Method and apparatus for determining relative pose, device and medium

1. A method for determining a relative pose, comprising:

obtaining millimeter-wave data for obstacles in a scene collected by a millimeter-wave radar sensor in a vehicle and first data for the obstacles in the scene collected by a first sensor in the vehicle, wherein, millimeter-wave radar signal boosters are disposed in the scene as target obstacles and the obstacles comprise the target obstacles;

determining first obstacle data for each target obstacle in a millimeter-wave radar coordinate system and second obstacle data for each target obstacle in a first sensor coordinate system according to the millimeter-wave data and the first data for the obstacles in the scene; and

determining the relative pose between the millimeter-wave radar sensor and the first sensor according to the first obstacle data and the second obstacle data;

wherein determining the relative pose between the millimeter-wave radar sensor and the first sensor according to the first obstacle data and the second obstacle data, comprises:

determining a position error of each target obstacle according to a relative pose variable between the millimeter-wave radar sensor and the first sensor, and first position data and second position data of each target obstacle; and

determining a target function according to the position error of each target obstacle, performing a minimizing process on the target function, and using a relative pose corresponding to a minimized target function as the relative pose between the millimeter-wave radar sensor and the first sensor.

2. The method of claim 1 , before determining the position error of each target obstacle according to the relative pose variable between the millimeter-wave radar sensor and the first sensor, and the first position data and the second position data of each target obstacle, further comprising:

recognizing a first target obstacle included in the scene according to the millimeter-wave data for the obstacles in the scene;

recognizing a second target obstacle included in the scene according to the first data for the obstacles in the scene; and

determining a corresponding relationship between the first target obstacle included in the scene and the second target obstacle included in the scene.

3. The method of claim 2 , wherein recognizing the first target obstacle included in the scene according to the millimeter-wave data for the obstacles in the scene, comprises:

determining velocities and positions of respective objects in the scene according to the millimeter-wave data for the obstacles in the scene, and filtering out objects of which the velocities are not zero;

determining reflection intensities of respective objects in the scene according to the millimeter-wave data for the obstacles in the scene, and filtering out objects of which the reflection intensities are lower than an intensity threshold; and

determining whether the remaining objects are target obstacles according to actual relative distances between different target obstacles and positions of the remaining objects, so as to obtain respective target obstacles included in the obstacle data.

4. The method of claim 2 , wherein, when the first sensor is a laser radar sensor, recognizing the second target obstacle included in the scene according to the first data for the obstacles in the scene, comprises:

detecting a ground based on point-cloud data for the obstacles in the scene, and removing point-cloud data for the ground, the point-cloud data for the obstacles in the scene being the first data;

determining point-cloud data and positions for respective objects in the scene according to a number of the point-clouds and/or a distribution rule of the point-clouds; and

recognizing respective second target obstacles included in the scene based on the point-cloud data and the positions for the objects in the scene.

5. An apparatus for determining a relative pose, comprising:

one or more processors;

a millimeter-wave radar sensor, configured to collect millimeter-wave data for obstacles in a scene;

a first sensor, configured to collect first data for the obstacles in the scene; and

a storage device, configured to store one or more programs,

wherein, when the one or more programs are executed by one or more processors, the one or more processors is caused to implement a method for determining a relative pose, comprising:

obtaining millimeter-wave data for obstacles in a scene collected by a millimeter-wave radar sensor in a vehicle and first data for the obstacles in the scene collected by a first sensor in the vehicle, wherein, millimeter-wave radar signal boosters are disposed in the scene as target obstacles and the obstacles comprise the target obstacles;

determining first obstacle data for each target obstacle in a millimeter-wave radar coordinate system and second obstacle data for each target obstacle in a first sensor coordinate system according to the millimeter-wave data and the first data for the obstacles in the scene; and

determining the relative pose between the millimeter-wave radar sensor and the first sensor according to the first obstacle data and the second obstacle data;

wherein, when the one or more processors are configured to determine the relative pose between the millimeter-wave radar sensor and the first sensor according to the first obstacle data and the second obstacle data, the one or more processors are configured:

to determine a position error of each target obstacle according to a relative pose variable between the millimeter-wave radar sensor and the first sensor, and first position data and second position data of each target obstacle; and

to determine a target function according to the position error of each target obstacle, to perform a minimizing process on the target function, and to use a relative pose corresponding to a minimized target function as the relative pose between the millimeter-wave radar sensor and the first sensor.

