Thread: Converting Photo to Orthographic Drawing

Thanks Ian.

I may have made some progress, and the following text and attachment attempts to describe my method, albeit with a very simple thin rectangle.

The perspective sketch represents a hypothetical billboard viewed at an intermediate height from an oblique angle.
This view is the only input data to the solution.

1- Project billboard top and bottom until they meet and define this point as the Vanishing Point, VP.
2- Draw the Horizontal Line, HL, through the VP perpendicular to the sides. This line is also an edge view of the Picture Plane, PP, on which the perspective sketch is generated from the orthogonal drawing.
3- Nominate(ie guess!) the Observation Line perpendicular to the HL. This a commonsense judgement reflecting the view of the perspective sketch.
4- Nominate (another guess) the Observation Point, OP, on the OL.
5- Draw the Sight Lines from the OP to the Perspective Sketch at the intersection points of the HL and the sides. These sight lines will also go through the ends of the (yet unknown) plan.
6- Draw circles centered on the HL intersections with radii set to the side heights above/below the HL.
7- Draw lines from the OP tangential to the circles. These lines provide the scaling factor between the perspective sketch and the orthogonal drawing.
8- Nominate (another guess) a billboard height suitable for the proposed orthogonal drawing and use this value to draw offsets from the Sight Lines. Drop development lines from the intersections of the Scaling Lines and offsets perpendicular to the Sight Lines.
9- The intersections of the development lines with the Sight Lines denote the ends of the plan of the billboard.
10- The elevation would be projected perpendicular to the Produced Plan.

So there it is at the moment.
A bit convoluted and happy to receive criticism/comments/suggestions.
I will now try to extend the method to two planes at right angles.

If the end result is not sensible, I will have to go back and vary the nominated values.
In any real situation, some sizing data will probably be available to provide the sensibility check.

Perspective 4.pdf

John

I have made some small steps towards the final solution.

First I have to admit to an error in the previous description.
In step 7, the Scale Line is not drawn tangential to the circle, instead, a radius is constructed perpendicular to the Sight Line (SL) to the circle and the Scale Line is drawn from the Observation Point (OP) to the end of the radius.
A very subtle difference.
The drawing itself is still OK, I just cleaned it up a bit too much for the post.

In the billboard example, if the aspect ratio Width/Height is known, you can search for the correct Observation Point because any incorrect location does not give the correct aspect ratio.
But, a lot of lines to get there.

If the object is two perpendicular planes, the correct Observation Point will produce a plan view with the edges of the planes correctly at 90 degrees, if the OP is too close, or off to one side, the construct produces a plan with the included angle >90 degrees.

Since the objects of interest to me are essentially boxes or arrangements of boxes, I feel I will soon be able to tackle the real problem.

John

I have made some progress, after a setback for a day or so.

My CAD is DraftSight under Windows XP on a moderately old Dell laptop, and it has worked fine for a couple of years.
That is, until the display went black, and it took me a day to realize there was still an image there, it just needed a flashlight held 'just so' to see a glimmer, apparently the backlight had failed.

A bit of searching revealed the method to select an external second monitor, and half a day to discover neither my spare LCD or CRT monitors operated DraftSight correctly.
So, get a new driver from HP Compac, no joy.

Then, (feel free to kick me over this) I uninstalled DraftSight and went to Dassault Systems to get a new copy.
But, Dassault Systems have released a new version, 2015 SP3, that is not suitable for XP, and, no sign on their site of an archive facility for old versions.
No prize for guessing that I did not have a copy of the original install.

So, search the web and get a V1R4 off soft32, install Ok, seems good, but, DraftSight reports an unstable graphics driver and that is what it looks like.
A quick link to Dell lets them interrogate my machine and they download a new graphics driver that installs Ok, and joy, everything now operates Fine.

Except - I now get an occasional warning from my protection that there is a virus trying to do something.

The subject progress is depicted in the attached drawing.
This shows two perpendicular planes in plan view, P1, with an observation point, OP1, generating the perspective sketch shown (with assumed height and an arbitrary horizon line through it).
Then, the projections are worked backwards from the perspective sketches using various observation points, OP2 to OP5, to produce the plans P2 tp P5.
The only projection lines shown are those for the sketch-to-P5 conversion, and you can see just how messy the drawing can get.
Plans P1 to P5 show different sized lengths of the plan legs, and an included angle that is not 90 degrees.

