简体繁体 English

3D object 从 2D 图像定位

[英]3D object localization from 2D image

原文 2019-09-26 09:47:45 0 1 python-3.x/ opencv/ object-detection

I have a 2D image with vehicles localized by 2D bounding boxes in it.我有一个 2D 图像，其中的车辆由 2D 边界框定位。 Also I have the 3x3 homography matrix and the reprojection error from camera calibration.我也有 3x3 单应矩阵和相机校准的重投影误差。 Are these information enough to find the 3D position of the vehicles?这些信息是否足以找到车辆的3D position？

This is my very first question in here, so please let me know if I need to provide more information.这是我在这里的第一个问题，所以如果我需要提供更多信息，请告诉我。

1 个解决方案

It is impossible to get 3D coordinates from 2D image.从二维图像中获取 3D 坐标是不可能的。 You can map each pixel to the ray that comes from camera nodal point and goes through that pixel but to get 3D coordinates you need also the depth value in this pixel.您可以将每个像素 map 指向来自相机节点并穿过该像素的光线，但要获得 3D 坐标，您还需要该像素中的深度值。 Consider following example: The vehicle of the size S is located at the distance d to the camera.考虑以下示例：尺寸为S的车辆位于距离相机d处。 Similar vehicle that is two times bigger (size 2S ) is located at the distance 2d to the camera.两倍大的类似车辆（尺寸2S ）位于距相机2d的距离处。 They are projected exactly the same on the image.它们在图像上的投影完全相同。 So you can get 3D positions up to unknown scale factor only.因此，您可以获得 3D 位置，直到未知比例因子。 However, you can estimate the location of each vehicle that if you know the real world size of them.但是，如果您知道它们的真实世界大小，则可以估计每辆车的位置。