Method for reconstructing 3D information from 2D image and method and apparatus for measuring distance between vehicles using thereof

An apparatus for measuring a distance between vehicles, comprising:

• an image acquiring unit which acquires a 2D image obtained by capturing a non-ego vehicle using a camera mounted in an ego vehicle in driving;
• a bounding box detecting unit which detects a 2D bounding box which encloses the non-ego vehicle in the 2D image;
• a 3D information reconstructing unit which detects a 3D coordinate of at least one 3D vertex of a virtual 3D bounding box which encloses the non-ego vehicle in a 3D space, from a coordinate of a vertex of the 2D bounding box, using unprojection conversion from the 2D image to the 3D space, based on an intrinsic parameter of the camera and an extrinsic parameter with respect to the ego vehicle;
• and
• a distance measuring unit which measures a distance between the ego vehicle and the non-ego vehicle using the 3D coordinate of the at least one 3D vertex,
• wherein the extrinsic parameter is determined with respect to a 3D space coordinate system in which a first axis, a second axis, and a third axis are set according to a length direction, a width direction, and a height direction of the ego vehicle,
• wherein the 3D information reconstructing unit acquires a first 3D coordinate and a second 3D coordinate from a first vertex and a second vertex on the lower edge of the 2D bounding box, by means of the unprojection conversion, acquires a 3D intersection point of a first straight line which passes through the first 3D coordinate and is parallel to the first axis and a second straight line which passes through the second 3D coordinate and is parallel to the second axis, in the 3D space, and acquires a 3D coordinate of a first 3D vertex using a fourth vertex above the second vertex, from the 3D intersection point and the 2D bounding box.

A method for reconstructing 3D information from a 2D image, comprising:

• acquiring a 2D image obtained by capturing a non-ego vehicle using a camera mounted in an ego vehicle in driving;
• detecting a 2D bounding box which encloses the non-ego vehicle in the 2D image;
• and
• detecting a 3D coordinate of at least one 3D vertex of a virtual 3D bounding box which encloses the non-ego vehicle in a 3D space, from a coordinate of a vertex of the 2D bounding box, using unprojection conversion from the 2D image to the 3D space, based on an intrinsic parameter of the camera and an extrinsic parameter with respect to the ego vehicle,
• wherein the extrinsic parameter is determined with respect to a 3D space coordinate system in which a first axis, a second axis, and a third axis are set according to a length direction, a width direction, and a height direction of the ego vehicle,
• wherein the detecting of a 3D coordinate includes:
• acquiring a first 3D coordinate and a second 3D coordinate from a first vertex and a second vertex on the lower edge of the 2D bounding box, by means of the unprojection conversion;
• acquiring a 3D intersection point of a first straight line which passes through the first 3D coordinate and is parallel to the first axis and a second straight line which passes through the second 3D coordinate and is parallel to the second axis, in the 3D space;
• and
• acquiring a 3D coordinate of a first 3D vertex using a fourth vertex above the second vertex, from the 3D intersection point and the 2D bounding box.