1. IntroSo basically we detect a bunch of interest points and tract them over couple of frames to get their 3D world coordinates. Simply as that. |
||
2.Finding interest points and tracking themUsing harris corner detection, we could get 500 interest points like this:
Now we begin tracking, yay~~~
There are some points that are gonna fall off the frame bounds at some points, we have to toss them away. Here are all the points, points that stay and points that are tossed away:
|
||
3.Structure from Motion
Now that we have the points and their movements along the frames, let's build the model, so much fun! The red fan-shaped object is actually the vectors of the cameras of each frame, they are so close together that, they appear this way. Since we constructed and orthogonal camera, it really doesn't have a precise location. What matters is just the direction. |
||
Tan "Charles" Zhang