OBJECT TRACKING BY UNSUPERVISED LEARNING

• US 20180203446A1
• Filed: 01/18/2017
• Published: 07/19/2018
• Est. Priority Date: 01/18/2017
• Status: Active Grant

• Current Assignee: Ford Global Technologies, LLC (Ford Motor Company)
• Original Assignee: Ford Global Technologies, LLC (Ford Motor Company)
• Inventors: Wyffels, Kevin
• Granted Patent: US 10,449,956 B2
• Time in Patent Office: Days
• CPC Class Codes: B60W 2420/408 Radar; Laser, e.g. lidar
• CPC Class Codes: B60W 30/09 Taking automatic action to avoid collision, e.g. braking and steering
• CPC Class Codes: G01S 17/42 Simultaneous measurement of distance and other co-ordinates indirect measurement G01S17/46
• CPC Class Codes: G01S 17/66 Tracking systems using electromagnetic waves other than radio waves
• CPC Class Codes: G01S 17/931 of land vehicles
• CPC Class Codes: G01S 7/4808 Evaluating distance, position or velocity data
• CPC Class Codes: G05D 1/024 in combination with a laser lasers per se H01S

Abstract

A computing device in a vehicle can determine one or more objects based on 3D data points by determining a joint Bayesian probability of each of the one or more objects, conditioned on previously determined objects, and pilot the vehicle based on the determined one or more objects, wherein the objects have parameters including locations, sizes, poses, speeds, directions and predicted paths.

Claims

1. A method, comprising:

determining one or more current objects based on 3D data points by determining joint Bayesian probabilities for each of the one or more current objects, conditioned on previously determined objects; and

piloting a vehicle based on the determined one or more current objects.

2. The method of claim 1, wherein the current objects have parameters including locations, sizes, poses, speeds, directions and predicted paths.

3. The method of claim 1, wherein the previously determined objects have parameters including locations, sizes, poses, speeds, directions, predicted paths, and boundaries.

4. The method of claim 1, wherein the 3D data points are determined by LIDAR.

5. The method of claim 1, wherein the current objects are determined by recursively fitting object model parameters in a joint object model using Bayesian probabilities to model object parameters associated with each of the current objects, wherein each object parameters model is conditioned based on respective previously determined objects.

6. The method of claim 5, wherein the recursive model fitting is performed by determining partial derivatives with respect to each of the model object parameters in the joint object model.

7. The method of claim 6, wherein the joint object model is determined to cause the recursive fit to converge in a predetermined number of steps.

8. The method of claim 7, wherein the joint object model includes an expected number of objects, based on the previously determined objects.

9. The method of claim 8, wherein the joint object model includes one or more new objects.

10. The method of claim 9, wherein the one or more new objects are located at occlusion boundaries associated with previously determined objects or at predetermined limits of LIDAR sensor range.

11. An apparatus, comprising:

a processor;

a memory, the memory storing instructions executable by the processor to;

determine one or more current objects based on 3D data points by determining joint Bayesian probabilities for each of the one or more current objects, conditioned on previously determined objects; and

pilot a vehicle based on the determined one or more current objects.

12. The apparatus of claim 11, wherein the current objects have parameters including locations, sizes, poses, speeds, directions and predicted paths.

13. The apparatus of claim 11, wherein the previously determined object states have parameters including locations, sizes, poses, speeds, directions, predicted paths, and boundaries.

14. The apparatus of claim 11, wherein the 3D data points are determined by LIDAR.

15. The apparatus of claim 11, wherein the current objects are determined by recursively fitting object model parameters in a joint object model using Bayesian probabilities associated with each of the object model parameters, each respectively conditioned with object model parameters based on the respective previously determined objects.

16. The apparatus of claim 15, wherein the recursive model fitting is performed by determining partial derivatives with respect to each of the object model parameters in the joint object model.

17. The apparatus of claim 16, wherein the joint object model is arranged to cause the recursive fit to converge in a predetermined number of steps.

18. The apparatus of claim 17, wherein the joint object model includes an expected number of objects, based on the previously determined objects.

19. The apparatus of claim 18, wherein the joint object model includes one or more new objects.

20. The apparatus of claim 19, wherein the new object is located at an occlusion boundary associated with a previously determined object or at predetermined limits of LIDAR sensor range.