6. The apparatus of claim 5 , before the one or more processors determine the position error of each target obstacle according to the relative pose variable between the millimeter-wave radar sensor and the first sensor, and the first position data and the second position data of each target obstacle, the one or more processors are further configured:

to recognize a first target obstacle included in the scene according to the millimeter-wave data for the obstacles in the scene;

to recognize a second target obstacle included in the scene according to the first data for the obstacles in the scene; and

to determine a corresponding relationship between the first target obstacle included in the scene and the second target obstacle included in the scene.

7. The apparatus of claim 6 , wherein when the one or more processors are configured to recognize the first target obstacle included in the scene according to the millimeter-wave data for the obstacles in the scene, the one or more processors are configured:

to determine velocities and positions of respective objects in the scene according to the millimeter-wave data for the obstacles in the scene, and to filter out objects of which the velocities are not zero;

to determine reflection intensities of respective objects in the scene according to the millimeter-wave data for the obstacles in the scene, and to filter out objects of which the reflection intensities are lower than an intensity threshold; and

to determine whether the remaining objects are target obstacles according to actual relative distances between different target obstacles and positions of the remaining objects, so as to obtain respective target obstacles included in the obstacle data.

8. The apparatus of claim 6 , wherein, when the first sensor is a laser radar sensor, and the one or more processors are configured to recognize the second target obstacle included in the scene according to the first data for the obstacles in the scene, the one or more processors are configured:

to detect a ground based on point-cloud data for the obstacles in the scene, and to remove point-cloud data for the ground, the point-cloud data for the obstacles in the scene being the first data;

to determine point-cloud data and positions for respective objects in the scene according to a number of the point-clouds and/or a distribution rule of the point-clouds; and

to recognize respective second target obstacles included in the scene based on the point-cloud data and the positions for the objects in the scene.

9. A non-transitory computer readable storage medium having a computer program stored thereon, wherein, when the computer program is executed by a processor in a vehicle, the processor is caused to:

obtain millimeter-wave data for obstacles in a scene collected by a millimeter-wave radar sensor in the vehicle and first data for the obstacles in the scene collected by a first sensor in the vehicle, wherein, millimeter-wave radar signal boosters are disposed in the scene as target obstacles and the obstacles comprise the target obstacles;

determine first obstacle data for each target obstacle in a millimeter-wave radar coordinate system and second obstacle data for each target obstacle in a first sensor coordinate system according to the millimeter-wave data and the first data for the obstacles in the scene; and

determine the relative pose between the millimeter-wave radar sensor and the first sensor according to the first obstacle data and the second obstacle data;

wherein the processor is further caused to:

determine a position error of each target obstacle according to a relative pose variable between the millimeter-wave radar sensor and the first sensor, and first position data and second position data of each target obstacle; and

determine a target function according to the position error of each target obstacle, perform a minimizing process on the target function, and use a relative pose corresponding to a minimized target function as the relative pose between the millimeter-wave radar sensor and the first sensor.

10. The non-transitory computer readable storage medium of claim 9 , before determining the position error of each target obstacle according to the relative pose variable between the millimeter-wave radar sensor and the first sensor, and the first position data and the second position data of each target obstacle, the processor is further caused to:

recognize a first target obstacle included in the scene according to the millimeter-wave data for the obstacles in the scene;

recognize a second target obstacle included in the scene according to the first data for the obstacles in the scene; and

determine a corresponding relationship between the first target obstacle included in the scene and the second target obstacle included in the scene.

11. The non-transitory computer readable storage medium of claim 10 , wherein the processor is further caused to:

determine velocities and positions of respective objects in the scene according to the millimeter-wave data for the obstacles in the scene, and filter out objects of which the velocities are not zero;

determine reflection intensities of respective objects in the scene according to the millimeter-wave data for the obstacles in the scene, and filter out objects of which the reflection intensities are lower than an intensity threshold; and

determine whether the remaining objects are target obstacles according to actual relative distances between different target obstacles and positions of the remaining objects, so as to obtain respective target obstacles included in the obstacle data.

12. The non-transitory computer readable storage medium of claim 10 , wherein, when the first sensor is a laser radar sensor, the processor is further caused to:

detect a ground based on point-cloud data for the obstacles in the scene, and remove point-cloud data for the ground, the point-cloud data for the obstacles in the scene being the first data;

determine point-cloud data and positions for respective objects in the scene according to a number of the point-clouds and/or a distribution rule of the point-clouds; and

recognize respective second target obstacles included in the scene based on the point-cloud data and the positions for the objects in the scene.