Map data alignment method and device, medium and computer equipment

Abstract

The invention discloses a map data alignment method and device, a medium and computer equipment. The method comprises the following steps: acquiring incremental map data; detecting the incremental map data to obtain information of vector points on a plurality of map element instances in the incremental map data; encoding the information of the vector points to obtain encoding information of the vector points, and aggregating the encoding information of the vector points on the same map element instance to obtain a descriptor of the map element instance; other map element instances located in the same local area with the map element instance are determined, and the descriptors of the map element instance are updated based on the descriptors of the other map element instances; and based on the similarity between the updated descriptors of the plurality of map element instances and the descriptors of the map element instances in the base map, aligning the incremental map data with the base map.

Claims

1. A method for aligning map data, characterized in that: The method comprises: Get incremental map data; Detecting the incremental map data to obtain information of vector points on several map element instances in the incremental map data; Encoding the information of the vector points to obtain the encoded information of the vector points, and aggregating the encoded information of the vector points on the same map element instance to obtain a descriptor of the map element instance; Determining other map element instances in the same local area as the map element instance, and updating the descriptor of the map element instance based on the descriptors of the other map element instances; The incremental map data is aligned with the base map based on similarities between the updated descriptors of the plurality of map element instances and the descriptors of the map element instances in the base map.

2. The method according to claim 1, characterized in that The expected information of the vector point includes information of multiple dimensions; the method further includes: The information of the vector points obtained by the detection is input into a feature completion network, so that when the information of at least one dimension of the multiple dimensions is missing in the information of the vector points, the missing information is completed by the feature completion network.

3. The method according to claim 2, characterized in that The feature completion network is trained based on the following method: Acquire sample information, where the sample information includes information of the multiple dimensions; Information on at least one dimension of the sample information is masked; Predicting the masked information in the sample information through the original feature completion network; The original feature completion network is trained based on the prediction result to obtain the feature completion network.

4. The method according to claim 2, characterized in that: The information in the multiple dimensions includes: position information, semantic information, shape information, feature information and perceptual uncertainty of the vector point.

5. The method according to claim 1, characterized in that The determining of other map element instances in the same local area as the map element instance includes: Generate a graph network based on the descriptors of the map feature instances in the incremental map data, wherein the graph network includes a plurality of nodes, the nodes are connected by edges, and each node corresponds to a descriptor; The updating of the descriptor of the map element instance based on the descriptor of the other map element instance includes: For each target node in the graph network, a descriptor corresponding to the target node is updated based on descriptors corresponding to other nodes connected to the target node.

6. The method according to claim 5, characterized in that The updating of the descriptor corresponding to the target node based on the descriptors corresponding to other nodes connected to the target node includes: Concatenate the descriptor corresponding to the target node and the descriptor corresponding to the other nodes to obtain a concatenated feature, and decode the concatenated feature through a multi-layer perceptron to obtain the value of the edge between the target node and the other nodes; Determining the weight of the edge connecting the target node and the other nodes based on the similarity between the descriptor corresponding to the target node and the descriptor corresponding to the other nodes and the adjacency matrix; the adjacency matrix is ​​used to indicate whether there is a connection relationship between the target node and the other nodes; Weighting the value of the edge based on the weight of the edge to obtain a weighted value of the edge; The descriptor of the target node is updated based on the weighted value of the edge.

7. The method according to claim 1, characterized in that The aligning the incremental map data with the base map based on the similarity between the updated descriptors of the plurality of map element instances and the descriptors of the map element instances in the base map comprises: Determine a similarity matrix, each element in the similarity matrix is ​​used to represent the similarity between the updated descriptor of a map element instance in the incremental map data and the descriptor of a map element instance in the base map; A mapping relationship between the plurality of map element instances and each map element instance in the base map is determined based on the similarity matrix.

8. The method according to claim 7, characterized in that The determining, based on the similarity matrix, the mapping relationship between the plurality of map element instances and each map element instance in the base map comprises: Determine from the base map a map element instance with the greatest similarity to each map element instance in the incremental map data, and determine from the incremental map data a map element instance with the greatest similarity to each map element instance in the base map; If the map element instance determined to have the highest similarity to the j-th map element instance in the incremental map data is the i-th map element instance in the base map, and the map element instance determined to have the highest similarity to the i-th map element instance in the base map is the j-th map element instance in the incremental map data, the j-th map element instance in the incremental map data and the i-th map element instance in the base map are determined to be the same element instance, and a mapping relationship between the j-th map element instance in the incremental map data and the i-th map element instance in the base map is established.

9. The method according to claim 7, characterized in that: The determining of the similarity matrix comprises: Determine an initial similarity matrix; Adding a plurality of learnable parameters to the initial similarity matrix; the plurality of learnable parameters are used to represent the probability that the incremental map element is not associated with the map element instance in the base map; The similarity matrix with the learnable parameters added is optimized to obtain the similarity matrix.

10. A computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images