Autonomous machine having vision system for navigation and method of using same
Vision systems for autonomous machines and methods of using same during machine localization are provided. Exemplary systems and methods may reduce computing resources needed to perform vision-based localization by selecting the most appropriate camera from two or more cameras, and optionally selecting only a portion of the selected camera's field of view, from which to perform vision-based location correction. Other embodiments may provide camera lens coverings that maintain optical clarity while operating within debris-filled environments.
1 . A method for autonomous machine navigation comprising:
placing a ground maintenance machine within a work region defined by one or more boundaries, the machine comprising:
two or more vision sensors, each vision sensor adapted to capture image data within a field of view (FOV) defined by the vision sensor, wherein each vision sensor defines a different FOV relative to the machine; and
a controller in communication with each of the vision sensors;
designating at least one of the vision sensors as a localization vision sensor; and
determining with the controller a vision-based pose of the machine, the pose representing one or both of a position and an orientation of the machine relative to the work region, wherein determining the pose of the machine is based upon matching features in the image data received from the localization vision sensor to features associated with a previously-identified three-dimensional point cloud (3DPC) used to define the work region,
wherein designating at least one of the vision sensors as the localization vision sensor comprises either:
(a) analyzing the image data from each of the vision sensors;
identifying a first vision sensor from among the two or more vision sensors, wherein the FOV of the first vision sensor contains a greater number of feature matches with features of the 3DPC than the FOV of any of the other vision sensors; and
selecting the first vision sensor as the localization vision sensor; or
(b) selecting which of the vision sensors has the FOV that is predicted by the controller to contain either: a greater number of features of the 3DPC; or a feature or cluster of features having a recognition score greater than a recognition score of the features or cluster of features predicted to be within the FOV of any of the other vision sensors.
2 . The method of claim 1 , wherein designating the localization vision sensor comprises selecting which of the vision sensors has the FOV that is predicted by the controller in accordance with element (b) above and further comprises:
dividing the FOV of each vision sensor into a first zone and a second zone;
predicting, based upon the pose of the machine as estimated by the controller, either the number of features, or the recognition score of features, of the 3DPC contained within the first zone of the FOV of each of the vision sensors; and
designating the localization vision sensor to be the vision sensor having the first zone predicted by the controller to contain either: the greatest number of features of the 3DPC; or the greatest recognition score of features of the 3DPC.
3 . The method of claim 1 , wherein matching features in the image data comprises:
dividing the FOV of one or more of the vision sensors into a first zone and a second zone; and
matching features in the image data to features associated with the 3DPC using only the image data from the first zone.
4 . The method of claim 3 , wherein dividing the FOV comprises dividing the FOV such that the first zone is located vertically above the second zone.
5 . The method of claim 1 , further comprising:
collecting visual odometry data from one or more of the vision sensors; and
correcting the pose of the machine based upon the visual odometry data.
6 . An autonomous ground maintenance machine comprising:
a housing supporting a ground maintenance implement;
drive wheels supporting the housing in rolling engagement with a ground surface of a work region;
a propulsion system coupled to the drive wheels and adapted to control rotational speed and direction of the drive wheels;
a vision system comprising at least two cameras, wherein each camera captures image data within a field of view (FOV) different than the FOV of the other camera(s), and wherein one of the cameras is designated as a localization camera and one of the cameras is designated as a visual odometry camera; and
a controller operatively coupled to the vision system, wherein the controller is adapted to determine a vision-based pose of the machine, the pose representing one or both of a position and an orientation of the machine relative to the work region, and wherein the determination of the pose is based upon matching features in the image data from the localization camera to features associated with a previously-identified three-dimensional point cloud (3DPC) used to define the work region,
wherein the visual odometry camera is adapted to provide visual odometry data to the controller.
7 . The machine of claim 6 , wherein the visual odometry camera has a FOV directed transverse to a direction of travel of the machine.
8 . The machine of claim 6 , wherein the localization camera and the visual odometry camera are the same camera.
9 . The machine of claim 6 , wherein the localization camera and the visual odometry camera are different cameras.
10 . The machine of claim 6 , wherein the controller is adapted to:
determine whether the drive wheels have slipped relative to the ground surface based upon the visual odometry data; and
update the vision-based pose of the machine in response to determining that the drive wheels have slipped.
11 . The machine of claim 6 , wherein the FOV of each camera is divided into a first zone and a second zone.
12 . The machine of claim 11 , wherein the image data of the first zone of the FOV of each camera is provided to the controller for matching features with the 3DPC, and the image data of the second zone of the FOV of each camera is adapted to provide the visual odometry data to the controller.
13 . The machine of claim 6 , wherein the FOV of each camera extends 30-60 degrees above a horizontal plane.
14 . The machine of claim 6 , wherein one or more of the cameras comprises a lens covering, the lens covering comprising one or more of hydrophilic lens covering, a hydrophobic lens covering, an anti-reflective lens covering, an anti-glare lens covering, a polarizing filter, and one or more removable layers.