Method and apparatus for detecting a trailer, tow-ball, and coupler for trailer hitch assistance and jackknife prevention

1. A method for tracking a trailer, comprising:

obtaining a sequence of images from a camera positioned at a rear end of a vehicle,

wherein a field of view of the camera comprises at least a tow-ball attached to the vehicle and a trailer located in a scene behind the vehicle;

wherein the sequence of images is captured continuously at a sequence of predefined time stamps while the vehicle is moving; and

wherein a first image is captured at a first time stamp and a second image is captured at a second time stamp;

detecting the trailer and the tow-ball in the first image of the sequence of images using a trained learning model based on the first image and at least one scaled version of the first image, the detection comprising:

scaling the first image using a first scaling factor and a second scaling factor to generate a first scaled image and a second scaled image, respectively;

detecting the trailer in the first image using the trained learning model and a first detector to generate a first detected location;

detecting the trailer in the first scaled image using the trained learning model and a second detector to generate a second detected location;

detecting the trailer in the second scaled image using the trained learning model and a third detector to generate a third detected location; and

combining the first, the second and the third detected locations to generate a bounding box corresponding to the location of the trailer in the first image;

estimating an area of interest using a tracking algorithm and locations of the detected trailer and the tow-ball in the first image, wherein the area of interest corresponds to an area in the second image that will include the trailer with a probability that is higher than a first threshold;

estimating a distance between the vehicle and the trailer based on the detected location of the trailer in the first image; and

upon determining that the distance between the vehicle and the trailer is less than a second threshold:

continuously detect locations of the trailer, tow-ball, and a coupler attached to the trailer in each image of the sequence of images using the learning model and the estimated area of interest;

update the estimated area of interest to correspond to an area of interest in a following image in the sequence of images; and

actuating an action of the vehicle based on the detected locations of the trailer, tow-ball, and the coupler.

2. The method of claim 1 , wherein the area of interest comprises an area that is larger than an estimated bounding box for the trailer.

3. The method of claim 1 , wherein estimating the distance between the vehicle and the trailer based on the detected location of the trailer in the first image comprises: calculating the distance between the vehicle and the trailer using the detected location and one or more calibration values of the camera.

4. The method of claim 3 , wherein the one or more calibration values are determined in advance based on a plurality of calibrating images captured by at least a second camera located on a side of the vehicle.

5. The method of claim 1 , further comprising:

generating a graphical representation of one or more of the detected trailer, coupler, and the tow-ball, and a trajectory overlay for connecting the tow-ball to the coupler,

wherein the action of the vehicle comprises displaying the generated graphical representation to a driver of the vehicle to assist in moving the vehicle in reverse direction towards the trailer.

6. The method of claim 1 , further comprising: calculating an angle between the trailer and the vehicle while the vehicle is moving by continuously detecting and tracking movement of the trailer; and wherein actuating an action of the vehicle comprises actuating a jackknife prevention action to avoid a trailer jackknife situation.

7. The method of claim 1 , wherein the action of the vehicle comprises: determining a trajectory from the vehicle to the trailer using detected locations of the tow-ball and the coupler; and autonomously moving the vehicle in reverse direction towards the trailer by following the trajectory from the tow-ball to the coupler.

8. An apparatus for tracking a trailer, comprising a memory and at least one processor coupled to the memory, the at least one processor configured to:

obtain a sequence of images from a camera positioned at a rear end of a vehicle,

wherein a field of view of the camera comprises at least a tow-ball attached to the vehicle and a trailer located in a scene behind the vehicle;

wherein the sequence of images is captured continuously at a sequence of predefined time stamps while the vehicle is moving; and

wherein a first image is captured at a first time stamp and a second image is captured at a second time stamp;

detect the trailer and the tow-ball in the first image of the sequence of images using a trained learning model based on the first image and at least one scaled version of the first image, the detection comprising:

scaling the first image using a first scaling factor and a second scaling factor to generate a first scaled image and a second scaled image, respectively;

detecting the trailer in the first image using the trained learning model and a first detector to generate a first detected location;

detecting the trailer in the first scaled image using the trained learning model and a second detector to generate a second detected location;

detecting the trailer in the second scaled image using the trained learning model and a third detector to generate a third detected location; and

combining the first, the second and the third detected locations to generate a bounding box corresponding to the location of the trailer in the first image;

