JokerEyeAdas / AdasSourrondView

The C++ code demo for surround view of car
181 stars 58 forks source link

Steps to find project_keypoints #5

Open derekoutdu opened 11 months ago

derekoutdu commented 11 months ago

Hi, I know that to perform perspective transform we need source points and reference points. Source points is picked manually from undistorted source image. What is the procedure to get reference points that is passed in params.cpp. As it has to resemble the same on final birds eye view output.

Secondly, at what height was the fisheye cameras mounted?

zhk779 commented 11 months ago

You should be able to find out from this repo: https://github.com/neozhaoliang/surround-view-system-introduction

derekoutdu commented 11 months ago

Hi zhk779, thank you for responding. I am following the same repository, although the reference point values are hard coded in the specified repo under params.py

zhk779 commented 11 months ago

In the repo I mentioned, the documentation says that these key points correspond to the pixels of the image one by one. In my opinion, they are determined by the end output‘s resolution and the original image, but it is not necessary to set these points precisely, just make them resonable is okay.(in my proj, I didn't even make any changes). Because eventually you have to make several fine adjustments when you run the code to get projection matrix . These pictures below are my process of selecting key points and the end result, although it is not perfect. (As you can see, when I calibrate these points, the right part is slightly shifted to the right, which requires many manual attempts.) image

a5deab307e821ce79cec83c56ca13ff

In addition, the height of the camera seems to only need to be at a reasonable height, you can refer to the height of the reversing image camera of the family car.