Method and apparatus for recognizing obstacle of vehicle

1. A method for recognizing an obstacle to a vehicle, wherein the vehicle comprises a calibrated lidar sensor and an image capturing apparatus, and the method comprises:

acquiring point cloud data of a target region by using the lidar sensor and acquiring image data of the target region by using the image capturing apparatus;

recognizing obstacle information in the point cloud data by using a preset point cloud recognition model, and labeling a recognition result to obtain a first labeled result;

recognizing obstacle information in the image data by using a preset image recognition model, and labeling a recognition result to obtain a second labeled result;

comparing the first labeled result with the second labeled result to decide whether the first labeled result and the second labeled result are identical; and

in response to deciding that the first labeled result and the second labeled result are not identical, determining a correct labeled result from the first labeled result and the second labeled result and outputting the correct labeled result.

2. The method according to claim 1 , wherein the determining a correct labeled result from the first labeled result and the second labeled result comprises:

determining a difference between the first labeled result and the second labeled result;

determining a first obstacle in the first labeled result indicated by the difference and a second obstacle in the second labeled result indicated by the difference;

determining a labeled volume determined by a three-dimensional coordinate set of the first obstacle and a first distance between the first obstacle and the lidar sensor in the first labeled result;

determining an expected volume range of the first obstacle in the point cloud data based on the first labeled result and the first distance; and

in response to the labeled volume being beyond the expected volume range, determining the first labeled result as being incorrect.

3. The method according to claim 2 , wherein the determining a correct labeled result from the first labeled result and the second labeled result comprises:

determining a labeled pixel area covered by the second obstacle in the second labeled result;

determining a second distance between the second obstacle and the image capturing apparatus based on the first distance, and a calibration parameter between the lidar sensor and the image capturing apparatus;

determining an expected pixel area range of the second obstacle in the image data based on the second labeled result and the second distance; and

in response to the labeled pixel area being beyond the expected pixel area range, determining the second labeled result as being incorrect.

4. The method according to claim 3 , wherein the determining a correct labeled result from the first labeled result and the second labeled result comprises:

in response to the labeled volume being within the expected volume range and the labeled pixel area being within the expected pixel area range, performing following steps:

determining an actual volume determined by the three-dimensional coordinate set of the first obstacle in the point cloud data;

projecting the actual volume into the image data to obtain an expected pixel area corresponding to the actual volume;

deciding whether the labeled pixel area is equal to the expected pixel area; and

determining the first labeled result as being correct when deciding that the labeled pixel area is not equal to the expected pixel area.

5. The method according to claim 4 , wherein the determining a correct labeled result from the first labeled result and the second labeled result comprises:

determining the second labeled result as being correct when deciding that the second pixel area is equal to the expected pixel area.

6. The method according to claim 4 , wherein the determining a correct labeled result from the first labeled result and the second labeled result comprises:

deciding whether the labeled volume is equal to the actual volume; and

determining the first labeled result as being correct when deciding that the labeled volume is equal to the actual volume.

7. The method according to claim 4 , wherein the determining a correct labeled result from the first labeled result and the second labeled result comprises:

determining the second labeled result as being correct when deciding that the labeled volume is not equal to the actual volume.

8. The method according to claim 1 , further comprising at least one of:

in response to the first labeled result being correct, revising the second labeled result based on the first labeled result and retraining the image recognition model by using the revised second labeled result; and

in response to the second labeled result being correct, revising the first labeled result based on the second labeled result and retraining the point cloud recognition model by using the revised first labeled result.

9. An apparatus for recognizing an obstacle to a vehicle, wherein the vehicle comprises a calibrated lidar sensor and an image capturing apparatus, and the apparatus comprises:

at least one processor; and

a memory storing instructions, which when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:

acquiring point cloud data of a target region by using the lidar sensor and acquire image data of the target region by using the image capturing apparatus;

recognizing obstacle information in the point cloud data by using a preset point cloud recognition model, and label a recognition result to obtain a first labeled result;

recognizing obstacle information in the image data by using a preset image recognition model, and label a recognition result to obtain a second labeled result;

deciding whether the first labeled result and the second labeled result are identical by comparing the first labeled result with the second labeled result; and

in response to deciding that the first labeled result and the second labeled result are not identical, determining a correct labeled result from the first labeled result and the second labeled result.

10. The apparatus according to claim 9 , the operations further comprising:

determining a difference between the first labeled result and the second labeled result;

determining a first obstacle in the first labeled result indicated by the difference and a second obstacle in the second labeled result indicated by the difference;

determining a labeled volume determined by a three-dimensional coordinate set of the first obstacle and a first distance between the first obstacle and the lidar sensor in the first labeled result;

determining an expected volume range of the first obstacle in the point cloud data based on the first labeled result and the first distance; and

in response to the labeled volume being beyond the expected volume range, determining the first labeled result as being incorrect.

11. The apparatus according to claim 10 , the operations further comprising:

determining a labeled pixel area covered by the second obstacle in the second labeled result;

determining a second distance between the second obstacle and the image capturing apparatus based on the first distance, and a calibration parameter between the lidar sensor and the image capturing apparatus;

determining an expected pixel area range of the second obstacle in the image data based on the second labeled result and the second distance; and

in response to the labeled pixel area being beyond the expected pixel area range, determining the second labeled result as being incorrect.

12. The apparatus according to claim 11 , the operations further comprising:

in response to the labeled volume being within the expected volume range and the labeled pixel area being within the expected pixel area range, performing following steps:

determining an actual volume determined by the three-dimensional coordinate set of the first obstacle in the point cloud data;

projecting the actual volume into the image data to obtain an expected pixel area corresponding to the actual volume;

deciding whether the labeled pixel area is equal to the expected pixel area; and

determining the first labeled result as being correct when deciding that the labeled pixel area is not equal to the expected pixel area.

13. The apparatus according to claim 12 , the operations further comprising:

determining the second labeled result as being correct when deciding that the second pixel area is equal to the expected pixel area.

14. The apparatus according to claim 12 , the operations further comprising:

deciding whether the labeled volume is equal to the actual volume; and

determining the first labeled result as being correct when deciding that the labeled volume is equal to the actual volume.

15. The apparatus according to claim 14 , the operations further comprising:

determining the second labeled result as being correct when deciding that the labeled volume is not equal to the actual volume.

16. The apparatus according to claim 9 , the operations further comprising at least one of:

in response to the first labeled result being correct, revising the second labeled result based on the first labeled result and retrain the image recognition model by using the revised second labeled result; and

in response to the second labeled result being correct, revising the first labeled result based on the second labeled result and retrain the point cloud recognition model by using the revised first labeled result.

17. A non-transitory computer storage medium storing a computer program, which when executed by one or more processors, cause the one or more processors to perform operations for recognizing an obstacle to a vehicle, wherein the vehicle comprises a calibrated lidar sensor and an image capturing apparatus, and the operations comprises:

acquiring point cloud data of a target region by using the lidar sensor and acquiring image data of the target region by using the image capturing apparatus;

recognizing obstacle information in the point cloud data by using a preset point cloud recognition model, and labeling a recognition result to obtain a first labeled result;

recognizing obstacle information in the image data by using a preset image recognition model, and labeling a recognition result to obtain a second labeled result;

comparing the first labeled result with the second labeled result to decide whether the first labeled result and the second labeled result are identical; and

in response to deciding that the first labeled result and the second labeled result are not identical, determining a correct labeled result from the first labeled result and the second labeled result and outputting the correct labeled result.