There is no blueprint available for this concept car, so I searched for a new method trying to be accurate from reference pictures.
What I've come up with is using insight3d, a Free Software capable of doing 3D model from a set of pictures. From the focal length of each picture, and being given a set of common points (e.g. the ends of body edges) it can compute the location & orientation of cameras in 3D space for each picture, and therefore put all the common points in 3D (triangulation from the pictures)
It's a quite robust method : spotting a point on a picture may not be very accurate, but as I use many pictures, there is a reliable, average position which can be determined. I currently use 9 picture, and will add more for front and rear ends of the car. Adding a picture is easy, you just have to place some of the points you placed on previous pictures to locate the camera in 3D space.
I've located about forty points so far, many more to do. It's a long long way but then the workflow is quite simple : export the camera location and points cloud to my favorite 3D package, then use that to create spline cage. Later, as reference pictures are located in 3D space, modeling can be performed directly on top of the refpics.
My progress is quite low at the moment, so only one screenshot from insight3d attached. But I thought it's still worth posting because I wanted to hear about your own strategies for blueprint-less modeling.
Thanks for the support, I'm glad of the enthusiasm you show, guys !
Here are my progress. I added some pictures, I now have 12 pictures used for 3D reconstruction, with 86 points. And some areas of the car aren't mapped yet...
The first picture below shows the current look of my project within insight3d.
The second picture shows the imported VRML file into Blender : it's just a cloud of points, because insight3d doesn't export edges, only faces. As I created no face, I have only points.
The third picture shows the edges reconstructed, it's still hard to see a car because of the random orientation of the model.
The 4th picture shows some specific edges used to properly rotate the model : the yellow is for horizontal, going from front wheel center to rear wheel, and the white one is the middle of the car.
Last picture shows the model, properly aligned with a mirror modifier. It looks quite fine, except you can see the reconstruction isn't exempt of mistakes : for example, the rear wing isn't symmetric.
hey tom, good to see you starting a project, looks nice so far, but wouldnt it be easier, to do it by eye? with just splines ? it would probably take 2-3 hours tops.
Maybe it will be easier for someone else, but my low-level skills made me think doing it by eye is not easy for me
Anyway, I'm done with insight3d. 15 photos and 171 reference points later, I was able to import the model in Blender and align it very properly. It turned out pretty well, IMHO. I attached viewports pictures, I will now start actual modeling (a fully detailed cage won't be useful I think)
Great splinecage man, looking forward to the model itself Tho i have to disagree on the full detailed spline cage. I personally think it's very usefull, and i use splines almost always i make surfaces! So if you can find some time, i would highly recommend you'd make one, it will only be good for your surfaces
Well you may be right. Actually I never used real splines for such a cage, always a mesh with edges only, but with enough vertices to smooth curves and a lot of attention spent to get coherent curvature. If Blender was able to do NURBS I would use them but currently I stick to meshes.
Visit Homepage
Reply With Quote
