Taking my first steps in to photo matching. Was going to upload the file but got a message that it was too big?!
My model is of a new build that I am trying to show in position on the vacant plot where it is intended to build. Image 2 shows the model. Anyway, I’m getting some really weird perspective axis when I open up a photo to pop the model in to. The axis are in really strange orientations and sometimes even below the model entirely or with wildly exaggerated perspective. No amount of manoeuvring seems to bring them anywhere near the perspective of the photo. The model consists of a few groups so I have tried photo matching each group individually with a selection of different photos of the plot with the same strange configurations.
But the element is facing in the wrong direction. I assume this is because in the model is geo located, including rotation, to ensure it sits in a real world position, for shadows etc.
When I try and do similar on the whole model the initial positioning of the photo is underneath the whole model. Now I have a pretty good understanding of perspective but how I get those axis adjusted to even get the image above the model is beating me. This is what I’m getting:
so you already have the model built and you want to drop it into this photo? ( photomatch)
MAybe you just need to get a couple of key points and then manually move your model to that position? you could then later position it (rotate it) to conform with the actual site for correct sun position.
In styles there is a photomatch setting to change the transparency of the foreground and background photo. You might turn the foreground one off so that the photo image is always behind the model.
if you want to share the image and model (or just the shell of it) maybe I could better understand it?
Bad advise, but If you want to go rogue ( as I’ve had to on occaison where there are no lines to mp to with photomatch) you could load the photo as a watermark, and attempt to get a match by positioning (rotating moving etc) but that can be hit or miss and is helpful to know the focal length of the camera that took the photo.
Yeah that’s the idea. The plan is to get the model matching the perspective in the photo then export it out in to photoshop to do a composite with the photo and sort out the foreground etc. The shadows are not a big deal, I can sort those out later. Just a bit disappointing that I may have to do a separate version of the model just to photo match.
Anyway,I’ve done a stripped back version of the model and popped in to my Dropbox account. Hopefully you can get to it:
I had the same thoughts about your ‘rogue’ suggestion and have tried that. A bit messy and awkward. I do have the focal length of the photo and may well pursue this further. However, was hoping the photo match would be a bit more straight forward than this!
Thanks for your help so far, much appreciated.
Forgot to add the image:
Wow! I wish!! I never got anywhere near that. How have you done that? Clearly I’m doing something fundamentally wrong?
I’m off on holiday as of today for a week so will only be able to respond in text and not upload any images. Perhaps leave this until I get back (Sept 6th)
For the record, and to not exactly address your question…I’ve found that using an image as a background Watermark in the Styles menu, then rotating and adjusting the Field of View works for 99% of the the time I need to match a photo…the Photomatch tool comes from an earlier day when Google wanted everyone to model buildings for Google Earth and therefore have photo textures representing details to keep model geometry light. Might just be me but if photomatch is taking more than 10 seconds to line up, try my suggestion. I’d be happy to walk you through it if of interest.
I’d mentioned this as an option, and like yourself have had to go this way in the past. Positioning isn’t so bad but its usually the focal length that gets gnarly. If photomatch can get you close it saves juggling all those other parameters.
I see now that you mentioned that at the bottom. Yes if accuracy is top priority then photomatch it should be. For me, watermark “close enough” and usually works just fine. Either way, good to have options just in case.
Yea, I recently did a job and forgot to save the scene that I used for a render and subsequently had to re-render that same (now non-existent) scene. I had to use the render as a watermark as there were no visibly straight lines to get anything happening in photomatch to realign it with the model However, there was a foreground element and it took ages to juggle position, scale and focal length to get it to match, I’m pretty good with this stuff but it was horrible…made worse as I was under the gun for time.
Back from my holidays now and ready to tackle this little conundrum again!
I’ve managed to get the model somewhere close to what I’m after by returning it to the origin and rotating it back to the origin axis. However, the photo image is at a larger scale than my model. Do I have to rescale the model in some way?
Trying to match a one point perspective will make you crazy. (MatchPhoto wasn’t designed to do it) Here’s an example of what I mean by one point perspective.
Having a vanishing point straight ahead of you is difficult. @Keniff, your aerial has the red axis vanishing point straight ahead, though the POV is elevated. Looks like @whiterabbitdesigncompany found just enough rotation to get a green axis vanishing point to work with.
I’ll have to try that watermark method sometime. Sounds like what I had to do in Upfront years ago before SketchUp came along with MatchPhoto. I do wish for more advanced tools for photogrammetry.
Getting really disappointed at how difficult this is proving to be. PhotoMatch is not that useful to me thus far.
Eric-s I’d be really interested it having a go with the watermark technique, if you have the time to walk us through it? Cheers
Just for the record, in case I’m coming at this completely wrong, as mentioned previously what I’m trying to do is drop my model into a photo for visualisations. The photos have been taken by a drone, which I am currently trialling to see if I want to purchase it. If I can’t get my models easily to match the photos from the drone then there’s no point me buying it!
I think you will be able to achieve this by getting the right angles on the site. If you are using the drone it may be useful to place some orange cones at key/known places that might help with configuring position data and dimensions.
Motion drone footage can also be camera tracked with after effects or blender which will also produce positional info to lay structures on.
As for the photo match alternative using watermark…, see rough gif. The drone image is set as a watermark (background). Then its a matter of moving the model… rotating etc to get it to line up on the BG. One key part of this is to know the lens (FOV) of the drone camera and match that with the camera on the SU model as its one less parameter to manage. Also can be helpful to go into x-ray mode to see through model. Align the sun to match.
With this technique you can leave as is and export the 2d image comp. In the image I sent I rendered the image with an alpha channel and then comped it with the photo in PS. There you have a little more control in color matching the two elements etc.
You can add elements like trees in the model to overlap where the BG images should be in the foreground…or can do this in PS, but doing it with models in SU helps keep perspective cleaner.
White Rabbit, you’re a star! That looks fab and a lot more like I was looking for. Seems pretty straight forward too. Still disappointed that SU doesn’t do a better job with photo match, but hey ho it is what it is.
Thanks everyone for all your help, very much appreciated
Copy that. I’d still persevere with photomatch as it does take a lot of the guess work out.
Sometimes it also comes down to how accurate you need to be too. For a concept rendition it may be OK to be a little less accurate and a little more “creative”? Whereas an addition or something that’s more integrated may need to be more accurate.
