Granted Patent B2

US 12,412,390 · App. 17/600,597 · Granted Sep 9, 2025

Inventors: Yossef Israel Buda (Kfar-Adumim, IL); Tal Israel (Mazkeret-Batya, IL)

Assignee: CEPTION TECHNOLOGIES LTD.

G06V20/20G01C21/3841G01C21/3852G06V20/00

Abstract

Claims (145)

1. A computer-implemented method of determining at least one of a location and an orientation of a sensor, the computer-implemented method comprising:

including, in a primary database ( 131 ), representations of elements in a first space;

using an artificial intelligence engine that produces a machine learning model adapted to extract or predict, from input obtained from (i) a first perception sensor ( 310 ) or (ii) a first modality, one or more of location indicative elements of said elements that are characteristic or defining of a geographical location or position;

creating a mapping between (i) the representations of the elements including said one or more location indicative elements in the first space and (ii) the input from the first perception sensor or the first modality;

utilizing, in a second space, (i) the mapping and ( 11 ) input from a second perception sensor or a second modality that is different from the first perception sensor or the first modality, to identify one or more of the elements, including one or more of the one or more location indicative elements, whose representations are included in the database; and

using attributes of the identified one or more of the elements including said one or more location indicative elements to determine at least one of a location and an orientation of the second perception sensor or the second modality in the second space;

wherein the computer-implemented method comprises:

(a) automatically and autonomously creating said mapping of said space, based on input from a set of different sensors, by automatically creating in a primary database ( 131 ) the mapping of ( 1 ) said space and (ii) elements in said space, and

automatically creating element representations ( 132 ) based on input from said set of different sensors, and

automatically creating mapping between (I) element representations and (II) input from said set of sensors;

wherein objects or features in said space are usable for determining a location and an orientation of a sensor;

wherein after the mapping and element representations are created,

input from a sensor can be mapped to element representations, and the element representations can be used to determine the location and the orientation of the sensor;

wherein step (a) comprises:

storing in the primary database ( 131 ) a representation of an element in the first space;

creating a mapping between ( 1 ) the representation of the element, and (ii) input from the first perception sensor ( 310 ):

subsequently,

using (I) said mapping and (II) input from the second perception sensor, in the second space, to locate or identify one or more elements in the primary database ( 131 ), (b) providing a positioning solution based on matching between features or objects that are captured by perception sensors to features or objects that are represented in said primary database ( 131 ),

by matching between ( 1 ) an element as seen or captured from a first point-of-view, with ( 11 ) an element or feature as seen or captured from a second point-of-view;

(c) correlating between (i) representations of elements that are included in said primary database ( 131 ) based on a top-view thereof, and (ii) elements that were captured or seen using street-view images, and matching elements in a top-view with elements in a ground surface view;

(d) using auto-encoding features to learn ( 1 ) a first correlation between multiple Instances of a scene, and also (ii) a second correlation between multiple points of view, and also (iii) a third correlation between representations of a scene by a plurality of modalities;

(e) using Machine Learning,

creating a representation of an object based on input from a first type of sensor, and then identifying said object based on input from a second, different, type of sensor.

2. The computer-implemented method according to claim 1 , comprising:

based on an estimation of a location of a vehicle,

downloading some of the element representations ( 132 ) from said primary database ( 131 ) to a vehicular database ( 232 ) of a specific vehicle ( 230 ),

wherein the vehicular database ( 232 ) of the specific vehicle ( 230 ) only downloads from the primary database ( 131 ) and stores element representations ( 132 ) that describe elements that are physically located near the specific vehicle ( 230 ).

3. The computer-implemented method according to claim 1 , comprising:

at a localization Artificial Intelligence engine ( 220 ),

training a Machine Learning model ( 221 ),

by

(i) receiving, at the localization Artificial Intelligence engine ( 220 ), from a dataset ( 215 ), data that includes images;

wherein each data element ( 218 ) of said dataset ( 215 ) is associated with a geographical position and/or orientation;

(ii) receiving, at the localization Artificial Intelligence engine ( 220 ), data originating from a camera and data originating from a LIDAR sensor;

(iii) training, at the localization Artificial Intelligence engine ( 220 ), a Machine Learning model corresponding to data included in said dataset ( 215 ), and producing a Machine Learning model corresponding to a second type of data that originated from one or more sensors.

4. The computer-implemented method according to claim 3 , comprising:

(I) at the localization Artificial Intelligence engine ( 220 ),

producing a Machine Learning model that is adapted to receive an input data element of a specific type of sensor;

wherein the Machine Learning model is adapted to predict, from said received input data element, a location-indicative element ( 320 ) which is a characteristic of a geographical location;

wherein the location-indicative element is unintelligible or incoherent to a human observer, and does not pertain to any specific physical object, and is a produce of the Machine Learning model ( 221 ) that can be recognized by the localization Artificial Intelligence engine ( 220 ) as indicative of geographical locations;

(II) based on the Machine Learning model ( 221 ),

producing, from an incoming data element, at least one road code ( 250 ) that is associated with said location-indicative element.

