Top-view image generating apparatus and method

1. A top-view image generating method comprising:

detecting a ground feature point from at least one of front, rear, left, and right images respectively provided from a plurality of cameras installed in a vehicle;

estimating a first posture angle of a camera providing the at least one image, based on the detected ground feature point;

generating a first top-view image by rotation-converting the at least one image, based on the estimated first posture angle;

detecting a first lane from the first top-view image;

detecting a second lane succeeding the first lane from another image, from which the ground feature point is not detected, of the front, rear, left, and right images;

estimating a second posture angle of a camera providing the other image from which the second lane is detected, based on a correspondence relationship between a feature pattern of the detected first lane and a feature pattern of the detected second lane;

generating a second top-view image by rotation-converting the other image, based on the estimated second posture angle; and

synthesizing the first top-view image and the second top-view image to generate a final top-view image.

2. The top-view image generating method of claim 1 , further comprising:

before the detecting of the ground feature point,

obtaining the at least one image from a camera, installed in the vehicle for an optical axis to have a first angle with respect to a ground, among the plurality of cameras; and

obtaining the other image from a camera, installed in the vehicle for the optical axis to have a second angle less than the first angle with respect to the ground, among the plurality of cameras.

3. The top-view image generating method of claim 1 , wherein the estimating of the first posture angle comprises:

calculating an amount of pitch-direction rotation and an amount of yaw-direction rotation so that a geometric relationship between a previous ground feature point detected from the at least one image obtained from the plurality of cameras at a previous time and a current ground feature point detected from the at least one image obtained from the plurality of cameras at a current time satisfies a predetermined geometric condition;

generating a middle top-view image by rotation-converting the at least one image by the calculated amount of pitch-direction rotation and the calculated amount of yaw-direction rotation;

calculating a movement vector of the ground feature point by using a previous ground feature point detected from a middle top-view image generated at a previous time and a current ground feature point detected from a middle top-view image generated at a current time, and calculating a prediction movement vector of the ground feature point based on a movement of the vehicle;

calculating an amount of roll-direction rotation allowing a direction of the movement vector to match a direction of the prediction movement vector; and

estimating the first posture angle including the amount of pitch-direction rotation, the amount of yaw-direction rotation, and the amount of roll-direction rotation.

4. The top-view image generating method of claim 1 , wherein the estimating of the second posture angle comprises estimating a second posture angle of a camera providing the other image, for a feature pattern of the detected second lane to match a feature pattern of the first lane shown in the first top-view image.

5. The top-view image generating method of claim 4 , wherein the feature pattern comprises a lane direction and a lane thickness.

6. The top-view image generating method of claim 5 , wherein the estimating of the second posture angle comprises:

calculating an amount of pitch-direction rotation and an amount of yaw-direction rotation allowing a lane direction and a lane thickness of the second lane detected from the other image to match a lane direction and a lane thickness of the first lane detected from the first top-view image;

generating a middle top-view image by rotation-converting the other image by the calculated amount of pitch-direction rotation and the calculated amount of yaw-direction rotation;

calculating a movement vector of a lane feature point by using a previous lane feature point detected from a middle top-view image generated at a previous time and a current lane feature point detected from a middle top-view image generated at a current time, and calculating a prediction movement vector of the lane feature point based on a movement of the vehicle;

calculating an amount of roll-direction rotation allowing a direction of the movement vector to match a direction of the prediction movement vector; and

estimating the second posture angle including the amount of pitch-direction rotation, the amount of yaw-direction rotation, and the amount of roll-direction rotation.

7. The top-view image generating method of claim 1 , wherein the generating of the final top-view image comprises:

arranging the first top-view image and the second top-view image at designated positions in a top-view image coordinate system;

generating the final top-view image by moving the first top-view image or the second top-view image in up, down, left, and right directions so that a lane of the first top-view image and a lane of the second top-view image is displayed as one continuous lane.

