Center Path Estimator for Agricultural Windrows
AssigneeSouthwest Research Institute
InventorsEdmund M DuPont, Prasanna R. Kolar
Life of this patent
Estimating the term…
Abstract
A windrow path detector for use by baling equipment. A camera system is mounted on the baling equipment and provides image data representing a stereo view of a windrow ahead of the baling equipment. A pre-processor to generates a depth image. A neural network receives the depth image and the image data from at least one camera, and is trained to use that input to generate a predicted center line over the windrow. A center path process uses the predicted center line to define a navigation path, which it delivers to a navigation system on-board the baling equipment.
Claims10
A windrow path detector for use by baling equipment, the baling equipment being equipped with a navigation system, comprising:
- a camera system mounted on the baling equipment and having at least two spaced cameras, the cameras operable to provide image data representing a stereo view of a windrow ahead of the baling equipment;
- an image data pre-processor operable to receive the image data from each camera and to generate a depth image;
- a neural network operable to receive input data comprising the depth image and the image data from at least one camera, the neural network being trained to use the input data to generate a predicted center line over the windrow;
- and
- a center path process operable to use the predicted center line to define a navigation path, using a spline fit and converting pixels into three-dimensional points relative to the cameras and projected onto the ground surface;
- the center path process further operable to deliver the navigation path to the navigation system.
A method of training a neural network to perform windrow path detector for use by baling equipment, comprising:
- driving a vehicle having a camera system over a number of test windrows, the camera system having at least two spaced cameras operable to provide image data representing a stereo view of each windrow when ahead of the vehicle;
- combining the image data with depth data to provide a perspective image of the windrow
- using a navigation system of the vehicle to create localization data representing a driven path over the windrows;
- overlaying the perspective image data with the localization data, thereby generating labeled image data;
- inputting the labeled image data to the neural network;
- and
- performing a training process on the neural network to estimate the center of the windrow.
Dependencies are inferred from the claim text.