Granted Patent B2

US 12,669,593 · App. 18/089,279 · Granted Jun 30, 2026

Inventors: Wei-Yuan Hsieh (Taipei City, TW); Ming-Xuan Wu (New Taipei City, TW); Chia-Jui Hu (New Taipei City, TW)

Assignee: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

G01S7/497G01S17/89

Abstract

Claims (69)

1 . A point cloud positioning error detection method, performed by a processing device, comprising:

obtaining a plurality of pieces of first point data and a target point cloud map, wherein the target point cloud map comprises a plurality of pieces of target point data, wherein the plurality of pieces of first point data are generated by a point cloud generator sensing a surrounding of a vehicle;

registering the pieces of first point data and the pieces of target point data to obtain a plurality of pieces of second point data with a plurality of iterations;

selecting a plurality of pieces of third point data from the pieces of second point data according to a first default distance;

calculating a plurality of matching scores of the pieces of third point data relative to the pieces of target point data;

obtaining a plurality of step vectors corresponding to the pieces of third point data, respectively, when said registering converges, and obtaining a plurality of effective values according to directions of the step vectors; and

outputting a localization fault detection result based on an intersection of the matching scores and the effective values,

wherein outputting the localization fault detection result based on the intersection of the matching scores and the effective values comprises:

obtaining a distribution of the matching scores relative to the effective values;

setting a first point at the distribution according to an average value of the matching scores;

setting a second point at the distribution according to a plurality of positive effective values among the effective values;

setting a third point at the distribution according to a plurality of negative effective values among the effective values;

setting an ineffective range with the first point, the second point and the third point;

calculating an ineffective ratio of a part of the pieces of third point data falling in the ineffective range relative to the pieces of third point data; and

outputting the localization fault detection result according to the ineffective ratio and an ineffective threshold;

wherein the localization fault detection result is outputted to an automobile computer, for the automobile computer to use the localization fault detection result for positioning.

2 . The point cloud positioning error detection method according to claim 1 , wherein calculating the matching scores of the pieces of third point data relative to the pieces of target point data comprises:

for each of the pieces of third point data as a target piece of third point data, performing:

determining one or more pieces of target point data among the pieces of target point data, wherein a distance between each of the one or more pieces of target point data and the target piece of third point data is not greater than a second default distance;

calculating one or more distances between the one or more pieces of target point data and the target piece of third point data; and

designating the matching score according to a comparison result of the distances and a third default distance,

wherein the matching score is inversely proportional to the distances.

3 . The point cloud positioning error detection method according to claim 1 , wherein selecting the pieces of third point data from the pieces of second point data according to the first default distance comprises:

for each of the pieces of second point data as a target piece of second point data, performing:

determining a piece of target point data among the pieces of target point data closest to the target piece of second point data;

determining whether a distance between the piece of target point data and the target piece of second point data is greater than the first default distance;

if the distance is greater than the first default distance, abandoning the target piece of second point data; and

if the distance is not greater than the first default distance, using the target piece of second point data as one of the pieces of third point data.

4 . The point cloud positioning error detection method according to claim 1 , wherein obtaining the effective values according to the directions of the step vectors comprises:

enclosing the pieces of third point data according to a default three-dimensional block to obtain a plurality of point cloud blocks;

determining a plurality of normal vectors of the point cloud blocks;

determining a component of each one of the step vectors on a corresponding one of the normal vectors;

if a direction of the component is in a positive direction, setting the effective value to be positive; and

if the direction of the component is in a negative direction, setting the effective value to be negative.

5 . A point cloud positioning error detection system, comprising:

a point cloud generator configured to generate a plurality of pieces of first point data by sensing a surrounding of a vehicle; and

a processing device connected to the point cloud generator, and configured to perform:

registering the pieces of first point data and a plurality of pieces of target point data of a target point cloud map to obtain a plurality of pieces of second point data with a plurality of iterations;

selecting a plurality of pieces of third point data from the pieces of second point data according to a first default distance;

calculating a plurality of matching scores of the pieces of third point data relative to the pieces of target point data;

obtaining a plurality of step vectors corresponding to the pieces of third point data, respectively, when said registering converges, and obtaining a plurality of effective values according to directions of the step vectors; and

outputting a localization fault detection result based on an intersection of the matching scores and the effective values,

wherein the processing device performing outputting the localization fault detection result based on the intersection of the matching scores and the effective values comprises:

obtaining a distribution of the matching scores relative to the effective values;

setting a first point at the distribution according to an average value of the matching scores;

setting a second point at the distribution according to a plurality of positive effective values among the effective values;

setting a third point at the distribution according to a plurality of negative effective values among the effective values;

setting an ineffective range with the first point, the second point and the third point;

calculating an ineffective ratio of a part of the pieces of third point data falling in the ineffective range relative to the pieces of third point data; and

outputting the localization fault detection result according to the ineffective ratio and an ineffective threshold;

wherein the localization fault detection result is outputted to an automobile computer, for the automobile computer to use the localization fault detection result for positioning.

6 . The point cloud positioning error detection system according to claim 5 , wherein the processing device performing calculating the matching scores of the pieces of third point data relative to the pieces of target point data comprises:

for each of the pieces of third point data as a target piece of third point data, performing:

determining one or more pieces of target point data among the pieces of target point data, wherein a distance between each of the one or more pieces of target point data and the target piece of third point data is not greater than a second default distance;

calculating one or more distances between the one or more pieces of target point data and the target piece of third point data; and

designating the matching score according to a comparison result of the distances and a third default distance,

wherein the matching score is inversely proportional to the distances.

7 . The point cloud positioning error detection system according to claim 5 , wherein the processing device performing selecting the pieces of third point data from the pieces of second point data according to the first default distance comprises:

for each of the pieces of second point data as a target piece of second point data, performing:

determining a piece of target point data among the pieces of target point data closest to the target piece of second point data;

determining whether a distance between the piece of target point data and the target piece of second point data is greater than the first default distance;

if the distance is greater than the first default distance, abandoning the target piece of second point data; and

if the distance is not greater than the first default distance, using the target piece of second point data as one of the pieces of third point data.

8 . The point cloud positioning error detection system according to claim 5 , wherein the processing device performing obtaining the effective values according to the directions of the step vectors comprises:

enclosing the pieces of third point data according to a default three-dimensional block to obtain a plurality of point cloud blocks;

determining a plurality of normal vectors of the point cloud blocks;

determining a component of each one of the step vectors on a corresponding one of the normal vectors;

if a direction of the component is in a positive direction, setting the effective value to be positive; and

if the direction of the component is in a negative direction, setting the effective value to be negative.

Assignments (1)

Priority Claims (1)

Continuity (2)

References Cited (24)