Determining multi-degree-of-freedom pose to navigate a vehicle
AssigneeFord Global Technologies, LLC
InventorsPunarjay Chakravarty, Chandana Neerukonda, Sagar Manglani
Life of this patent
Estimating the term…
Abstract
An imaging system and method of providing localization data to a vehicle using the imaging system is disclosed. The method may comprise: capturing, from an electronic depth camera, one or more images, wherein the one or more images include at least a portion of the vehicle; and using a deep neural network and the one or more images, determining a multi-degree of freedom (MDF) pose of the vehicle, wherein an optical axis of the electronic depth camera is oriented along a Z-axis according to a Cartesian coordinate system (comprising an X-axis, a Y-axis, and the Z-axis), wherein the Z-axis is plumb with respect to Earth.
Claims20
A method of providing localization data to a vehicle, comprising:
- capturing, from a first electronic depth camera, one or more images, wherein the one or more images include at least a portion of the vehicle;
- and
- using a deep neural network and the one or more images, determining a multi-degree of freedom (MDF) pose of the vehicle,
- wherein an optical axis of the first electronic depth camera is oriented along a Z-axis according to a Cartesian coordinate system (comprising an X-axis, a Y-axis, and the Z-axis), wherein the Z-axis is plumb with respect to Earth.
An imaging system for guiding a vehicle, comprising:
- a first red-green-blue-depth (RGBD) camera comprising a housing couplably fixed to a structure and an optical assembly carried by the housing, wherein an optical axis of the optical assembly is oriented normal to ground, wherein the first RGBD camera is located in a region that is absent a satellite-based positioning signal or the satellite positioning signal is attenuated less than a threshold, wherein the first RGBD camera is oriented along a vertical axis;
- a wireless transceiver;
- and
- a computer communicatively coupled to the first RGBD camera and the transceiver, wherein the computer comprises one or more processors and memory storing instructions executable by the one or more processors, wherein the instructions comprise to:
- (a) identify a vehicle to which to provide localization data;
- (b) capture, via the first RGBD camera, one or more image frames of the vehicle;
- (c) using a deep neural network and the one or more image frames of the RGBD camera, determine a multi-degree of freedom (MDF) pose of the vehicle;
- (d) using the MDF pose, determine a new waypoint for the vehicle;
- and
- (e) transmitting, via the transceiver, a command to the vehicle to navigate to the new waypoint.
A method for cross-modal supervision, comprising:
- capturing, from a first electronic depth camera, a depth image and a red-green-blue image, wherein the depth image and a red-green-blue image includes at least a portion of a vehicle, wherein the red-green-blue image includes a depiction of fiducial marker positioned on an exterior surface of the vehicle;
- calculating a multi-degree of freedom (MDF) pose of the vehicle based on the red-green-blue image;
- assigning a label to the depth image, wherein the label comprises the calculated MDF pose of the vehicle;
- and
- training a deep neural network with the labeled depth image.
Dependencies are inferred from the claim text.