8. A top-view image generating apparatus comprising:

a ground feature point detector detecting a ground feature point from at least one of front, rear, left, and right images respectively provided from a plurality of cameras installed in a vehicle;

a first posture angle estimator estimating a first posture angle of a camera providing the at least one image, based on the detected ground feature point;

a first rotation converter generating a first top-view image by rotation-converting the at least one image, based on the estimated first posture angle;

a lane detector detecting a first lane from the first top-view image and detecting a second lane succeeding the first lane from another image, from which the ground feature point is not detected, of the front, rear, left, and right images;

a second posture angle estimator estimating a second posture angle of a camera providing the other image from which the second lane is detected, based on a correspondence relationship between a feature pattern of the detected first lane and a feature pattern of the detected second lane;

a second rotation converter generating a second top-view image by rotation-converting the other image, based on the estimated second posture angle; and

an image synthesizer synthesizing the first top-view image and the second top-view image to generate a final top-view image.

9. The top-view image generating apparatus of claim 8 , wherein

the at least one image is obtained from a camera, installed in the vehicle for an optical axis to have a first angle with respect to a ground, among the plurality of cameras, and

the other image is obtained from a camera, installed in the vehicle for an optical axis to have a second angle less than the first angle with respect to the ground, among the plurality of cameras.

10. The top-view image generating apparatus of claim 9 , wherein

the at least one image is an image where an extraction success rate for the ground feature point is high, based on an installation environment where a camera is installed in the vehicle for an optical axis to have a first angle with respect to a ground, and

the other image is an image where an extraction fail rate for the ground feature point is high, based on an installation environment where a camera is installed in the vehicle for an optical axis to have a second angle less than the first angle with respect to the ground.

11. The top-view image generating apparatus of claim 8 , wherein the first posture angle estimator comprises:

a first rotation amount calculator calculating an amount of pitch-direction rotation and an amount of yaw-direction rotation so that a geometric relationship between a previous ground feature point detected from the at least one image obtained from the plurality of cameras at a previous time and a current ground feature point detected from the at least one image obtained from the plurality of cameras at a current time satisfies a predetermined geometric condition;

a rotation converter generating a middle top-view image by rotation-converting the at least one image by the calculated amount of pitch-direction rotation and the calculated amount of yaw-direction rotation;

a vector calculator calculating a movement vector of the ground feature point by using a previous ground feature point detected from a middle top-view image generated at a previous time and a current ground feature point detected from a middle top-view image generated at a current time, and calculating a prediction movement vector of the ground feature point based on a movement of the vehicle;

a second rotation amount calculator calculating an amount of roll-direction rotation allowing a direction of the movement vector to match a direction of the prediction movement vector; and

a conversion matrix generator estimating the first posture angle including the amount of pitch-direction rotation, the amount of yaw-direction rotation, and the amount of roll-direction rotation.

12. The top-view image generating apparatus of claim 8 , wherein

the second posture angle estimator estimates a second posture angle of a camera providing the other image, for a feature pattern of the detected second lane to match a feature pattern of the first lane shown in the first top-view image, and

the feature pattern comprises a lane direction and a lane thickness.

13. The top-view image generating apparatus of claim 12 , wherein the second posture angle estimator comprises:

a first rotation amount calculator calculating an amount of pitch-direction rotation and an amount of yaw-direction rotation allowing a lane direction and a lane thickness of the second lane detected from the other image to match a lane direction and a lane thickness of the first lane detected from the first top-view image;

a rotation converter generating a middle top-view image by rotation-converting the other image by the calculated amount of pitch-direction rotation and the calculated amount of yaw-direction rotation;

a vector calculator calculating a movement vector of a lane feature point by using a previous lane feature point detected from a middle top-view image generated at a previous time and a current lane feature point detected from a middle top-view image generated at a current time, and calculating a prediction movement vector of the lane feature point based on a movement of the vehicle;

a second rotation amount calculator calculating an amount of roll-direction rotation allowing a direction of the movement vector to match a direction of the prediction movement vector; and

a conversion matrix generator generating a conversion matrix representing the second posture angle, including the amount of pitch-direction rotation, the amount of yaw-direction rotation, and the amount of roll-direction rotation, in a matrix type.