System and method for improving the representation of line features

Patent metadata

Field	Value
Publication number	US-10740645-B2
Application number	US-201816023645-A
Country	US
Kind code	B2
Filing date	Jun 29, 2018
Priority date	Jun 29, 2018
Publication date	Aug 11, 2020
Grant date	Aug 11, 2020

Abstract

System, methods, and other embodiments described herein relate to improving an electronic representation of lines. In one embodiment, a method includes, in response to acquiring sensor data from at least one sensor representing a surrounding environment of a robotic device, extracting a feature representation of an observed line feature in the sensor data by providing a probability distribution that is defined based, at least in part, on feature parameters that overparameterize the observed line feature. The method includes converting the feature representation of the observed line feature into reduced parameters to avoid the feature parameters overparameterizing the observed line feature. The reduced parameters include an observation uncertainty for the line feature that is based, at least part, on the probability distribution. The method includes providing a detection distribution according to a correlation between the reduced parameters and a mapped line feature.

First claim

What is claimed is:

1. A translation system for improving electronic representation of lines, comprising:
one or more processors;
a memory communicably coupled to the one or more processors and storing:
a sensor module including instructions that when executed by the one or more processors cause the one or more processors to, in response to acquiring sensor data from at least one sensor representing a surrounding environment of a robotic device, extract a feature representation of an observed line feature in the sensor data by providing a probability distribution that is defined based, at least in part, on feature parameters that overparameterize the observed line feature; and
a transform module including instructions that when executed by the one or more processors cause the one or more processors to convert the feature representation of the observed line feature into reduced parameters to avoid the feature parameters overparameterizing the observed line feature, wherein the reduced parameters include an observation uncertainty for the observed line feature that is based, at least part, on the probability distribution, and
wherein the transform module includes instructions to provide a detection distribution according to a correlation between the reduced parameters and a mapped line feature.

2. The translation system of claim 1 , wherein the transform module includes instructions to convert the feature representation including instructions to apply an unscented transform to estimate a non-linear transform between the feature parameters and the reduced parameters and provide the feature representation according to the reduced parameters including a vector with four values representing the observed line feature and a covariance matrix that is four-by-four in size and full-rank representing the observation uncertainty.

3. The translation system of claim 2 , wherein the transform module includes instructions to apply the unscented transform including instructions to:
i) compute an error between samples of the probability distribution by comparing inverse representation and standard representations of the observed line feature from the probability distribution, wherein computing the error includes determining differences between angular and positional values, wherein the samples are represented using the reduced parameters, and
ii) compute an average for the observed line feature over the samples, and
wherein applying the unscented transform produces the vector and the covariance matrix that is full-rank as an electronic output.

4. The translation system of claim 2 , wherein the transform module includes instructions to provide the covariance matrix as an electronic output that quantifies innovation between the observed line feature and the mapped line feature; and
wherein the transform module includes instructions to provide the detection distribution including instructions to estimate a location of the robotic device based, at least in part, on the covariance matrix.

5. The translation system of claim 1 , wherein the sensor module includes instructions to identify the mapped line feature from a feature map, the mapped line feature being represented according to the reduced parameters, wherein the transform module includes instructions to provide the detection distribution to indicate at least an estimation uncertainty for a correlation between the mapped line feature and the observed line feature that is a function of the observation uncertainty.

6. The translation system of claim 1 , wherein the sensor module includes instructions to acquire the sensor data including noise along with the observed line feature that at least partially distorts the observed line feature within the sensor data, and
wherein the reduced parameters include position and orientation parameters defining an azimuth and an elevation for the observed line feature.

7. The translation system of claim 1 , wherein the sensor data includes at least one of: a three-dimensional representation of the surrounding environment as a point cloud, and a camera image, and
wherein the feature parameters include at least six values defining endpoints of the observed line feature in Euclidean three-space.

8. The translation system of claim 1 , wherein the probability distribution is quantified in a six-by-six covariance matrix, and
wherein the reduced parameters include at least a full-rank covariance matrix that quantifies innovation of the observed line feature.

9. A non-transitory computer-readable medium for improving an electronic representation of lines and including instructions that when executed by one or more processors cause the one or more processors to:
in response to acquiring sensor data from at least one sensor representing a surrounding environment of a robotic device, extract a feature representation of an observed line feature in the sensor data by providing a probability distribution that is defined based, at least in part, on feature parameters that overparameterize the observed line feature;
convert the feature representation of the observed line feature into reduced parameters to avoid the feature parameters overparameterizing the observed line feature, wherein the reduced parameters include an observation uncertainty for the observed line feature that is based, at least part, on the probability distribution; and
provide a detection distribution according to a correlation between the reduced parameters and a mapped line feature.

10. The non-transitory computer-readable medium of claim 9 , wherein the instructions to convert the feature representation include instructions to apply an unscented transform to estimate a non-linear transform between the feature parameters and the reduced parameters and provide the feature representation according to the reduced parameters including a vector with four values representing the observed line feature and a full-rank covariance matrix that is four-by-four in size representing the observation uncertainty.

11. The non-transitory computer-readable medium of claim 10 , wherein the instructions to apply the unscented transform include instructions to:
i) compute an error between samples of the probability distribution by comparing inverse representation and standard representations of the observed line feature from the probability distribution, wherein computing the error includes determining differences between angular and positional values, wherein the samples are represented using the reduced parameters, and ii) compute an average for the observed line feature over the samples, and
wherein applying the unscented transform produces the vector and the full-rank covariance matrix as an electronic output.

12. The non-transitory computer-readable medium of claim 10 , wherein the instructions to convert the observed line feature include instructions to provide the full-rank covariance matrix as an electronic output that quantifies innovation between the observed line feature and the mapped line feature; and
wherein the instructions to provide the detection distribution include instructions to estimate a location of the robotic device based, at least in part, on the full-rank covariance matrix.

13. The non-transitory computer-readable medium of claim 9 , wherein the instructions include instructions to acquire the sensor data including noise along with the observed line feature that at least partially distorts the observed line feature within the sensor data, and
wherein the reduced parameters include position and orientation parameters defining an azimuth and an elevation for the observed line feature.

14. A method

Assignees

• Toyota Res Inst Inc

Inventors

• Ozog Paul J

Classifications

• G06V20/588 Primary

  Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road · CPC title

• G06F18/251

  of input or preprocessed data · CPC title

• G06V10/44

  Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components · CPC title

• G06T7/74

  involving reference images or patches · CPC title

• G06T2207/30256

  Lane; Road marking · CPC title