For the real problem, a non-perpendicular result indicates that the selected OP is not correct, and I hope there may be an indication of which way to move the OP to get a better result.
There is also the possibility that the elevations, when drawn, could provide some information as to which way to move the OP.

John
Perspective 6.pdf

A small advance after some review this morning.

For a box-like object viewed at an oblique angle mid-way up the height, there are Vanishing Points on the Horizon Line to the left and to the right.

Commonsense says that a ray projected from the Observation Point parallel to the two visible sides will also recede to the same Vanishing Points.
Such rays on the Plan are lines drawn parallel to the two sides.

The attached drawing shows the previous drawing with the two parallels added.
On the left Vanishing Point the co-incidence is very good, on the right VP not so good.
But, knowing some of the difficulties producing the drawing, the error could be operator-error.

John
Perspective 7.pdf

I did a new projection carefully and accurately and proved that the rays referred to in the previous post do meet at the vanishing points, the possible operator error is in there somewhere.

Further commonsense, and a check with Mr. Pythagoras, reveals that the Observation Point, OP, must lie on the circumference of a circle constructed with the Horizon Line from Vanishing Point Left to Vanishing Point Right as the diameter.

The circle relationship should reduce considerably the work involved when searching for the OP when converting a perspective sketch into an ortho drawing.

Thinking about it, if a photo has not been cropped, the Normal Sight Line from the OP to the Horizon Line must be in the centre of the photo.
The choice of centre would seem to be a logical first guess.

I wonder if there are any more fundamental relationships yet to be teased out of these geometric projections.

From the number of views, this subject is of some interest.
You can see from the attachment below that CAD is probably necessary for most real situations, but the projections for a simple exercise can be done on a drawing board, give it a go.

Perspective 9.pdf
John.

Some bad news to report.
I did some exploring of the effect of moving the Observation Point along the locus of the large circle circumference.

The attached extract shows the correct normal sight line and observation point shown in blue.
I know it is correct because the perspective sketch was generated with these.

The projections were then worked backwards to get the correct two perpendicular planes in plan view with the two legs parallel with the orthogonal lines to the vanishing points.

Then I erased all the projections and shifted the OP a little to the left, as shown by the small circle, and did it all again.
The two planes are 89.4 degrees apart, showing the OP was not correct.

The next test was a little to the right.
This gave 90.000 degrees between the two planes with the planes parallel to the vanishing point orthogonal axis.
Not so good, cannot see my error.

I repeated the exercise a little further to the right, leaving all the projections visible.
Same result, perfectly (at least to 4 decimal places) perpendicular and orthogonal.

I am at a loss to explain it,
There must be something wrong with the theory, or I am making some sort of stupid mistake.
I will go back and rethink and recheck, firstly examining if greater precision on the angular measurement reveals something.

At the very least I am getting some good practice on the CAD.
Perspective 10.pdf
John

Hi John

I think there may be an error in your theory.

a photo contains two adjustments.

• the POV resolves to three vanishing points -- one to the right, one to the left and a third either above or below the image

• then the lens itself induces a number of "distortions" as it converts the 3-D scene onto a flat plane which may be tilted relative to the POV


I'm coming to the view that what you want to do requires a 3-D solid modelling program -- Solidworks is the only 3-D modelling program I've used.
In Solidworks the process for what you want to do would be
1. create a 3-D model of what you think the object in the photo should look like -- start with a model composed of stacked cubes, rectangles, and cylinders
2. adopt a POV that matches the POV in the photo
3. compare the Solidworks perspective image to your photo noting critical differences -- the most likely cause of critical differences between the two images will be what you though were rectilinear 3-D objects are actually tapered 3-D objects.
4. adjust the 3-D model and repeat steps 2, 3 and 4 till the critical differences between the images "vanish"

Yes Ian, I think my expectations cannot be met with the input data of the problem I set.

Note that that the problem was restricted to a perspective sketch with verticals in the sketch, so only two vanishing points.

