2d-border-to-3d-border

Abstract: convert the 2d border of cars to 3d border in an image (knowing the alpha angle of the detected vehicle already)

Envirnoment: Windows 10, Pycharm 2019, Python 3.7, OpenCV 3.4.1, Numpy 1.16, Scipy 1.4

How to configure the environment?

1. install libs in python

   Use the command pip install xxx to install the libs in python

How to run this project?

1. run 2d-3d.py

Steps:

1. read image and data file

   Read image from 004007.png and data file from 004007.txt

   The 5th column is x_min, 7th column is x_max, 6th column is y_min, 8th column is y_max, 4th column is alpha angle

2. definition of alpha

   

3. catagorize according to alpha

   

4. calculate points' coordinates (3 of them)

   e.g. if -90°<alpha<=0°, x_2=x_max, x_4=x_min, y_3=y_max.

5. definition of rotation matrix

   

6. definition of introvert matrix

   

   In this project, f_x=f_y=721.5, s=0, x0=609.5, y0=172.9.

7. slove rotation and translation equation

   

   

   I suppose the length of a car is 3.96m, the width is 1.59m and the height is 1.5m. Thus, I can calculate the x_w, x_y and x_z which are points in the world coordinates. The camera is installed above the car, and its height is 2.22m. So t_y=2.22m. t_x, t_z and ry are parameters we don't know, but we have already known 3 coordinates in the image (step 4). We have 3 functions and those unknown numbers can be solved!

8. solve all corners' coordinates in the image

   Using the function mentioned in step 6, we can solve the coordinates of all corners in the image coordinate.

9. draw outlines on the image