Object detection system

• US 10,527,730 B2
• Filed: 10/21/2016
• Issued: 01/07/2020
• Est. Priority Date: 10/22/2015
• Status: Active Grant

• Current Assignee: Toyota Jidosha Kabushiki Kaisha (Toyota Motor Corporation)
• Original Assignee: Toyota Motor Engineering & Manufacturing North America Incorporated (Toyota Motor Corporation)
• Inventors: Uehara, Yasuo
• Primary Examiner(s): Tran, Khoi H
• Assistant Examiner(s): Hannan, B M M
• Application Number:
  US15/331,025
• Publication Number:
  US 20170115396A1
• Time in Patent Office: 1,173 Days
• Field of Search: 701 25, 701 323, 701 324, 701300, 701416
• CPC Class Codes: G01S 17/86 Combinations of lidar systems with systems other than lidar, radar or sonar, e.g. with direction finders
• CPC Class Codes: G01S 17/89 for mapping or imaging
• CPC Class Codes: G01S 17/931 of land vehicles
• CPC Class Codes: G01S 7/4802 using analysis of echo signal for target characterisation; Target signature; Target cross-section
• CPC Class Codes: G01S 7/4808 Evaluating distance, position or velocity data

Abstract

In a method involving the recognition of ambiguous objects from samples of 3D points, an initial constant is set for properties of prospective points of an object, and a sample of 3D points is received from a 3D sensor. When the sensed locations in space of points of objects represented by the 3D points are the same as those for the prospective points of an object, they are redefined as an actual points of objects, and the initial constants are replaced with corresponding properties represented by the 3D points. An ambiguous object is then recognized from a sparsely populated cluster of actual points of an object identified based on the replacement of the initial constants. The ambiguous object is confirmed based on signals representing its surface properties received from a polarization image sensor.

Claims

1. A method, comprising:

setting, using a detection module executable by at least one processor, an initial constant for at least one property of each of a plurality of spatially predefined prospective points of an object having respective locations in space and prospectively representable by 3D points from a 3D sensor;

receiving, using the detection module executable by the at least one processor, from the 3D sensor, a sample of 3D points each representing a sensed location in space of a point of an object, and at least one sensed property of the point of an object;

using the detection module executable by the at least one processor, for each 3D point in the sample, when the sensed location in space of the point of an object represented by the 3D point is the same as the location in space of a prospective point of an object, redefining the prospective point of an object as an actual point of an object, and replacing the initial constant set for the at least one property of the actual point of an object with the at least one sensed property of the point of an object represented by the 3D point;

clustering, using the detection module executable by the at least one processor, actual points of an object with associated un-redefined prospective points of an object based on the replacement of the initial constant set for the at least one property of the actual points of an object in the cluster with the at least one sensed property of the points of objects represented by the 3D points;

recognizing, using the detection module executable by the at least one processor, an ambiguous object from a sparse population of the actual points of an object in the cluster relative to the associated un-redefined prospective points of an object in the cluster; and

confirming, using the detection module executable by the at least one processor, the ambiguous object based on signals representing its polarization signature received from a polarization image sensor.

2. The method of claim 1, wherein the at least one property of each of the plurality of prospective points of an object, and the at least one sensed property of the point of an object, are each at least one of an amplitude, an intensity and a reflectivity.

3. The method of claim 2, wherein the initial constant set for at the at least one property of each of the plurality of prospective points of an object is a maximum or minimum value of the at least one of an amplitude, an intensity and a reflectivity.

4. The method of claim 1, further comprising:

driving, using a planning module executable by the at least one processor, a vehicle along a route based on the recognized and confirmed ambiguous object.

5. The method of claim 1, further comprising:

identifying, using the detection module executable by the at least one processor, a closest actual point of an object in the cluster; and

assigning, using the detection module executable by the at least one processor, the remaining actual points of an object in the cluster new simulated locations in space that correspond in closeness to the location in space of the identified closest actual point of an object in the cluster.

6. The method of claim 5, further comprising:

driving, using a planning module executable by the at least one processor, a vehicle along a route based on the new simulated locations in space of the actual points of an object in the cluster.