5. The computer-implemented method according to claim 1 , comprising:

adapting a radar to sense an object, and to output a data element corresponding to the sensed object;

producing, by said radar, signals or data elements that correspond to a set of values or readings that represent a frequency or an amplitude or other attributes of reflected Radio Frequency radiation or electromagnetic waves;

mapping said set of values to an element representation object;

subsequently, when the set of values is received by another radar, mapping the set of values to said element representation object.

6. The computer-implemented method according to claim 1 ,

wherein said mapping comprises generating a signature based on a set of values received from a sensor;

wherein the method comprises:

utilizing said signature as a key for finding the correct element representation object in the vehicular database ( 232 ), by performing, at an Onboard Localization and Orientation Unit ( 233 ):

(i) examining input from a sensor,

(ii) producing a signature based on the input from the sensor,

(iii) using said signature to find an element representation object in the vehicular database ( 232 ) of the specific vehicle ( 230 );

(iv) once the element representation object is found, extracting location attributes of the element from the element representation object.

7. The computer-implemented method according to claim 1 , comprising:

creating a first mapping according to a first condition, and

creating a second mapping according to a second condition, and

selecting to use one of the first mapping and the second mapping according to a condition.

8. The computer-implemented method according to claim 7 ,

wherein the first condition is day time,

wherein the second condition is night time.

9. The computer-implemented method according to claim 7 ,

wherein the first condition is a rainy day,

wherein the second condition is a sunny day.

10. The computer-implemented method according to claim 1 , comprising:

classifying element representations according to a condition,

wherein elements that are best detected at night are associated with a first class,

wherein elements that are best detected in sunlight are associated with a second class.

11. The computer-implemented method according to claim 1 , comprising, comprising:

automatically selecting a set of elements to be used for determining location and/or orientation, based on a condition;

wherein,

if it rains when location of a sensor ( 311 ) is to be determined, then selecting a set of elements that are classified as best for poor visibility for determining location and/or orientation of said sensor ( 311 ).

12. The computer-implemented method according to claim 1 , comprising:

classifying element representations according to a suitability for sensor type, wherein a first class of elements is suitable for use with a camera, wherein a second class of elements is suitable for use with a LIDAR sensor.

13. The computer-implemented method according to claim 1 , comprising:

creating a plurality of mappings ( 330 ) for a respective plurality of sensor types; and

determining location and orientation of a sensor by selecting one of the plurality of mappings ( 330 ) based on the type of the sensor.

14. The computer-implemented method according to claim 1 , comprising:

(i) creating a plurality of mappings ( 330 ) for a respective plurality of sensor types;

(ii) dynamically selecting a mapping to be used, based on the type of sensor for which location and/or orientation are determined;

(iii) automatically selecting a subset of elements in the primary database ( 131 ), wherein the subset is selected based on the type of the sensor;

(iv) selecting mappings and element representations by selecting the set of element representations ( 132 ) and mappings ( 330 ) that are downloaded to the vehicular database ( 232 ) of said specific vehicle ( 230 ) and are the optimal set for the sensor used;

wherein the computer-implemented method comprises:

selecting a subset of element representations ( 132 ) and mappings ( 330 ) based on: (I) sensor type, and (II) suitability of the sensor to sense specific elements, and (III) a condition;

if input from the sensor cannot be mapped to any element representation ( 132 ), then:

creating in the primary database ( 131 ) a new element representation ( 132 ).

15. The computer-implemented method according to claim 1 , comprising:

encoding (i) data received from a plurality of sensor types, to (ii) a unified format representation to be included in said primary database ( 131 ); and

by encoding different data types coming from different sensor types into a unified format representation,

representing a single element in space based on input from a plurality of different modalities.

