We have 12 constant cameras, arranged as a bow, taking pictures of a moving car.
The goal is to build a 3D model by using the images. The steps are:
1- Background subtraction.
2- Using an algorithm to build a cloud points for each camera.
3- Consolidate between the modules.
At the first part of our project we used background subtraction algorithm, this algorithm is appropriate for problems where objects that are related to foreground are moving from frame to others, and all other objects are constant. This algorithm fits our requirements because the car is the only object that moves between frame to frame. To this algorithm we add Morphologically close image to close all holes that are created inside the car because the car itself is a flat object in some areas.
In the second part we used structure from motion algorithm to build a 3D model: In the beginning we fit this algorithm to work with more than one camera, we did not get good results because we get little common points between two different photos, so we settled to building a 3D model for each camera, but we could not succeed to consolidate between the models.
in the first part of background subtraction we got good results, whereas in building model 3D there were problems that are related to lighting and the fact that a car is flat, moreover we searched for algorithms for feature detection that are more computable for our issue but we did not succeed.