Method and system for determining poses of vehicle-mounted cameras for in-road obstacle detection
Poses of a movable camera relative to an environment are obtained by determining point correspondences from a set of initial images and then applying 2-point motion estimation to the point correspondences to determine a set of initial poses of the camera. A point cloud is generated from the set of initial poses and the point correspondences. Then, for each next image, the point correspondences and corresponding poses are determined, while updating the point cloud.
1. A method for determining poses of a camera relative to an environment, wherein the camera is movable, comprising:
determining point correspondences from a set of initial images acquired of the environment by the camera;
applying 2-point motion estimation to the point correspondences to determine a set of initial poses of the camera; and
generating a point cloud from the set of initial poses and the point correspondences, and for each next image, performing the steps of:
determining the point correspondences from the next image;
determining the pose of the camera from the point correspondences of the next image and the point cloud; and
updating the point cloud according to the pose and the point correspondences of the next image, wherein the steps are performed in a processor, and wherein the 2-point motion estimation further comprises:
obtaining projection rays from the point correspondences;
transforming the projection rays in a first camera coordinate system to a first intermediate coordinate system, and a second camera coordinate system to a second intermediate coordinate system;
determining motion between the projection rays in the first and second intermediate coordinate systems using a quadratic equation subject to coplanarity and orthonormality constraints; and
determining motion between the projection rays in the first and second camera coordinate systems using the motion computed in the first and second intermediate coordinate systems.
2. The method of claim 1 , wherein the point cloud is sparse.
3. The method of claim 1 , wherein the applying further comprises:
solving analytically a quadratic equation for the 2-point motion estimation.
4. The method of claim 1 , further comprising:
applying a plane sweep procedure to the images with the poses to produce a dense depth map.
5. The method of claim 3 , wherein the 2-point motion estimation is done in a random sample consensus (RANSAC) framework.
6. The method of claim 5 , wherein hypotheses in the RANSAC framework are generated using at least two point correspondences.
7. The method of claim 1 , wherein the camera is mounted on a vehicle.
8. The method of claim 4 , further comprising:
detecting obstacles using the dense depth map.