estimate an area of interest using a tracking algorithm and locations of the detected trailer and the tow-ball in the first image, wherein the area of interest corresponds to an area in the second image that will include the trailer with a probability that is higher than a first threshold;

estimate a distance between the vehicle and the trailer based on the detected location of the trailer in the first image; and

upon determining that the distance between the vehicle and the trailer is less than a second threshold:

continuously detect locations of the trailer, tow-ball and a coupler attached to the trailer in each image of the sequence of images using the learning model and the estimated area of interest;

update the estimated area of interest to correspond to an area of interest in a following image in the sequence of images, and

actuate an action of the vehicle based on the detected locations of the trailer, tow-ball and the coupler.

9. The apparatus of claim 8 , wherein the first area of interest comprises an area that is larger than an estimated bounding box for the trailer.

10. The apparatus of claim 8 , wherein the at least one processor configured to estimate the distance between the vehicle and the trailer based on the detected location of the trailer in the first image is further configured to calculate the distance between the vehicle and the trailer using the detected location and one or more calibration values of the camera.

11. The apparatus of claim 10 , wherein the one or more calibration values are determined in advance based on a plurality of calibrating images captured by at least a second camera located on a side of the vehicle.

12. The apparatus of claim 8 , wherein the at least one processor is further configured to generate a graphical representation of one or more of the detected trailer, coupler, and the tow-ball, and a trajectory overlay for connecting the tow-ball to the coupler, wherein the action of the vehicle comprises displaying the generated graphical representation to a driver of the vehicle to assist in moving the vehicle in reverse direction towards the trailer.

13. The apparatus of claim 8 , wherein the at least one processor is further configured to:

calculate an angle between the trailer and the vehicle while the vehicle is moving by continuously detecting and tracking movement of the trailer; and

actuate a jackknife prevention action to avoid a trailer jackknife situation.

14. The apparatus of claim 8 , wherein the at least one processor is further configured to:

determine a trajectory from the vehicle to the trailer using detected locations of the tow-ball and the coupler;

autonomously move the vehicle in reverse direction towards the trailer by following the trajectory from the tow-ball to the coupler.

15. A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors of a vehicle computer system, cause the one or more processors to perform the following:

obtain a sequence of images from a camera positioned in rear of a vehicle,

wherein a field of view of the camera comprises at least a tow-ball attached to the vehicle and a trailer located in a scene behind the vehicle;

wherein the sequence of images is captured continuously at a sequence of predefined time stamps while the vehicle is moving; and

wherein a first image is captured at a first time stamp and a second image is captured at a second time stamp;

detect the trailer and a tow-ball in the first image of the sequence of images using a trained learning model based on the first image and at least one scaled version of the first image, the detection comprising:

scaling the first image using a first scaling factor and a second scaling factor to generate a first scaled image and a second scaled image, respectively;

detecting the trailer in the first image using the trained learning model and a first detector to generate a first detected location;

detecting the trailer in the first scaled image using the trained learning model and a second detector to generate a second detected location;

detecting the trailer in the second scaled image using the trained learning model and a third detector to generate a third detected location; and

combining the first, the second and the third detected locations to generate a bounding box corresponding to the location of the trailer in the first image;

estimate an area of interest using a tracking algorithm and locations of the detected trailer and the tow-ball in the first image, wherein the area of interest corresponds to an area in the second image that will include the trailer with a probability that is higher than a first threshold;

estimate a distance between the vehicle and the trailer based on the detected location of the trailer in the first image;

upon determining that the distance between the vehicle and the trailer is less than a second threshold:

continuously detect locations of the trailer, tow-ball and a coupler attached to the trailer in each image of the sequence of images using the learning model and the estimated area of interest;

update the estimated area of interest to correspond to an area of interest in a following image in the sequence of images; and

actuate an action of the vehicle based on the detected locations of the trailer, tow-ball and the coupler.

16. The non-transitory computer-readable storage medium of claim 15 , wherein the first area of interest comprises an area that is larger than an estimated bounding box for the trailer.

17. The non-transitory computer-readable storage medium of claim 15 , wherein estimating the distance between the vehicle and the trailer based on the detected location of the trailer in the first image comprises calculate the distance between the vehicle and the trailer using the detected location and one or more calibration values of the camera.