Identifying and avoiding obstructions using depth information in a single image
A farming machine includes one or more image sensors for capturing an image as the farming machine moves through the field. A control system accesses an image captured by the one or more sensors and identifies a distance value associated with each pixel of the image. The distance value corresponds to a distance between a point and an object that the pixel represents. The control system classifies pixels in the image as crop, plant, ground, etc. based on depth information in in the pixels. The control system generates a labelled point cloud using the labels and depth information, and identifies features about the crops, plants, ground, etc. in the point cloud. The control system generates treatment actions based on any of the depth information, visual information, point cloud, and feature values. The control system actuates a treatment mechanism based on the classified pixels.
1. A method for avoiding an obstruction by a machine that moves through an operational environment, the method comprising:
accessing a single image of the operational environment from an image sensor as the machine moves through the operational environment, the single image comprising one or more pixels representing the obstruction;
applying a depth identification model to the single image, the depth identification model including a convolutional neural network configured to identify distances between sensors and representative pixels in single images by:
encoding the single image onto a first layer of the convolutional neural network as an encoded image;
transforming the encoded image to a reduced image on a second layer of the convolutional neural network;
classifying latent features in the reduced image as a distance between the image sensor and an object represented by the pixel;
decoding the reduced image to a decoded image corresponding to the single image, each pixel in the decoded image representing the distance between the object represented by the pixel and the image sensor in the single image; and
classifying, based on the distance represented by each pixel, a first set of pixels in the decoded image as the obstruction; and
modifying an operating parameter of the machine, the modified operating parameter changing a speed of the machine to avoid the classified obstruction, and the modified operating parameter selected based on the distance for pixels in the first set of pixels representing the obstruction.
2. The method of claim 1 , wherein changing the speed of the machine to avoid the classified obstruction changes the speed to zero.
3. The method of claim 1 , further comprising modifying a sensor parameter of the image sensor based on the distance represented by each of the pixels in the decoded image.
4. The method of claim 1 , wherein modifying the operating parameter to change the speed of the machine to avoid the classified obstruction is an automated process of the machine.
5. The method of claim 1 , wherein the single image further comprises one or more pixels representing a plant and applying the depth identification model further comprises:
classifying latent features in the reduced image as a distance between the image sensor and the plant represented by the pixel;
decoding the reduced image to the decoded image corresponding to the single image, one or more pixels in the decoded image representing the distance between the plant represented by the pixel and the image sensor in the single image; and
classifying, based on the distance represented by each pixel, a second set of pixels in the decoded image as the plant;
wherein the modified operating parameter is additionally selected based on the distance for pixels in the second set of pixels representing the plant.
6. The method of claim 5 , wherein the distance of the obstruction is closer to the machine than the distance of the plant.
7. The method of claim 5 , wherein the distance of the obstruction is farther from the machine than the distance of the plant.
8. The method of claim 1 , further comprising generating a depth map comprising the distances represented by pixels in the decoded image and wherein the depth map comprises the distances of the obstruction.
9. The method of claim 1 , further comprising:
generating a point cloud representing the obstruction based on the distances for pixels in the first set of pixels of the decoded image representing the obstruction.
10. A machine configured to perform machine actions and avoid an obstruction as the machine moves through an operational environment, the machine comprising:
an image sensor configured to capture images of the operational environment as the machine moves through the operational environment;
one or more processors; and
a non-transitory computer readable storage medium comprising computer program instructions that, when executed by the one or more processors, cause the machine to:
access a single image of the operational environment from the image sensor, the single image comprising one or more pixels representing the obstruction;
apply a depth identification model to the image, the depth identification model including a convolutional neural network configured to identify distances between sensors and representative pixels in single images by:
encoding the single image onto a first layer of the convolutional neural network as an encoded image;
transforming the encoded image to a reduced image on a second layer of the convolutional neural network;
classifying latent features in the reduced image as a distance between the image sensor and an object represented by the pixel;
decoding the reduced image to a decoded image corresponding to the single image, each pixel in the decoded image representing the distance between the object represented by the pixel and the image sensor in the single image; and
classifying, based on the distance represented by each pixel, a first set of pixels in the decoded image as the obstruction; and
modifying an operating parameter of the machine, the modified operating parameter changing a speed of the machine to avoid the classified obstruction, and the modified operating parameter selected based on the distance for pixels in the first set of pixels representing the obstruction.
11. The machine of claim 10 , wherein changing the speed of the machine to avoid the classified obstruction changes the speed to zero.
12. The machine of claim 10 , wherein the computer program instructions, when executed, cause the one or more processors to modify a sensor parameter of the image sensor based on the distance represented by each of the pixels in the decoded image.
13. The machine of claim 10 , wherein executing the computer program instructions to modify the operating parameter to change the speed of the machine to avoid the classified obstruction is an automated process of the machine.
14. The machine of claim 10 , wherein the single image further comprises one or more pixels representing a plant and applying the depth identification model to the single image further comprises:
classifying latent features in the reduced image as a distance between the image sensor and the plant represented by the pixel;
decoding the reduced image to the decoded image corresponding to the single image, one or more pixels in the decoded image representing the distance between the plant represented by the pixel and the image sensor in the single image; and
classifying, based on the distance represented by each pixel, a second set of pixels in the decoded image as the plant;
wherein the modified operating parameter is additionally selected based on the distance for pixels in the second set of pixels representing the plant.
15. The machine of claim 14 , wherein the distance of the obstruction is closer to the machine than the distance of the plant.
16. The machine of claim 14 , wherein the distance of the obstruction is farther from the machine than the distance of the plant.
17. The machine of claim 10 , wherein the computer program instructions, when executed, further cause the one or more processors to generate a depth map comprising the distances represented by pixels in the decoded image and wherein the depth map comprises the distances of the obstruction.
18. The machine of claim 10 , wherein the computer program instructions, when executed, further cause the one or more processors to generate a point cloud representing the obstruction based on the distances for pixels in the first set of pixels of the decoded image representing the obstruction.
19. A non-transitory computer readable storage medium comprising computer program instructions for performing actions to avoid an obstruction as a machine moves through an operational environment, the computer program instructions, when executed by one or more processors, causing the one or more processors to:
accessing a single image of the operational environment from an image sensor as the machine moves through the operational environment, the single image comprising one or more pixels representing the obstruction;
applying a depth identification model to the image, the depth identification model including a convolutional neural network configured to identify distances between sensors and representative pixels in single images by:
encoding the single image onto a first layer of the convolutional neural network as an encoded image;
transforming the encoded image to a reduced image on a second layer of the convolutional neural network;
classifying latent features in the reduced image as a distance between the image sensor and an object represented by the pixel;
decoding the reduced image to a decoded image corresponding to the single image, each pixel in the decoded image representing the distance between the object represented by the pixel and the image sensor in the single image; and
classifying, based on the distance represented by each pixel, a first set of pixels in the decoded image as the obstruction; and
modifying an operating parameter of the machine, the modified operating parameter changing a speed of the machine to avoid the classified obstruction, and the modified operating parameter selected based on the distance for pixels in the first set of pixels representing the obstruction.