search

neozhaoliang / surround-view-system-introduction

A full Python implementation for real car surround view system
MIT License
858 stars 303 forks source link

About the splicing of 4 pictures #80

Closed tigerdhl closed 6 months ago

tigerdhl commented 7 months ago

Thank you for sharing. But there is a question that I don’t understand. For example, when splicing two images, top and left, how the grids can be aligned neatly, the usual method is through feature matching, but the author does not seem to mention this step, only smoothing the splicing after the splicing is completed. @neozhaoliang

neozhaoliang commented 7 months ago

Sorry I don't know much about feature matching. Do you think it can stitch the images in real time?

tigerdhl commented 7 months ago

Sorry I don't know much about feature matching. Do you think it can stitch the images in real time?

What I mean is how the two corrected images from adjacent fisheye cameras are spliced ​​together. I see that your documentation directly talks about the transition problem after splicing.

neozhaoliang commented 7 months ago

They can match together because, at the projection calibration step, you specified the parameters to project the images.

tigerdhl commented 6 months ago

They can match together because, at the projection calibration step, you specified the parameters to project the images.

ok,I get it. For example,the four pixel points in the BEV image is Well designed？

neozhaoliang commented 6 months ago

You need to choose >=4 points to make sure your projected image covers the right region on the ground. The feature-matching method should also work as long as it can give the correct projection matrix. However, I'm not sure how well that method behaves because the projected images are taken by the front and left cameras respectively, they are not taken by the same camera from different viewpoints. So I doubt a more advanced method should be used here.