Method for labelling an epipolar-projected 3D image
A method for labeling a 3D image of a scene acquired by a 3D sensor including: receiving: a 2D image of the same scene acquired by a camera, coordinates, in the 2D image, of a set of pixels delineating the region of interest, and of a reference point belonging to the region of interest, and determining the depth of the reference point in a coordinate system associated with the camera; assigning, to the pixels delineating the region of interest in the 2D image, a depth corresponding to the depth of the reference point; computing the coordinates, in the 3D image, of the pixels delineating the region of interest, based on the coordinates of the pixels delineating the region of interest in the 2D image and on the depth assigned to the pixels delineating the region of interest.
1. A method for labeling a 3D image of a scene acquired by a 3D sensor comprising identifying at least one region of interest in the 3D image, the method being implemented by a computer and comprising:
receiving:
a 2D image of the same scene, acquired by a camera,
coordinates, in the 2D image, of a set of pixels delineating the region of interest,
coordinates, in the 2D image, of a reference point belonging to the region of interest, and
data relating to the relative position and relative orientation of the camera with respect to the 3D sensor,
determining the depth of the reference point in a coordinate system associated with the camera, said step comprising:
based on the coordinates of the reference point in the 2D image, determining the two-dimensional coordinates of a plurality of first points in the 3D image, each first point corresponding to a possible position of the reference point in the 3D image,
obtaining, for each first point, a third depth coordinate with respect to the 3D sensor,
for each first point of the 3D image, obtaining the coordinates of the corresponding point in the 2D image, based on the depth coordinate of the first point,
selecting, in the 2D image, the first point closest to the reference point, and,
assigning, to the reference point, a depth corresponding to the depth of the first selected point,
assigning, to the pixels delineating the region of interest in the 2D image, a depth corresponding to the depth assigned to the reference point, and
computing the coordinates, in the 3D image, of the pixels delineating the region of interest, based on the coordinates of the pixels delineating the region of interest in the 2D image, on the depth assigned to the pixels delineating the region of interest and on the data relating to the relative position and relative orientation of the camera with respect to the 3D sensor.
2. The method as claimed in claim 1 , wherein:
determining the two-dimensional coordinates of the plurality of first points in the 3D image comprises:
assigning a maximum depth to the reference point,
assigning a minimum depth to the reference point,
computing two-dimensional coordinates, in the 3D image, of a first furthest point corresponding to the reference point to which the maximum depth was assigned,
computing two-dimensional coordinates, in the 3D image, of a first closest point corresponding to the reference point to which the minimum depth was assigned, and
determining the two-dimensional coordinates, in the 3D image, of at least one first point located between the first closest point and the first furthest point.
3. The method as claimed in claim 2 , wherein determining the two-dimensional coordinates, in the 3D image, of at least one first point located between the first closest point and the first furthest point comprises determining two-dimensional coordinates of at least one point located, in the 3D image, on a segment connecting the first furthest point and the first closest point.
4. The method as claimed in claim 2 , wherein determining the two-dimensional coordinates, in the 3D image, of at least one first point located between the first closest point and the first furthest point comprises:
assigning, to the reference point, at least an intermediate depth comprised between the maximum depth and the minimum depth, and
computing two-dimensional coordinates, in the 3D image, of a first point corresponding to the reference point to which the intermediate depth was assigned.
5. The method as claimed in claim 1 , wherein determining the two-dimensional coordinates of the plurality of first points in the 3D image comprises:
assigning, to the reference point, a depth of one of a minimum depth or maximum depth,
computing two-dimensional coordinates, in the 3D image, of a first end point corresponding to the reference point to which one of a minimum or maximum depth was assigned, and
computing, based on the horizontal resolution of the 3D sensor and on the distance between the 3D sensor and the camera, a maximum disparity corresponding to a maximum number of pixels in the 3D image separating the first end point and a point in the 3D image corresponding to the reference point to which the other, minimum or maximum, depth was assigned,
determining the two-dimensional coordinates of each point of the 3D image comprised between the first end point and a point laterally separated from the first end point by the maximum disparity.
6. The method as claimed in claim 1 , wherein the set of pixels delineating the region of interest in the 2D image comprises four pixels delineating a rectangle.
7. The method as claimed in claim 1 , wherein the region of interest, in the 2D image, has a predetermined geometric shape and the method further comprises a step of defining a region of interest in the 3D image having the same geometric shape as the region of interest in the 2D image.
8. A computer program comprising instructions for implementing the method as claimed in claim 1 when this program is executed by a computer.
9. A non-transient computer-readable storage medium on which is stored a program for implementing the method as claimed in claim 1 when this program is executed by a computer.