CN-118424303-B - High-precision map data association method based on joint probability

CN118424303BCN 118424303 BCN118424303 BCN 118424303BCN-118424303-B

Abstract

The invention discloses a high-precision map data association method based on joint probability, which comprises the following steps of firstly, parameterizing a high-precision map landmark and a vehicle detection landmark; setting a correlation threshold according to the current position of the vehicle, calculating the correlation probability of the joint data, establishing a dynamic joint probability matrix, and correcting the time sequence. According to the method, the method for establishing the correct matching between the observation landmark and the map landmark by the intelligent vehicle is determined by introducing the joint probability between the semantic similarity, the local spatial similarity and the global structural similarity between the landmarks, so that the intelligent vehicle can be accurately and robustly positioned by utilizing a high-precision map, and the method has important significance for improving the reliability and the safety of an intelligent driving system.

Inventors

CHENG SHAOWU
Gu Zitian

Assignees

哈尔滨工业大学

Dates

Publication Date: 20260508
Application Date: 20240425

Claims (3)

1. The high-precision map data association method based on the joint probability is characterized by comprising the following steps of: Step one, parameterizing high-precision map landmarks and vehicle detection landmarks: Step one, according to the high-precision map, setting landmark semantic categories contained in the map; Step one, classifying landmarks detected by the intelligent vehicle according to map landmark semantics; step one, parameterizing the map landmark and the vehicle observation landmark, and assuming sharing A plurality of observation landmarks, The parameterization equations for the map landmarks are as follows: in the formula, A set of landmarks detected for the vehicle; Is a map landmark set; as a two-dimensional coordinate point of a landmark in a high-precision map, Semantic categories that are landmarks; step two, setting an association threshold according to the current position of the vehicle: step two, generating landmark point pairs to be matched within the forward 50m range of the current vehicle position; step two, generating an association threshold for each map point, wherein the association threshold is an ellipse with a long axis along the running direction of the vehicle : In the formula, And (3) with Is the first one in the running process of the intelligent vehicle Two-dimensional coordinates of the individual landmark points to be mapped, Is a threshold value; step two, if a certain detected landmark is not in the association threshold of the map landmark point, the matching probability of the detected landmark is directly set as a minimum value, and if the certain detected landmark is in the association threshold of the map landmark point, the step three is executed; step three, calculating association probability of the joint data: Definition of the definition Is the first in a certain frame The detection landmark point Probability of establishing a match between individual map landmark points, given a determined data association Landmark set detected by vehicle , Assume semantic similarity between landmarks to be matched Global structural similarity Similarity to local space Independently of each other, the joint association probability is calculated from the product of the above similarities: global structural similarity The calculation formula of (2) is as follows: in the formula, 、 Respectively, detecting landmark points Map landmark point Is defined by the geometric center of the (c), 、 Respectively, detecting landmark points Map landmark point Is used to determine the global structural metric value of (a), Is a super parameter for weighting the structural similarity of the current landmark point and other landmark points; step four, establishing a dynamic joint probability matrix: step four, a dynamic joint probability matrix is established for each data association , ; Step four, combining dynamic probability matrix Each column is normalized, and the data association of the current frame is determined according to the following maximum likelihood principle: in the formula, For the matching relationship of the landmark points to be determined, i.e. the first The detection landmark point A match is established between the individual map landmark points, Refers to a dynamic joint probability matrix Middle (f) All of the elements of the column, Refers to a dynamic joint probability matrix Middle (f) All elements of a row; step five, correcting the time sequence: recording dynamic joint probability matrix at current moment When (when) Is a certain association probability When the previous frame is also established, smoothing correction is performed according to the association probability calculated at the previous moment, and a specific correction formula is as follows: in the formula, Is the associated probability after the timing correction, In order to correct the length of time it takes, As a smoothing factor, the smoothing factor is used, Is in front of The associated probability corresponding to the moment; so far, the dynamic joint probability matrix at the current moment is corrected through time sequence smoothing, and then the final correct data association is selected according to the fourth step.
2. The joint probability-based high-precision map data association method according to claim 1, wherein in the third step, semantic similarity is generated due to the discreteness of landmarks The method is modeled as a probability quality function, when the semantic types of the detected landmark points and the map landmark points are the same, the semantic similarity is equal to the average correct rate of the intelligent vehicle detection system, and otherwise, the semantic similarity is equal to the average false detection rate.
3. The joint probability-based high-precision map data correlation method according to claim 1, wherein in the third step, local spatial similarity is obtained The coordinate difference value between the landmark points to be matched is calculated to obtain, and the calculation formula is as follows: in the formula, 、 Respectively, detecting landmark points Map landmark point Is used for the two-dimensional coordinates of (c), To detect covariance matrix of landmark coordinates.