7. The method of claim 1, wherein the 3D sensor is a LIDAR sensor.

8. A method, comprising:

setting, using a detection module executable by at least one processor, an initial constant for at least one property of each of a plurality of spatially predefined prospective points of an object having respective locations in space and prospectively representable by 3D points from a 3D sensor;

receiving, using the detection module executable by the at least one processor, from the 3D sensor, a sample of 3D points each representing a sensed location in space of a point of an object, and at least one sensed property of the point of an object;

using the detection module executable by the at least one processor, for each 3D point in the sample, when the sensed location in space of the point of an object represented by the 3D point is the same as the location in space of a prospective point of an object, redefining the prospective point of an object as an actual point of an object, and replacing the initial constant set for the at least one property of the actual point of an object with the at least one sensed property of the point of an object represented by the 3D point;

clustering, using the detection module executable by the at least one processor, actual points of an object with associated un-redefined prospective points of an object based on the replacement of the initial constant set for the at least one property of the actual points of an object in the cluster with the at least one sensed property of the points of objects represented by the 3D points;

recognizing, using the detection module executable by the at least one processor, an ambiguous object from a sparse population of the actual points of an object in the cluster relative to the associated un-redefined prospective points of an object in the cluster;

identifying, using the detection module executable by the at least one processor, a closest actual point of an object in the cluster; and

assigning, using the detection module executable by the at least one processor, the remaining actual points of an object in the cluster new simulated locations in space that correspond in closeness to the location in space of the identified closest actual point of an object in the cluster.

9. The method of claim 8, wherein the at least one property of each of the plurality of prospective points of an object, and the at least one sensed property of the point of an object, are each at least one of an amplitude, an intensity and a reflectivity.

10. The method of claim 9, wherein the initial constant set for at the at least one property of each of the plurality of prospective points of an object is a maximum or minimum value of the at least one of an amplitude, an intensity and a reflectivity.

11. The method of claim 8, further comprising:

driving, using a planning module executable by the at least one processor, a vehicle along a route based on the new simulated locations in space of the actual points of an object in the cluster.

12. The method of claim 8, further comprising:

confirming, using the detection module executable by the at least one processor, the ambiguous object based on signals representing its polarization signature received from a polarization image sensor.

13. The method of claim 8, wherein the 3D sensor is a LIDAR sensor.

14. A method, comprising:

setting, using a detection module executable by at least one processor, an initial constant for at least one property of each of a plurality of spatially predefined prospective points of an object having respective locations in space and prospectively representable by 3D points from a 3D sensor;

receiving, using the detection module executable by the at least one processor, from the 3D sensor, a sample of 3D points taken according to a sampling rate for a set of samples of 3D points taken over a time period in which the poses of objects are projected to stay substantially the same, the 3D points each representing a sensed location in space of a point of an object, and at least one sensed property of the point of an object;

using the detection module executable by the at least one processor, for each 3D point in the sample, when the sensed location in space of the point of an object represented by the 3D point is the same as the location in space of a prospective point of an object, redefining the prospective point of an object as an actual point of an object, and replacing the initial constant set for the at least one property of the actual point of an object with the at least one sensed property of the point of an object represented by the 3D point;

clustering, using the detection module executable by the at least one processor, actual points of an object with associated un-redefined prospective points of an object based on the replacement of the initial constant set for the at least one property of the actual points of an object in the cluster with the at least one sensed property of the points of objects represented by the 3D points;

recognizing, using the detection module executable by the at least one processor, an ambiguous object from a sparse population of the actual points of an object in the cluster relative to the associated un-redefined prospective points of an object in the cluster; and

increasing, using the detection module executable by the at least one processor, the sampling rate for the set of samples of 3D points based on the recognition of the ambiguous object.

15. The method of claim 14, wherein the at least one property of each of the plurality of prospective points of an object, and the at least one sensed property of the point of an object, are each at least one of an amplitude, an intensity and a reflectivity.

16. The method of claim 15, wherein the initial constant set for at the at least one property of each of the plurality of prospective points of an object is a maximum or minimum value of the at least one of an amplitude, an intensity and a reflectivity.

17. The method of claim 14, further comprising:

repeating, using the detection module executable by the at least one processor, the setting, receiving, redefining, replacing and recognizing steps for the set of samples of 3D points; and

identifying, using the detection module executable by the at least one processor, the consistency of and reducing the noise among the actual points of an object in the clusters.

18. The method of claim 14, further comprising:

repeating, using the detection module executable by the at least one processor, the setting, receiving, redefining, replacing and recognizing steps to take a temporal series of sets of samples of 3D points taken over consecutive times over which the poses of objects are projected to change; and

tracking, using the detection module executable by the at least one processor, the ambiguous object based on its repeated recognition across the temporal series of sets of samples of 3D points.

19. The method of claim 18, further comprising:

driving, using a planning module executable by the at least one processor, a vehicle along a route based on the tracked ambiguous object.

20. The method of claim 14, wherein the 3D sensor is a LIDAR sensor.