Rapid point cloud alignment and classification with basis set learning
地域：
WA
WA Seattle
摘要
A system configured to process an input point cloud, which represents an object using unstructured data points, to generate a feature vector that has an ordered structure and a fixed length. The system may process the input point cloud using a basis point set to generate the feature vector. For example, for each basis point in the basis point set, the system may identify a closest data point in the point cloud data and store a distance value or other information associated with the closest data point in the feature vector. The system may process the feature vector using a trained model to generate output data, such as performing point cloud registration to generate mesh data, point cloud classification to generate classification data, and/or the like.
說明書
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
With the advancement of technology, the use and popularity of electronic devices has increased considerably. Electronic devices are commonly used to capture and process image data.
For a more complete understanding of the present disclosure, reference is now made to the following description taken in conjunction with the accompanying drawings.
權利要求
What is claimed is:1. A computer-implemented method, the method comprising:receiving a plurality of unstructured data points representing an object, the plurality of unstructured data points including a first data point at a first position in a three-dimensional (3D) coordinate system and a second data point at a second position in the 3D coordinate system, the plurality of unstructured data points lacking organization between the first data point and the second data point; determining a plurality of structured data points with which to compare the plurality of unstructured data points, the plurality of structured data points including a third data point at a third position in the 3D coordinate system and a fourth data point at a fourth position in the 3D coordinate system, the plurality of structured data points having a known spatial arrangement between the third data point and the fourth data point; determining, using the plurality of unstructured data points and the plurality of structured data points, a first feature vector for the plurality of unstructured data points, wherein the determining the first feature vector comprises:determining that the third data point corresponds to a first element in the first feature vector, determining a first distance value between the third position and the first position, determining a second distance value between the third position and the second position, determining that the first distance value is lower than the second distance value, indicating that the first data point is nearest to the third data point of the plurality of unstructured data points, and storing the first distance value in the first element in the first feature vector; and processing, using a first neural network, the first feature vector to generate output data corresponding to a three-dimensional (3D) model of the object. 2. The computer-implemented method of claim 1 , further comprising:processing, using a second neural network, the first feature vector to generate second output data that includes a plurality of probability values including a first probability value; determining that the first probability value is highest of the plurality of probability values; identifying a first classification label of a plurality of classification labels that corresponds to the first probability value; and associating the first classification label with the object. 3. The computer-implemented method of claim 1 , wherein the processing further comprises:processing, using the first neural network, the first feature vector to generate output data corresponding to the 3D model of the object in a first pose that is different from a second pose represented by the plurality of unstructured data points. 4. The computer-implemented method of claim 1 , further comprising:receiving a second plurality of unstructured data points representing a second object, the second plurality of unstructured data points including a fifth data point at a fifth position in the 3D coordinate system; determining, using the second plurality of unstructured data points and the plurality of structured data points, a second feature vector for the second plurality of unstructured data points, a first element in the second feature vector indicating a third distance value between the third position and the fifth position, indicating that the fifth data point is nearest to the third data point of the second plurality of unstructured data points; and training the first neural network using the first feature vector and the second feature vector. 5. A computer-implemented method, the method comprising:receiving point cloud data representing an object, the point cloud data including a first data point at a first position in a three-dimensional (3D) coordinate system; determining basis point data with which to compare the point cloud data, the basis point data including a first number of data points that include a second data point at a second position in the 3D coordinate system and a third data point at a third position in the 3D coordinate system; determining, using the point cloud data and the basis point data, a plurality of distance values, wherein the determining further comprises:determining, using the first data point and the second data point, a first distance value corresponding to a first distance from the second position to the first position, determining that the first distance value represents a first minimum distance from the second position to one of a plurality of data points in the point cloud data, determining, using the third data point, a second distance value representing a second minimum distance from the third position to one of the plurality of data points in the point cloud data, and determining the plurality of distance values, the plurality of distance values having the first number of data points and including the first distance value and the second distance value; and processing, using a first model, the plurality of distance values to generate output data. 6. The computer-implemented method of claim 5 , wherein the processing further comprises:processing, using the first model, the plurality of distance values to generate output data that indicates a first classification label of a plurality of classification labels, the first classification label corresponding to the object. 7. The computer-implemented method of claim 5 , wherein the processing further comprises:processing, using the first model, the plurality of distance values to generate output data that includes a 3D model representing the object. 8. The computer-implemented method of claim 5 , wherein the processing further comprises:processing, using the first model, the plurality of distance values to generate output data that includes a 3D model representing the object in a first pose that is different from a second pose represented by the point cloud data. 9. The computer-implemented method of claim 5 , whereindetermining the plurality of distance values further comprises:determining a first distance value between the second position and the first position; determining a third distance value between the second position and a fourth position associated with a fourth data point included in the point cloud data; determining that the first distance value is smaller than the third distance value; determining that the first data point is closest to the second data point of a plurality of data points included in the point cloud data; and associating the first distance value with the second data point. 