Description

High-precision map data association method based on joint probability Technical Field The invention belongs to the technical field of intelligent vehicle positioning and information, relates to a data association method for high-precision map positioning, and in particular relates to a high-precision map data association method based on joint probability. Background Today, intelligent vehicles and autopilot technology are rapidly evolving, and high-precision map positioning is crucial for achieving safe and efficient autopilot. In a specific driving scenario, accurate and robust vehicle position output can be achieved using the position information of landmarks in the high-precision map when the sensor-based positioning results fail. How to accurately establish data association in real time (i.e. matching between vehicle observation landmarks and map landmarks) is a pain point problem based on high-precision map positioning. When the data association is established incorrectly, inaccurate positioning can be caused, so that the safety and reliability of the intelligent driving system are affected, and the risk of traffic accidents is increased. The traditional data association method is based on a nearest neighbor method by introducing landmark semantic features, and the problem of data association ambiguity is partially solved, but when a driving scene is complex, the nearest neighbor method is seriously influenced by landmark repetition and shielding, and the problem of data association ambiguity still exists. Disclosure of Invention In order to solve the problem that when a driving scene is complex, the traditional data association based on the nearest neighbor method is seriously influenced by landmark repetition and shielding, so that positioning is inaccurate, and further the safety and reliability of an intelligent driving system are influenced, the invention provides a high-precision map data association method based on joint probability. According to the method, the method for establishing the correct matching between the observation landmark and the map landmark by the intelligent vehicle is determined by introducing the joint probability between the semantic similarity, the local spatial similarity and the global structural similarity between the landmarks, so that the intelligent vehicle can be accurately and robustly positioned by utilizing a high-precision map, and the method has important significance for improving the reliability and the safety of an intelligent driving system. The invention aims at realizing the following technical scheme: A high-precision map data association method based on joint probability comprises the following steps: Step one, parameterizing high-precision map landmarks and vehicle detection landmarks: Step one, according to the high-precision map, setting landmark semantic categories contained in the map; Step one, classifying landmarks detected by the intelligent vehicle according to map landmark semantics; step one, parameterizing the map landmark and the vehicle observation landmark, and assuming sharing A plurality of observation landmarks,The parameterization equations for the map landmarks are as follows: in the formula, A set of landmarks detected for the vehicle; Is a map landmark set; as a two-dimensional coordinate point of a landmark in a high-precision map, Semantic categories that are landmarks; step two, setting an association threshold according to the current position of the vehicle: step two, generating landmark point pairs to be matched within the forward 50m range of the current vehicle position; step two, generating an association threshold for each map point, wherein the association threshold is an ellipse with a long axis along the running direction of the vehicle : In the formula,And (3) withIs the first one in the running process of the intelligent vehicleTwo-dimensional coordinates of the individual landmark points to be mapped,Is a threshold value; step two, if a certain detected landmark is not in the association threshold of the map landmark point, the matching probability of the detected landmark is directly set as a minimum value, and if the certain detected landmark is in the association threshold of the map landmark point, the step three is executed; step three, calculating association probability of the joint data: Definition of the definition Is the first in a certain frameThe detection landmark pointProbability of establishing a match between individual map landmark points, given a determined data associationLandmark set detected by vehicle,Assume semantic similarity between landmarks to be matchedGlobal structural similaritySimilarity to local spaceIndependently of each other, the joint association probability is calculated from the product of the above similarities: step four, establishing a dynamic joint probability matrix: step four, a dynamic joint probability matrix is established for each data association ,; Step four, combi