使用 opencv 的特定相机设置和校准过程
这是我的第一篇文章，很高兴成为这个社区的一员。
我有两个网络摄像头，想用它来检测白色网球中心的二维坐标。 并找到中心的 3d 坐标。我的相机是这样设置的。有一个圆形区域 半径为 7 英尺，并且摄像机放置在彼此相对的一端，这意味着如果摄像机 1 放置在圆形区域的 0 度处，然后将相机 2 放置在同一区域的 180 度处 因此它们在圆形区域的圆周上彼此完全相对的一条直线上。
我需要校准相机并需要找到内在和外在参数。我正在使用 opencv 这。
我可以使用 cvStereoCalibrate() 进行此相机设置吗？
我问这个问题是因为如果你查看相机设置，你会发现相机 1 上有一个点并且 相机 2 捕获的图像与两个相机的极点共线。所以极线是一个点。 这会成为校准程序的问题吗？如果是的话可以告诉我如何解决这个问题吗？
This is my first post and am delighted to be a part of this community.
I've two webcams and want to use it to detect 2d coordinates of the center of the white tennis balls.
And find the 3d coordinate of the centers. My cameras are setup like this. There's a circular region
of radius 7 feet and the cameras are placed on opposite end of each other which means if camera1 is
placed at 0 degree of the circular region then camera2 is placed at 180 degree of that same region
so they are on a straight line exactly opposite of each other on the circumference of the circular region.
I need to calibrate the cameras and need to find intrinsic and extrinsic parameters.I'm using opencv for
this.
Can I use cvStereoCalibrate() for this camera setup?
I'm asking this because if you look at the camera setup you will see that there's a point on camera1 and
camera2's captured image that is collinear with epipoles of both cameras. So the epipolar line is a point.
Will this be an issue for calibration procedure? If yes is it possible to tell me how to solve this problem?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
这是立体摄像机完全内倾的摄像机设置（真的是斗鸡眼:)。要校准，您需要每个图像的对应点。你只能在边缘找到这些......在每个相机的周边视觉中一直存在，即使如此，我认为它们也会反映算法所期望的（必须考虑这一点）。我还没有尝试过这个，但我认为这会将图像扭曲成甜甜圈，假设它没有完全混淆算法以至无法识别。我认为，立体对应算法肯定会出现问题，因为一处较远，另一处较近。
This is a camera setup with the stereo cameras toed-in completely (really cross-eyed :). To calibrate you need correspondence points from each image. You can only find these around the edges ... all the way around in the peripheral vision of each camera and even then I think they would be mirrored from what the algorithm would expect (have to think about that one). I have not tried this, but I think this would warp the image into a doughnut assuming it didn't completely confuse the algorithm beyond recognition. The stereo correspondence algorithms would definitely have problems with this, I think, Because also far in one is near in the other.