Camera calibration method and medium and 3D object reconstruction method and medium using the same

A method of reconstructing a 3-dimensional (3D) object, the method comprising:

• calibrating a camera using feature points in an input image sequence by detecting a feature point of a first frame in the input image sequence, predicting a feature point position in a next frame of the input image sequence corresponding to a detected feature point of the first frame, wherein the predicting of the feature point position is performed hierarchically by generating images of a plurality of resolution levels and predicting the feature of the next frame beginning from a lower resolution level and increasing gradually to higher resolution levels, matching feature points in the next frame based on the predicting of the feature point positions in the next frame, and obtaining 3D points with respect to the matched feature points by initializing a structure and motion of the camera;
• obtaining an outermost surface with respect to an object from two frames among input images, and dividing 3D points belonging to the outermost surface from the 3D points;
• modeling primitive objects included in the input images from the divided 3D points;
• and
• receiving outline information on non-primitive objects through user interaction, and modeling the non-primitive objects from the input images.

A method of reconstructing a non-primitive 3-dimensional (3D) object, the method comprising:

• receiving outline points of a 2-dimensional (2D) silhouette of the non-primitive object desired to be modeled in an input image sequence, wherein the outline points include points that have been selected because they are positioned at inflection points of the silhouette thereby minimizing a quantity of selected outline points;
• extracting 2D skeletons with respect to the 2D silhouette by using triangulation based on the received outline points;
• obtaining an outermost surface of a 3D object from the 2D skeletons and constructing a 3D skeleton by calculating 3D sections from the outermost surface;
• generating a volume model by performing a transform of the 3D skeleton into a voxel space;
• and
• constructing a 3D polygon model by performing the transform inversely for the volume model.

A method of reconstructing a 3-dimensional (3D) object at a roadside, comprising:

• arranging a hexahedron with respect to an identical and primitive object in two images of the roadside viewed from different camera orientations, respectively;
• extracting feature points with respect to the hexahedron;
• calibrating the camera using the feature points and obtaining 3D points for the images;
• modeling the primitive object from the 3D points;
• and
• receiving outline information of a non-primitive object through user interaction and modeling the non-primitive object from the images.

At least one non-transitory medium comprising computer readable code to control at least one processing element in a computer to implement a method of calibrating a camera, the method comprising:

• setting a feature track by tracking and matching features in an input image sequence, wherein the tracking and matching of the features comprises;
• detecting a feature point of a first frame of the input image sequence;
• predicting a feature point position in a next frame corresponding to the detected feature point of the first frame, wherein the predicting of the feature point position is performed hierarchically by generating images of a plurality of resolution levels and predicting the feature of the next frame by beginning from a lower resolution level and increasing sequentially to higher resolution levels;
• and
• matching feature points in the next frame based on the predicting of the feature point positions in the next frame;
• estimating 3-dimensional (3D) points in relation to the matched feature points by initializing a structure and motion of the camera;
• estimating a pose of the camera using 3D points and refining the structure of the camera based on the estimated pose;
• and
• refining the pose of the camera.

At least one non-transitory medium comprising computer readable code to control at least one processing element in a computer to implement a method of reconstructing a 3-dimensional (3D) object, the method comprising:

• calibrating a camera using feature points in an input image sequence by detecting a feature point of a first frame in the input image sequence, predicting a feature point position in a next frame of the input image sequence corresponding to a detected feature point of the first frame, wherein the predicting of the feature point position is performed hierarchically by generating images of a plurality of resolution levels and predicting the feature of the next frame beginning from a lower resolution level and increasing gradually to higher resolution levels, matching feature points in the next frame based on the predicting of the feature point positions in the next frame, and obtaining 3D points with respect to the matched feature points by initializing a structure and motion of the camera;
• obtaining an outermost surface with respect to an object from two frames among input images, and dividing 3D points belonging to the outermost surface from the 3D points;
• modeling primitive objects included in the input images from the divided 3D points;
• and
• receiving outline information on a non-primitive objects through user interaction, and modeling the non-primitive objects from the input images.

At least one non-transitory medium comprising computer readable code to control at least one processing element in a computer to implement a method of reconstructing a non-primitive 3-dimensional (3D) object, the method comprising:

• receiving outline points of a silhouette of the non-primitive object desired to be modeled in an input image sequence, wherein the outline points include points that have been selected because they are positioned at inflection points of the silhouette thereby minimizing a quantity of selected outline points;
• extracting 2-dimensional (2D) skeletons with respect to the silhouette by using triangulation based on the received outline points;
• obtaining an outermost surface of a 3D object from the 2D skeletons and constructing a 3D skeleton by calculating 3D sections from the outermost surface;
• generating a volume model by performing a transform of the 3D skeleton into a voxel space;
• and
• constructing a 3D polygon model by performing the transform inversely for the volume model.

At least one non-transitory medium comprising computer readable code to control at least one processing element in a computer to implement a method of reconstructing a 3-dimensional (3D) object at a roadside, the method comprising:

• arranging a hexahedron with respect to an identical and primitive object in two images of the roadside viewed from different camera orientations, respectively;
• extracting feature points with respect to the hexahedron;
• calibrating the camera using the feature points and obtaining 3D points for the images;
• modeling the primitive object from the 3D points;
• and
• receiving outline information of a non-primitive object through user interaction and modeling the non-primitive object from the images.

A camera calibration method comprising:

• detecting one or more feature points from each of a plurality of consecutive images, wherein the detecting comprises;
• detecting a feature point of a first frame of the plurality of consecutive images;
• predicting a feature point position in a next frame corresponding to the detected feature point of the first frame, wherein the predicting of the feature point position is performed hierarchically by generating images of a plurality of resolution levels and predicting the feature of the next frame by beginning from a lower resolution level and increasing sequentially to higher resolution levels;
• and
• matching feature points in the next frame based on the predicting of the feature point position in the next frame;
• estimating 3-dimensional (3D) points in relation to the matched feature points by initializing a structure and motion of the camera;
• estimating a pose of the camera using the 3D points and refining the structure of the camera based on the estimated pose;
• and
• refining the pose of the camera based on the refined structure of the camera.