/stereo

The project stereo

Primary LanguagePythonMIT LicenseMIT

stereo

About project Stereo

Instructions:

Simply download the entire stereo completely, and you need:

  • Python 3.6
  • Opencv 4.0

Calibration.py

This document completes the 6,7 sub-question, implements two functions of camera calibration and image correction.

Instructions:

Run calibration.py, you can see the calibrated picture in output/calibration/.

Results

The result case (one randomly selected, the rest can be viewed by running the program):

Image before correction:

left03

Draw the chessboard:

Image after correction:

Method 1:

calibresult1_03

Method 2:

calibresult2_03

Note: Some pictures cannot be corrected, and the size of the picture is 0.

epilines.py

After learning the Epipolar geometry, this document implements the functions of finding the corresponding points, calculating the fundamental matrix, calculating the epipolar line and marking them according to the official OpenCV document.

And tried two methods provided in OpenCV: brute force and FLANN.

And we use this to test whether camera rectificationis is good or not

Instructions:

Run calibration.py, you can see the calibrated picture in output/epilines/.

Results

One of the left image:

epilines_left03

Corresponding the right image:

epilines_right03

Corresponding the matches image:

epilines_03

However, there are some errors in the corresponding points, and some corresponding points are not on the board. The corresponding point relationship of the keyboard is correct, probably because the location is similar, but it is not the object we want to study.

calibrate_binocular.py

This document is used to implement Stereo calibration and rectification.

Instructions:

Run calibration.py, you can see the calibrated and rectified picture in output/calibration/.

Results

（The figure below is the same source file analysis as the corresponding point analysis chart before correction.）

The left original picture:

The right original image:

calibration and rectification:

(Obviously, it succeed.)

Some values about the parameters are as follows：

After stereo correction, perform corresponding point analysis:

We can find that the epipolar lines are parallel and the corresponding pixels are the same position in the two figures. It may prove that the rectification is successful.

Disparity

In this module, I use OpenCV to compute the disparity maps for the images I used for stereo calibration. And the result as follows.

Disparity_BM.py

Block matching method:

The left image:

The right image:

BMright04

Disparity_DP.py

Dynamic programming method:

The left image:

The right image:

DPright04

Disparity_SGBM.py

Semi-global block matching method:

The left image:

SGBMleft104

The right image:

SGBMright104

当增加SGBM中的blocksize值后，可以减少一些噪声，但是也会去除一些有用的信息：

SGBMright204

Modify

  1. 2019/05/03 First upload, completed camera calibration and image correction
  2. 2019/05/04 Fixed the difference between the generated image and the original image (because the sequence returned by glob.glob() is not based on the named alphabetical order)
  3. 2019/05/09 Add epilines.py
  4. 2019/05/10 Add calibrate_binocular.py
  5. 2019/05/11 Modify calibrate_binocular.py, add rectification function
  6. 2019/05/12 Add calibrate_binocular.m, which is from Tutorial on Rectification of Stereo Images
  7. 2019/05/14 Add Disparity_BM.py, Disparity_DP.py, Disparity_SGBM.py
  8. 2019/05/15 Add timing function
  9. 2019/05/17 Temporarily stop project update