - the final jpg for the new version is roughly the same size as the original despite the fact that the camera only pics are higher resolution each.

The second one with the stills clearly has more detail and is sharper, and was a lot less work! Here is an enlargement of the tennis court on each, with the stills only version first.

It's pretty obvious which is which...however, your first one from video frames loads faster.

 

[GadgetGuy]

From my own experience with video, any panning, no matter how slow, results in ripped frames. All movement must be eliminated to avoid the ripped frames. I think you could still accomplish this through video still extractions, but stop the panning every 60° of rotation for a full second, to get sharp video frames. Shouldn't take any longer than shooting the still frames, but you maintain continuous FPV and video throughout the process. It will take 15 seconds or so per rotation at each of the three elevation levels, but less than a minute overall. Select the 20 still frame video extractions with sharp detail. Any one of the frames during each of the one second pauses should be equally sharp.

 

[rhwd2003]

So how many total still pictures did you use? Did you follow the same procedure slowly turning at the same camera angle all the way around while taking pictures and then turn down the camera a little more and keep going? How many total pictures did you come up with? How did you put them all together at the end?

 

[jstn]

Was just thinking - isn't there an obvious SDK opportunity here? Does anyone know if with the new SDK a "script" could be written to move the P3 to a particular spot & altitude, position the gimbal, take a series or burst of pictures while rotating, move the gimbal, down, repeat, etc. That would be exactly what I did but would use the higher res stills and be precisely controlled for an exact overlap of pics. Also would be incredibly efficient in terms of time and number of shots. Heck I could see with such automation the shots necessary to do what I did but better be performed in less than a minute.

Heck, people would PAY for an app that did this....

I've actually written an app that does exactly this but it's unlikely that I'm going to release it - at least for now. The current state of the SDK makes it not worth it.

 

[Bob Denny]

When you view it in the iPad in Chrome, it pans and tilts with the tablet!! I think it shows what you would see through a window facing in that direction. I was so surprised!!!

 

[jadebox]

When you view it in the iPad in Chrome, it pans and tilts with the tablet!! I think it shows what you would see through a window facing in that direction. I was so surprised!!!

It does the same on my Android phone. Pretty cool! I am going to have to try creating an image like that some day.

-- Roger

 

[Domenic]

So how many total still pictures did you use? Did you follow the same procedure slowly turning at the same camera angle all the way around while taking pictures and then turn down the camera a little more and keep going? How many total pictures did you come up with? How did you put them all together at the end?

I took about 230 for the second one. But just did a third @ 1,600ft using only 160 pictures. Its MUCH easier just snapping pictures and not fooling with the video and extracting. I also agree that the quality is better with the regular photos.

http://theitalianpalace.com/tip1600/TIP1600.htm

 

[Domenic]

I've actually written an app that does exactly this but it's unlikely that I'm going to release it - at least for now. The current state of the SDK makes it not worth it.

So how do "applets" like this work with the SDK? How is it implemented / run from a user perspective? Outside of the app? In parallel somehow?

If you are working on such an app obviously it would be a tremendous contribution.

 

[GadgetGuy]

The pan and tilting works just like the Street View in Google Maps with the accelerometer turned on. That feature sold me on my first Smartphone! It was like standing in the middle of the street and turning around and seeing everything live!

As to the number of pictures used for the stitching, beyond the minimum necessary to exactly cover the grid of checkerboard, the additional photos are only to help with the stitching process (not to improve the image quality), and an overlap of 10-25% is all that is necessary, not 50-100%. So 24 photos are all that are ever needed, whether shot as jpegs, or extracted frames from the 4K video, paused at the correct intervals to add up to a grid of 360° across and 120° tall. The gimbal elevation in the app has 10° hatch marks to break the 120° up into 3 layers at 0°, 60°, and 120°, with the red hatch being 90°. The 360° field of view rotation has to be done manually, using the direction of red arrow over the map in the map view, but since the lens has a 94° FOV, anything from a 45° to 75° interval should work. The easiest to keep track of would be a 45° rotation each time, consisting of 8 shots total, so, 0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°, where you have front, right, back, left, and half way between each. 3 x 8 = 24 frames to stitch together with considerable overlap. 1-2 second pause at each location, to get sharp video frames, and the photographic/video process should only take 2 minutes tops. The rest is extraction of the 24 frames, and the stitching. Ready, set, go!

This is the manual process that your SDK would automate, but it's pretty simple to do it manually!

 

[Airborne]

agree, 24 photos can do it. i used 26...

 

[GadgetGuy]

agree, 24 photos can do it. i used 26...

Thanks for the validation!

 

[GadgetGuy]

I took about 230 for the second one. But just did a third @ 1,600ft using only 160 pictures. Its MUCH easier just snapping pictures and not fooling with the video and extracting. I also agree that the quality is better with the regular photos.

http://theitalianpalace.com/tip1600/TIP1600.htm

Shooting with the sun directly overhead also keeps it out of all the images!
1,600 feet looks really cool! I'll have to try it when the FAA isn't looking, and no aircraft are in the area!

 

[GadgetGuy]

Just took both stills and video to do my own from 500 feet!
Lots of processing to do! Should be fun! What I did find is that anything above 90° elevation is just empty sky, unless you are at the bottom of a valley next to a tall mountain range, so the top row at 120° is really unnecessary in most locations, and just introduces overexposure due to the sun.

 

[Domenic]

I guess I am getting better - did this one in only about 100 photos

Pano of Wilmington Delaware Riverfront

http://theitalianpalace.com/360/wilmington/WilmingtonRiverfront.htm

 

[chaghal]

incredible

 

[jstn]

So how do "applets" like this work with the SDK? How is it implemented / run from a user perspective? Outside of the app? In parallel somehow?

If you are working on such an app obviously it would be a tremendous contribution.

From a user perspective, it's a separate app from DJI Pilot / DJI Go that runs on your phone or tablet. In this particular case, you could either use the DJI app or the 3rd party app (since my app in particular displays the camera feed in-app) to properly position the drone, and then you would use the 3rd party app to kick off the photo taking sequence. The app in question would use the starting point as a reference to begin adjusting the yaw and gimbal position while taking photos so take the appropriate images.

 

[Domenic]

How about an initial waypoint including height then return / auto land for a complete no touch mission?

 

[GadgetGuy]

My first attempt at this directly overhead at 500 feet resulted in a loss of signal, and the Phantom repeatedly descending to Return Home! I suggest trying it at some distance out in front of you!