Building a DIY 360 camera for Street View. Has anyone tried it?

Cameras and Setups
#1

I was excited by the Mosaic51 camera, but for me it is too expensive (EUR 20k Does anyone here own a Mosaic 51 camera?).

Is there anyone else in the space building dedicated GSV camera, at a prosumer price tag? Less than $2000 USD.

The offerings from Ricoh seem to be focused on indoor tours, GoPro on action sports photography, LabPano on live streaming, Insta360 on VR movie’s…

It seems there is a space in the market for GSV, or more generically, mapping cameras. I am enquiring if anyone has considered building one? Or if they are doing so?

#2

The team at Open Path View built a camera using GoPro Hero’s: https://openpathview.fr/ (they open-sourced the design). They have some great shots of French National Parks taken from it.

The old GoPro Omni camera rigs might also be of interest: https://360rumors.com/review-26-gopro-omni-360-camera-rig-clone/ . 4 older Hero’s could be picked up for less than $1k and use your phone as a GPS logger.

@360dev (and the wider community); what would be the killer features of a 360 mapping cam?

#3

Also just discovered PICAM360:

https://www.picam360.com/

This looks like a cool project. Not sure it’s ready for GSV capture in current form, but perhaps a project for an engineer out there…

#4

For about the price of an action camera, you could probably put together a very high quality rig with 6 sensors. Have a read of this:

Would require a GPS module for the Pi and some housing, but for under $1000, you could easily get a 11k Street View Ready cam (with the right technical knowledge).

2 Likes
#5

Interesting!
Let’s calculate for a sec:
Lens: 3x$18,08 180 degree Arducam

Camera: 3x€54,25 Arducam 16MP Pi Camera

Board: 3x89,95 Pi 4, 8Gb

€513,84

Add serious memory, a F9K and pitch/yaw/roll and we have a massive setup!

The lens does 180 degrees vertical when I read correctly, then it’s probably a smart idea to let them point up by a few degrees… there is someone below there anyway… and we get space for the nadir :wink:

Ok, who’s gonna build it :wink: ?

[update]
When you think of it that you will need to add memory, build a case, a bunch of time programming the thing… well, I think a labpano one, GoPro max or a second hand insta pro is a better investment?

#6

Hi @Eesger, @360dev, @dgreenwood-trekview,

Both myself and Joscelin @OmniSynThesis360VR have been trying to persuade the folks at Labpano the last few months that there is a huge untapped mapping camera market who would purchase the right hardware if it was available.

There is the debate that higher resolution than 8K would be preferred, but currently unlikely as a prosumer level product. They only need to add a few additional software capture options and F9K GNSS hardware and we’d have a pretty capable product.

(Update)
And improve their image quality.

1 Like
#7

Yes indeed! We are doing that at https://github.om/OpenDroneMap/odm360. Our primary use case is photogrammetric surveys (3D reconstructions) but intend it to work well with other use cases, like Street View.

We started with 6 cameras, but it makes the alignment a little too fussy. 7 cameras gets you a little more overlap.

1145×852 75.8 KB

1 Like
#8

It looks like my 2nd and 3rd image might have gotten blocked, maybe for being a secondary domain. So I have edited that post to remove the other images

But: moar to be seen here:

image
image1024×768 321 KB

Ahhh, good cable management is a treat all its own.

2 Likes
#9

@smathermather what would we be looking at in terms of the cost of parts for this setup?

I’d love to hear more about the stitching workflow, and see some sample images if possible?

#10

I don’t have finished numbers on cost, but we are looking at ~1000USD, depending on numbers of cameras and quality of GPS. The below is for >75 megapixel stills.

For sense, we have $400 in RTK/PPK GPS (this can be tuned down depending on project needs), 7x $75 per camera+lens, 7x $12 Raspberry Pi 0s, 1x $70 for a Pi 4, and then various other smaller items.

1 Like
#11

This looks seriously interesting, not sure how portable one of these would be?

Whats particularly attractive is that it’s Raspberry Pi based and consequently completely open source. Making it possible, presumably (once you get home, anyway) to pre-process and upload the panos directly to something like OpenTrailView on-device.

1 Like
#12

The cameras, including all the frames and bolts to hold them in place total ~800 grams so far. Then you would have to add in the backpack (reasonable weight), battery, adapters. All in all, it shouldn’t be all that heavy with 7 cameras.

1 Like
#13

I did a terrible job reading all the questions: stitching workflow and sample images to come.

#14

I assume you know hugin? (For stitching the images?)

If you need, I can help you create a good template?

#15

That would be fantastic. That’s been at the back of my mind since this project started. What do you need from me? The requisite 7 images?

1 Like
#16

Sorry for the late reply (busy, busy).

I need the separate images and the focal length (… all the info you can give me about the lens actually;) )

1 Like
#17

Cool. No worries: we’re probably a week or two out from triggering the whole rig.

I can get full calibration parameters if you like. :slight_smile:

#18

Full calibration will not do the trick… that will get “lost in the noise” anyway… When a 360 is taken some stitching points for a good result have greater distance then others… a house nearby… a tree in the background. there lies the big trick…

This is the data Hugin asks for:
image
image
With that information I should have enough data to play with :wink:

The great thing about Hugin is, that it has a sort of “fuzy logic” to stitch the images together… when you have a good template with enough control points (not to few… but also not to many) you give the software enough for it to stitch them together. Hugin has tech for a sort of object detection… that is why it gives those damn good results (with the right template :wink:

PS: do keep in mind, that stitching with Hugin takes quite a bit of CPU power… (and thus time)

2 Likes
#19

I know the guys at OpenPathView tested extensively Hugin in their stitching workflow: https://openpathview.fr/

They too came across the CPU issues. I believe it was almost 30 seconds per photo in their workflow.

Might be worth chatting to them? A lot of their stuff is also public on Github: https://github.com/openpathview

1 Like
#20

Yes, they are on my list to reach out to: their work is pretty inspired.

1 Like

HQ | Blog | View / Upload Photos | Become a Trekker | Facebook | Instagram