Camera Projection and Geometric Transformation Library
This library provides a simple and intuitive library for basic mathematical and geometric transformations in 3D space. The library includes simple geometric functions such as finding line-line intersections, line-plane intersections, midpoint finding, unit vectors, translation transformations, and rotation transformations.
The library also includes a camera projection kit which handles projection from images to 3D space. These projections from 2D to 3D is made by projecting pixels into lines/rays in 3D space. I intended this library to be a general purpose library. If you want to see some applications of this library, please see my Unmanned Aerial Vehicle project here.
The library can be used for numerous applications related to computer vision and spatial intelligence.
For example, in my project, AerialAR, by receiving the position of Points of Interest from Google Maps API and the position and orientation of the camera, I am able to show visible Points of Interest in camera imagery. The program also handles sketch recognition. This allows users to circle a certain area in the image to find the area's GPS location as well as to determine the name and type of each building inside that area. This feature also uses camera projection, as it is projecting the sketched pixels into lines/rays in 3D space. Then, by finding the lines' intersection with the ground plane, the 3D location of the user's selection can be determined.
Another application can be seen in the test executable I have provided. The executable I have currently is a test for a private project which involves detecting lights in a building. Knowing the position and orientation of a camera at all times, a map of the lights in a building can be constructed by projecting pixels into rays. Because the camera is able to take pictures at a high rate while moving, two pictures at similar but not equal positions can be taken. This allows stereoscopic reconstruction of the 3D location of the light.
Currently, the library doesn't support computationally intensive methods such as finding intersections of 3D objects. Also, another useful implementation would be 3D location finding using stereoscopic methods. As of right now, the library by itself requires additional information in order to determine an accurate 3D location of an object in the image. This can be fixed by using two images to project into 3D space. However, that may require some form of feature detection, as the camera would need to differentiate each object in the image to avoid errors related to stereoscopic vision.
Also, implementing quaternion rotation would also be useful for several applications as it is agnostic of order of rotations.
All of these are planned implementations in the future
Besides downloading the library, you would also need to install Armadillo, a linear algebra library, along with its dependencies in order to use this library. You can get that here.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent