System and method for target behavior prediction in advanced driving assist system (ADAS), autonomous driving (AD), or other applications
A method includes obtaining sensor data associated with a target object at a host vehicle and identifying relative positions of the target object with reference to the host vehicle over time. The method also includes generating polynomials associated with the target object over time based on the relative positions, where coefficients of each polynomial are based on a magnitude of a covariance of the relative positions used to generate the polynomial. The method further includes generating a weighted combination of the polynomials for the target object, where the weighted combination is a representation of an estimated behavior of the target object. In addition, the method includes determining whether a collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials and, in response to determining that the collision is possible, initiating one or more corrective actions by the host vehicle.
1. A method comprising:
obtaining sensor data associated with a target object at a host vehicle;
identifying relative positions of the target object with reference to the host vehicle over time;
generating polynomials associated with the target object over time based on the relative positions of the target object, wherein coefficients of each polynomial are based on a magnitude of a covariance of the relative positions of the target object used to generate the polynomial;
comparing, for at least one polynomial of the polynomials being generated, the magnitude of the covariance to a covariance activation threshold in order to determine whether the coefficients of the at least one polynomial are determined or set to zero;
generating a weighted combination of the polynomials for the target object, wherein the weighted combination is a representation of an estimated behavior of the target object;
determining whether a collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials; and
in response to determining that the collision between the host vehicle and the target object is possible, initiating one or more corrective actions by the host vehicle.
2. The method of claim 1 , further comprising, for each polynomial being generated:
generating a covariance matrix based on the relative positions of the target object used to generate the polynomial;
generating the magnitude of the covariance of the relative positions of the target object used to generate the polynomial based on the covariance matrix; and
comparing the magnitude of the covariance to the covariance activation threshold in order to determine whether the coefficients of the polynomial are determined or set to zero.
3. The method of claim 2 , wherein:
the coefficients of the polynomial are set to zero when the magnitude of the covariance is below the covariance activation threshold; and
the coefficients of the polynomial are determined using the relative positions of the target object when the magnitude of the covariance is above the covariance activation threshold.
4. The method of claim 1 , wherein generating the weighted combination of the polynomials for the target object comprises using weights defined by a sigmoid function.
5. The method of claim 1 , wherein determining whether the collision between the host vehicle and the target object is possible comprises:
determining whether one coefficient of the weighted combination of the polynomials related to a lateral displacement between the host vehicle and the target object indicates that collision is possible; and
determining whether another coefficient of the weighted combination of the polynomials related to a curvature of an estimated path of the target object is towards a travel direction of the host vehicle and indicates that collision is possible.
6. The method of claim 1 , wherein the one or more corrective actions comprise at least one of:
adjusting at least one of: a steering of the host vehicle, a speed of the host vehicle, and a braking of the host vehicle; and
activating an audible, visible, or haptic warning.
7. The method of claim 1 , further comprising repeatedly:
generating the polynomials for the target object and the weighted combination of the polynomials for the target object over time; and
determining whether the collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials for the target object.
8. The method of claim 1 , further comprising repeatedly:
generating multiple polynomials for multiple target objects and weighted combinations of the multiple polynomials for the multiple target objects over time; and
determining whether a collision between the host vehicle and any of the target objects is possible based on the weighted combination of the polynomials for the multiple target objects.
9. An apparatus comprising:
at least one processing device configured to:
obtain sensor data associated with a target object at a host vehicle;
identify relative positions of the target object with reference to the host vehicle over time;
generate polynomials associated with the target object over time based on the relative positions of the target object, wherein coefficients of each polynomial are based on a magnitude of a covariance of the relative positions of the target object used to generate the polynomial;
compare, for at least one polynomial of the polynomials being generated, the magnitude of the covariance to a covariance activation threshold in order to determine whether the coefficients of the at least one polynomial are determined or set to zero;
generate a weighted combination of the polynomials for the target object, wherein the weighted combination is a representation of an estimated behavior of the target object;
determine whether a collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials; and
in response to determining that the collision between the host vehicle and the target object is possible, initiate one or more corrective actions by the host vehicle.
10. The apparatus of claim 9 , wherein the at least one processing device is further configured, for each polynomial being generated, to:
generate a covariance matrix based on the relative positions of the target object used to generate the polynomial;
generate the magnitude of the covariance of the relative positions of the target object used to generate the polynomial based on the covariance matrix; and
compare the magnitude of the covariance to the covariance activation threshold in order to determine whether the coefficients of the polynomial are determined or set to zero.
11. The apparatus of claim 10 , wherein:
the at least one processing device is configured to set the coefficients of the polynomial to zero when the magnitude of the covariance is below the covariance activation threshold; and
the at least one processing device is configured to determine the coefficients of the polynomial using the relative positions of the target object when the magnitude of the covariance is above the covariance activation threshold.