I put some thinking into the problem and now realize that any observation point on the circle between the vanishing points is a valid solution, unless additional data demonstrates otherwise.

A 3D model has the same difficulty, just a quicker way of reviewing the situation.
One of my ambitions is to get a 3D package running, when I get my new Windows 10 laptop.

When I have some results of my original quest available, the railway vehicle, I will submit the results for your review.
I now have to get the best possible scan of the photograph in question and attempt my first ever incorporation into a CAD package.

The discussion and method development has prompted me to put a document together describing the sequence and projections used.
It could prove useful to others and should be submitted in a week or so.

John

Originally Posted by electrosteam

Yes Ian, I think my expectations cannot be met with the input data of the problem I set.

Note that that the problem was restricted to a perspective sketch with verticals in the sketch, so only two vanishing points.

Hi John

I don't think this is strictly true.
If you look at a photo of a tall building, it typically leans into the image because of the effect of perspective. If the photo is taken at the mid height of the building the "leaning" effect is less noticeable, but still there.
I suspect that the general case is for the POV to be on teh surface of a sphere connecting the vanishing points.


In the case of your railway vehicle, it's unlikely that the camera was positioned at the mid height of the vehicle, but this would be a very very small correction compared to the big unknown of what sort of lens was used. If the lens had a field of view greater than the human eye, then the resulting image will be significantly distorted by the lens itself.

Wow, from all the views this is a subject of some interest.
Attached is the photo that started the thread.
The descriptive document briefly mention before is currently being reviewed in draft form by one of the forum members.

Ian,
You can see that the vehicle is essentially a rectangular box viewed at mid-height from an oblique angle, hence my restriction of verticals in the perspective requiring only two vanishing points.

The only known problem with the procedure described so far is, where does one nominate the direct face-on view to be ?
Effectively, this was the camera axis when the photo was taken.
I have spent some time exploring the image, but cannot confidently nominate the axis position.

Berninabahn 151 Compr.jpg
John

Originally Posted by electrosteam

Ian,
You can see that the vehicle is essentially a rectangular box viewed at mid-height from an oblique angle, hence my restriction of verticals in the perspective requiring only two vanishing points.

The only known problem with the procedure described so far is, where does one nominate the direct face-on view to be ?
Effectively, this was the camera axis when the photo was taken.
I have spent some time exploring the image, but cannot confidently nominate the axis position.

Berninabahn 151 Compr.jpg
John

John

Ok, from the photo name, the vehicle is on the Bernia rail line at Campocologno station (I think) -- the tower and surrounding buildings are rather distinctive and the number of sort of matches Google Street view.
The vehicle looks like two units
a power / goods unit and a goods wagon
The track gauge is a 1.000 metre
I suspect the camera is positioned at about or just below the head lights
also, I wouldn't assume that the train is on a flat section of track.

Look familiar? Thanks Ian.


Berninabahn.JPG

(This is South-East from St. Moritz, Switzerland, just before entering Italy.)

From your photograph, I thought the height of the camera would have been closer to the bottom of the windows rather than near the lights but this is only because I put it into a graphics program and the horizontal line I used fit there better than using the top of the lights. Seems rather high for a camera to me so go figure? The ground where the camera would have been positioned didn't look appreciably higher when looking at the above on computer. Have a look for yourself though to take it all in.

Craig

Originally Posted by electrosteam

The only known problem with the procedure described so far is, where does one nominate the direct face-on view to be ?
Effectively, this was the camera axis when the photo was taken.
I have spent some time exploring the image, but cannot confidently nominate the axis position.

John

As mentioned in a post further up the page, I also assume that the whole photograph is not shown. If this was the whole photograph, the centre of the image gives you the camera axis. Makes a mockery of what I just said regarding the height of the camera being about the height of the window bottoms.

Berninabahn Axis.JPG

Also, does a 5" scale mean 5" across the rails, that is, a one-eighth scale if the gauge is 1.0m as per Ian's post? Quite a large scale (ride-on?)

Craig

I have not read beyond post one and two okus a saw a pic of the tram.

What i do recall from school os how to draw a scale drawing from a picture / photo. The method taught would allow you to scale in two possibly three views

Dave TTC
Turning Wood Into Art