16. A non-transitory storage medium having stored thereon instructions that, when executed by a machine, cause the machine to perform a method comprising:

including, in a primary database ( 131 ), representations of elements in a first space;

using an artificial intelligence engine that produces a machine learning model adapted to extract or predict, from input obtained from (i) a first perception sensor ( 310 ) or (ii) a first modality, one or more of location indicative elements of said elements that are characteristic or defining of a geographical location or position;

creating a mapping between (i) the representations of the elements including said one or more location indicative elements in the first space and (ii) the input from the first perception sensor or the first modality;

utilizing, in a second space, (i) the mapping and (ii) input from a second perception sensor or a second modality that is different from the first perception sensor or the first modality, to identify one or more of the elements, including one or more of the one or more location indicative elements, whose representations are included in the database; and

using attributes of the identified one or more of the elements including said one or more location indicative elements to determine at least one of a location and an orientation of the second perception sensor or the second modality in the second space;

wherein the method comprises:

(a) automatically and autonomously creating said mapping of said space, based on input from a set of different sensors,

by automatically creating in a primary database ( 131 ) the mapping of (i) said space and (ii) elements in said space, and

automatically creating element representations ( 132 ) based on input from said set of different sensors, and

automatically creating mapping between (I) element representations and (II) input from said set of sensors;

wherein objects or features in said space are usable for determining a location and an orientation of a sensor;

wherein after the mapping and element representations are created,

input from a sensor can be mapped to element representations, and the element representations can be used to determine the location and the orientation of the sensor;

wherein step (a) comprises:

storing in the primary database ( 131 ) a representation of an element in the first space;

creating a mapping between (i) the representation of the element, and (ii) input from the first perception sensor ( 310 );

subsequently,

using (I) said mapping and (II) input from the second perception sensor, in a the second space, to locate or identify one or more elements in the primary database ( 131 ) (b) providing a positioning solution based on matching between features or objects that are captured by perception sensors to features or objects that are represented in said primary database ( 131 ),

by matching between (i) an element as seen or captured from a first point-of-view, with (ii) an element or feature as seen or captured from a second point-of-view;

(c) correlating between (i) representations of elements that are included in said primary database ( 131 ) based on a top-view thereof, and (ii) elements that were captured or seen using street-view images, and matching elements in a top-view with elements in a ground surface view;

(d) using auto-encoding features to learn (i) a first correlation between multiple instances of a scene, and also (ii) a second correlation between multiple points of view, and also (iii) a third correlation between representations of a scene by a plurality of modalities;

(e) using Machine Learning,

creating a representation of an object based on input from a first type of sensor,

and then identifying said object based on input from a second, different, type of sensor.

17. A system comprising:

one or more hardware processors that are configured to execute code,

and that are operably associated with a memory unit that is configured to store code and data,

wherein the one or more hardware processors are configured to perform a method comprising:

including, in a primary database ( 131 ), representations of elements in a first space;

using an artificial intelligence engine that produces a machine learning model adapted to extract or predict, from input obtained from (i) a first perception sensor ( 310 ) or (ii) a first modality, one or more of location indicative elements of said elements that are characteristic or defining of a geographical location or position;

creating a mapping between (i) the representations of the elements including said one or more location indicative elements in the first space and (ii) the input from the first perception sensor or the first modality;

utilizing, in a second space, (i) the mapping and (ii) input from a second perception sensor or a second modality that is different from the first perception sensor or the first modality, to identify one or more of the elements, including one or more of the one or more location indicative elements, whose representations are included in the database; and

using attributes of the identified one or more of the elements including said one or more location indicative elements to determine at least one of a location and an orientation of the second perception sensor or the second modality in the second space;

wherein the method comprises:

(a) automatically and autonomously creating said mapping of said space, based on input from a set of different sensors, by

automatically creating in a primary database ( 131 ) the mapping of (i) said space and (ii) elements in said space, and

automatically creating element representations ( 132 ) based on input from said set of different sensors, and

automatically creating mapping between (I) element representations and (II) input from said set of sensors;

wherein objects or features in said space are usable for determining a location and an orientation of a sensor;

wherein after the mapping and element representations are created,

input from a sensor can be mapped to element representations, and the element representations can be used to determine the location and the orientation of the sensor;

wherein step (a) comprises:

storing in the primary database ( 131 ) a representation of an element in the first space;

creating a mapping between (i) the representation of the element, and (ii) input from the first perception sensor ( 310 );

subsequently,

using (I) said mapping and (II) input from the second perception sensor, in the second space, to locate or identify one or more elements in the primary database ( 131 ),

(b) providing a positioning solution based on matching between features or objects that are captured by perception sensors to features or objects that are represented in said primary database ( 131 ),

by matching between (i) an element as seen or captured from a first point-of-view, with (ii) an element or feature as seen or captured from a second point-of-view;

(c) correlating between (i) representations of elements that are included in said primary database ( 131 ) based on a top-view thereof, and (ii) elements that were captured or seen using street-view images, and matching elements in a top-view with elements in a ground surface view;

(d) using auto-encoding features to learn (i) a first correlation between multiple instances of a scene, and also (ii) a second correlation between multiple points of view, and also (iii) a third correlation between representations of a scene by a plurality of modalities;

(e) using Machine Learning,

creating a representation of an object based on input from a first type of sensor,

and then identifying said object based on input from a second, different, type of sensor.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (111)