12. The apparatus of claim 9 , wherein, to generate the weighted combination of the polynomials for the target object, the at least one processing device is configured to use weights defined by a sigmoid function.
13. The apparatus of claim 9 , wherein, to determine whether the collision between the host vehicle and the target object is possible, the at least one processing device is configured to:
determine whether one coefficient of the weighted combination of the polynomials related to a lateral displacement between the host vehicle and the target object indicates that collision is possible; and
determine whether another coefficient of the weighted combination of the polynomials related to a curvature of an estimated path of the target object is towards a travel direction of the host vehicle and indicates that collision is possible.
14. The apparatus of claim 9 , wherein the one or more corrective actions comprise at least one of:
adjusting at least one of: a steering of the host vehicle, a speed of the host vehicle, and a braking of the host vehicle; and
activating an audible, visible, or haptic warning.
15. The apparatus of claim 9 , wherein the at least one processing device is further configured to repeatedly:
generate the polynomials for the target object and the weighted combination of the polynomials for the target object over time; and
determine whether the collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials for the target object.
16. The apparatus of claim 9 , wherein the at least one processing device is further configured to repeatedly:
generate multiple polynomials for multiple target objects and weighted combinations of the multiple polynomials for the multiple target objects over time; and
determine whether a collision between the host vehicle and any of the target objects is possible based on the weighted combination of the polynomials for the multiple target objects.
17. A non-transitory machine-readable medium containing instructions that when executed cause at least one processing device of a host vehicle to:
obtain sensor data associated with a target object;
identify relative positions of the target object with reference to the host vehicle over time;
generate polynomials associated with the target object over time based on the relative positions of the target object, wherein coefficients of each polynomial are based on a magnitude of a covariance of the relative positions of the target object used to generate the polynomial;
compare, for at least one polynomial of the polynomials being generated, the magnitude of the covariance to a covariance activation threshold in order to determine whether the coefficients of the at least one polynomial are determined or set to zero;
generate a weighted combination of the polynomials for the target object, wherein the weighted combination is a representation of an estimated behavior of the target object;
determine whether a collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials; and
in response to determining that the collision between the host vehicle and the target object is possible, initiate one or more corrective actions by the host vehicle.
18. The non-transitory machine-readable medium of claim 17 , further containing instructions that when executed cause the at least one processing device, for each polynomial being generated, to:
generate a covariance matrix based on the relative positions of the target object used to generate the polynomial;
generate the magnitude of the covariance of the relative positions of the target object used to generate the polynomial based on the covariance matrix; and
compare the magnitude of the covariance to the covariance activation threshold in order to determine whether the coefficients of the polynomial are determined or set to zero.
19. The non-transitory machine-readable medium of claim 18 , wherein:
the instructions when executed cause the at least one processing device to set the coefficients of the polynomial to zero when the magnitude of the covariance is below the covariance activation threshold; and
the instructions when executed cause the at least one processing device to determine the coefficients of the polynomial using the relative positions of the target object when the magnitude of the covariance is above the covariance activation threshold.
20. The non-transitory machine-readable medium of claim 17 , wherein the instructions that when executed cause the at least one processing device to generate the weighted combination of the polynomials for the target object comprise:
instructions that when executed cause the at least one processing device to use weights defined by a sigmoid function.
21. The non-transitory machine-readable medium of claim 17 , wherein the instructions that when executed cause the at least one processing device to determine whether the collision between the host vehicle and the target object is possible comprise:
instructions that when executed cause the at least one processing device to:
determine whether one coefficient of the weighted combination of the polynomials related to a lateral displacement between the host vehicle and the target object indicates that collision is possible; and
determine whether another coefficient of the weighted combination of the polynomials related to a curvature of an estimated path of the target object is towards a travel direction of the host vehicle and indicates that collision is possible.
22. The non-transitory machine-readable medium of claim 17 , wherein the one or more corrective actions comprise at least one of:
adjusting at least one of: a steering of the host vehicle, a speed of the host vehicle, and a braking of the host vehicle; and
activating an audible, visible, or haptic warning.
23. The non-transitory machine-readable medium of claim 17 , further containing instructions that when executed cause the at least one processing device to repeatedly:
generate the polynomials for the target object and the weighted combination of the polynomials for the target object over time; and
determine whether the collision between the host vehicle and the target object is possible based on the weighted combination of the polynomials for the target object.
24. The non-transitory machine-readable medium of claim 17 , further containing instructions that when executed cause the at least one processing device to repeatedly:
generate multiple polynomials for multiple target objects and weighted combinations of the multiple polynomials for the multiple target objects over time; and
determine whether a collision between the host vehicle and any of the target objects is possible based on the weighted combination of the polynomials for the multiple target objects.