Obstacle type recognizing method and apparatus, device and storage medium
View Patent ↗The present disclosure provides an obstacle type recognizing method and apparatus, a device and a storage medium, wherein the method comprises: obtaining 3D point cloud data corresponding to a to-be-recognized obstacle; mapping the 3D point cloud data and its dimension data to a four-dimensional array; recognizing a type of the obstacle through a deep learning algorithm based on the four-dimensional array. The solution of the present disclosure can be applied to determine the type of the obstacle such as a person, a bicycle or a motor vehicle; and recognize a small-sized vehicle, a medium-sized vehicle and a large-sized vehicle; and improve the accuracy of a recognition result.
1. An obstacle type recognizing method, wherein the method comprises:
obtaining 3D point cloud data corresponding to a to-be-recognized obstacle;
mapping the 3D point cloud data and its dimension data to a four-dimensional array;
recognizing a type of the obstacle through a deep learning algorithm based on the four-dimensional array,
wherein the four-dimensional array is stored in a database in a C*H*W array format, wherein C=4 and includes an R channel, a G channel and a B channel and a dimension feature channel, and
wherein the mapping the 3D point cloud data and its dimension data to the four-dimensional array comprises:
mapping the 3D point cloud data from a first view angle to an R channel of the four-dimensional array;
mapping the 3D point cloud data from a second view angle to a G channel of the four-dimensional array,
mapping the 3D point cloud data from a third view angle to a B channel of the four-dimensional array;
mapping the dimension data of the 3D point cloud data to the dimension feature channel of the four-dimensional array; and
generating the four-dimensional array according to mapping results.
2. The method according to claim 1 , wherein
the method further comprises:
obtaining each obstacle detected from the 3D point cloud data obtained by scanning;
considering each detected obstacle as the to-be-recognized obstacle;
wherein the 3D point cloud data is obtained by scanning an environment around a driverless vehicle.
3. The method according to claim 1 , wherein
the first view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the second view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the third view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the first view angle, the second view angle and the third view angle are different view angles.
4. The method according to claim 1 , wherein
the dimension data of the 3D point cloud data are values of length, width, height and volume after normalization.
5. The method according to claim 4 , wherein the mapping the dimension data of the 3D point cloud data to the dimension feature channel of the four-dimensional array further comprises:
respectively mapping values of length, width, height and volume after normalization to different geometrical shapes of the dimension feature channel.
6. A computer device, comprising a memory, a processor and a computer program which is stored on the memory and runs on the processor, wherein the processor, upon executing the program, implements the following operation:
obtaining 3D point cloud data corresponding to a to-be-recognized obstacle;
mapping the 3D point cloud data and its dimension data to a four-dimensional array;
recognizing a type of the obstacle through a deep learning algorithm based on the four-dimensional array,
wherein the four-dimensional array is stored in a database in a C*H*W array format, wherein C=4 and includes an R channel, a G channel and a B channel and a dimension feature channel, and
wherein the mapping the 3D point cloud data and its dimension data to the four-dimensional array comprises:
mapping the 3D point cloud data from a first view angle to an R channel of the four-dimensional array;
mapping the 3D point cloud data from a second view angle to a G channel of the four-dimensional array,
mapping the 3D point cloud data from a third view angle to a B channel of the four-dimensional array;
mapping the dimension data of the 3D point cloud data to the dimension feature channel of the four-dimensional array; and
generating the four-dimensional array according to mapping results.
7. The computer device according to claim 6 , wherein
the operation further comprises:
obtaining each obstacle detected from the 3D point cloud data obtained by scanning;
considering each detected obstacle as the to-be-recognized obstacle;
wherein the 3D point cloud data is obtained by scanning an environment around a driverless vehicle.
8. The computer device according to claim 6 , wherein
the first view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the second view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the third view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the first view angle, the second view angle and the third view angle are different view angles.
9. The computer device according to claim 6 , wherein
the dimension data of the 3D point cloud data are values of length, width, height and volume after normalization.
10. The computer device according to claim 9 , wherein the mapping the dimension data of the 3D point cloud data to the dimension feature channel of the four-dimensional array further comprises:
respectively mapping values of length, width, height and volume after normalization to different geometrical shapes of the dimension feature channel.
11. A non-transitory computer-readable storage medium on which a computer program is stored, wherein the program, when executed by a processor, implements the following operation:
obtaining 3D point cloud data corresponding to a to-be-recognized obstacle;
mapping the 3D point cloud data and its dimension data to a four-dimensional array;
recognizing a type of the obstacle through a deep learning algorithm based on the four-dimensional array,
wherein the four-dimensional array is stored in a database in a C*H*W array format, wherein C=4 and includes an R channel, a G channel and a B channel and a dimension feature channel, and
wherein the mapping the 3D point cloud data and its dimension data to the four-dimensional array comprises:
mapping the 3D point cloud data from a first view angle to an R channel of the four-dimensional array,
mapping the 3D point cloud data from a second view angle to a G channel of the four-dimensional array,
mapping the 3D point cloud data from a third view angle to a B channel of the four-dimensional array;
mapping the dimension data of the 3D point cloud data to the dimension feature channel of the four-dimensional array; and
generating the four-dimensional array according to mapping results.
12. The non-transitory computer-readable storage medium according to claim 11 , wherein
the operation further comprises:
obtaining each obstacle detected from the 3D point cloud data obtained by scanning;
considering each detected obstacle as the to-be-recognized obstacle;
wherein the 3D point cloud data is obtained by scanning an environment around a driverless vehicle.
13. The non-transitory computer-readable storage medium according to claim 11 , wherein
the first view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the second view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the third view angle is one of the following: a top view angle, a vehicle head front view angle, and a left view angle;
the first view angle, the second view angle and the third view angle are different view angles.
14. The non-transitory computer-readable storage medium according to claim 11 , wherein
the dimension data of the 3D point cloud data are values of length, width, height and volume after normalization.
15. The non-transitory computer-readable storage medium according to claim 14 , wherein the mapping the dimension data of the 3D point cloud data to the dimension feature channel of the four-dimensional array further comprises:
respectively mapping values of length, width, height and volume after normalization to different geometrical shapes of the dimension feature channel.