Device and method for detecting 3D object

1. An apparatus for detecting a three-dimensional object, the apparatus comprising:

a camera configured for obtaining a 2D image; and

a controller electrically connected to the camera,

wherein the controller is configured to:

extract 2D information and 3D information for at least one object from the 2D image; and

detect 3D object information by utilizing the 2D information and the 3D information,

wherein the 2D information includes 2D center coordinates, a width, and a height of a 2D anchor box,

wherein the 3D information includes 3D center coordinates, a width, a height, a length, and a yaw of a 3D anchor box, and

wherein the controller is further configured to:

generate the 2D anchor box on the 2D image;

generate the 3D anchor box by averaging 3D information of objects fitted to the 2D anchor box;

learn a difference between the 2D anchor box, the 3D anchor box, and object information;

select the 3D information using argmax; and

output the 3D object information by sequentially performing decode, softmax, sort, and non-maximum suppression (NMS) on the selected 3D information.

2. The apparatus of claim 1 , wherein the controller is configured to match the 2D information with the 3D information on the 2D image and to determine 3D position information of an object matching the 3D information based on a result of the matching.

3. The apparatus of claim 2 , wherein the controller is configured to:

determine distances between corner points of the 2D anchor box and corner points of the 3D anchor box; and

select two pairs of corner points with a shortest determined distance among the determined distances.

4. The apparatus of claim 3 , wherein the controller is configured to determine the 3D center coordinates of the 3D anchor box using the selected two pairs of corner points.

5. The apparatus of claim 1 , wherein the controller is configured to remove a corresponding 3D information when a result of scaling a possibility that the 3D information is a background with a predetermined ratio exceeds a possibility that the 3D information is an object.

6. The apparatus of claim 1 , wherein the camera is a monocular camera.

7. A method for detecting a three-dimensional object, the method comprising:

obtaining, by a controller, a 2D image using a camera;

extracting, by the controller, 2D information and 3D information for at least one object from the 2D image; and

detecting, by the controller, 3D object information by utilizing the 2D information and the 3D information,

wherein the 2D information includes 2D center coordinates, a width, and a height of a 2D anchor box,

wherein the 3D information includes 3D center coordinates, a width, a height, a length, and a yaw of a 3D anchor box,

wherein the extracting of the 2D information and the 3D information includes:

creating, by the controller, the 2D anchor box on the 2D image;

creating, by the controller, the 3D anchor box by averaging 3D information of objects fitted to the 2D anchor box;

learning, by the controller, a difference between the 2D anchor box, the 3D anchor box, and object information, and

wherein the detecting of the 3D object information includes:

selecting, by the controller, the 3D information using argmax; and

outputting, by the controller, the 3D object information by sequentially performing decode, softmax, sort, and non-maximum suppression (NMS) on the selected 3D information.

8. The method of claim 7 , wherein the detecting of the 3D object information includes:

matching, by the controller, the 2D information with the 3D information on the 2D image; and

determining, by the controller, 3D position information of an object matching the 3D information based on a result of the matching.

9. The method of claim 8 , wherein the matching of the 2D information with the 3D information includes:

determining, by the controller, distances between corner points of the 2D anchor box and corner points of the 3D anchor box; and

selecting, by the controller, two pairs of corner points with a shortest determined distance among the determined distances.

10. The method of claim 9 , wherein the determining of the 3D position information of the object includes:

determining, by the controller, the 3D center coordinates of the 3D anchor box using the selected two pairs of corner points.

11. The method of claim 7 , wherein the selecting of the 3D information includes:

removing, by the controller, a corresponding 3D information when a result of scaling a possibility that the 3D information is a background with a predetermined ratio exceeds a possibility that the 3D information is an object.