10. The computer-implemented method of claim 5 , further comprising:generating, using the plurality of distance values, a first feature vector, a first element of the first feature vector corresponding to the first distance value; receiving second point cloud data representing a second object, the second point cloud data including a fourth data point at a fourth position in the 3D coordinate system; determining, using the second point cloud data and the basis point data, a second plurality of distance values, the second plurality of distance values including a third distance value corresponding to a second distance between the fourth position and the second position; generating, using the second plurality of distance values, a second feature vector, a first element of the second feature vector corresponding to the third distance value; and training the first model using the first feature vector and the second feature vector. 11. The computer-implemented method of claim 5 , further comprising:generating, using the plurality of distance values, a feature vector, a first value in the feature vector including first coordinates representing the first position associated with the first data point, wherein processing the plurality of distance values further comprises processing, using the first model, the feature vector to generate the output data. 12. The computer-implemented method of claim 5 , wherein the point cloud data includes a variable number of unstructured data points that lack organization, and the basis point data includes a fixed number of structured data points having a known spatial arrangement sampled from a random uniform ball. 13. A system comprising:at least one processor; and memory including instructions operable to be executed by the at least one processor to cause the system to:receive point cloud data representing an object, the point cloud data including a first data point at a first position in a three-dimensional (3D) coordinate system; determine basis point data with which to compare the point cloud data, the basis point data including a second data point at a second position in the 3D coordinate system; determine, using the point cloud data and the basis point data, a plurality of distance values, the plurality of distance values including a first distance value corresponding to a first distance from the second position to the first position; and process, using a first model, the plurality of distance values to generate output data, wherein the point cloud data includes a variable number of unstructured data points that lack organization, and the basis point data includes a fixed number of structured data points having a known spatial arrangement. 14. The system of claim 13 , wherein the memory further comprises instructions that, when executed by the at least one processor, further cause the system to:process, using the first model, the plurality of distance values to generate output data that indicates a first classification label of a plurality of classification labels, the first classification label corresponding to the object. 15. The system of claim 13 , wherein the memory further comprises instructions that, when executed by the at least one processor, further cause the system to:process, using the first model, the plurality of distance values to generate output data that includes a 3D model representing the object. 16. The system of claim 13 , wherein the memory further comprises instructions that, when executed by the at least one processor, further cause the system to:process, using the first model, the plurality of distance values to generate output data that includes a 3D model representing the object in a first pose that is different from a second pose represented by the point cloud data. 17. The system of claim 13 , wherein the memory further comprises instructions that, when executed by the at least one processor, further cause the system to:determine a first distance value between the second position and the first position; determine a second distance value between the second position and a third position associated with a third data point included in the point cloud data; determine that the first distance value is smaller than the second distance value; determine that the first data point is closest to the second data point of a plurality of data points included in the point cloud data; and associate the first distance value with the second data point. 18. The system of claim 13 , wherein the memory further comprises instructions that, when executed by the at least one processor, further cause the system to:generate, using the plurality of distance values, a first feature vector, a first element of the first feature vector corresponding to the first distance value; receive second point cloud data representing a second object, the second point cloud data including a third data point at a third position in the 3D coordinate system; determine, using the second point cloud data and the basis point data, a second plurality of distance values, the second plurality of distance values including a second distance value corresponding to a second distance between the third position and the second position; generate, using the second plurality of distance values, a second feature vector, a first element of the second feature vector corresponding to the second distance value; and train the first model using the first feature vector and the second feature vector. 19. The system of claim 13 , wherein the memory further comprises instructions that, when executed by the at least one processor, further cause the system to:generate, using the plurality of distance values, a feature vector, a first value in the feature vector including first coordinates representing the first position associated with the first data point, wherein processing the plurality of distance values further comprises processing, using the first model, the feature vector to generate the output data. 20. A computer-implemented method, the method comprising:receiving point cloud data representing an object, the point cloud data including a first data point at a first position in a three-dimensional (3D) coordinate system; determining basis point data with which to compare the point cloud data, the basis point data including a second data point at a second position in the 3D coordinate system; determining a first distance value corresponding to a first distance from the second position to the first position; determining a second distance value corresponding to a second distance from the second position to a third position associated with a third data point included in the point cloud data; determining that the first distance value is smaller than the second distance value; determining that the first data point is closest to the second data point of a plurality of data points included in the point cloud data; associating the first distance value with the second data point; determining, using the point cloud data and the basis point data, a plurality of distance values, the plurality of distance values including the first distance value; and processing, using a first model, the plurality of distance values to generate output data.
意見反饋