Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.
1.1Stored ground survey frames
What ifWhat if the ground view comes from a stored survey captured before the current drive?
Claim language“being sensed in real-time”
A local archive stores timestamped ground photographs with camera poses and roadway coordinates. During the drive, the processor retrieves frames that overlap the vehicle's projected field of view and uses them to locate fixed features such as lane markings and curbs.
1.2Ground surveyed roadway vectors
What ifWhat if ground survey vectors supply the roadway geometry used to draw the overhead view?
Claim language“air based data”
A map tile stores lane boundaries, curb lines, and crosswalk polygons measured by a ground survey vehicle. The processor clips the vectors to the local driving region and renders them in a vehicle-centered overhead projection.
1.3Fixed geometric feature extraction
What ifWhat if fixed image filters and geometric fitting generate the road feature vectors?
Claim language“a machine learning process”
Calibrated projection maps ground image pixels onto the road plane, and fixed gradient filters extract edges from the projected ground image and the aerial image. A line-fitting routine encodes each lane marking candidate as a vector of endpoints, heading, and length for spatial matching.
1.4Removal of stale markings
What ifWhat if the processor removes stale lane markings from the overhead view when the ground view shows bare pavement?
Claim language“adding road elements”
The processor aligns the live ground view with an overhead marking layer and checks whether each projected marking lies on visible pavement. When five consecutive frames show bare pavement across the marking footprint, the processor clears the corresponding pixels from the marking layer and renders the remaining geometry.
1.4Direct obstacle footprint overlay
What ifWhat if the processor draws road obstacles from measured footprint coordinates?
Claim language“represented by at least one of the embeddings”
A scanning range sensor supplies three-dimensional returns, and a geometric routine removes ground points and groups the remaining returns into obstacle clusters. The processor fits a footprint polygon to each cluster, transforms its vertices into overhead coordinates, and draws the polygon in the view.
Section 2 · machine-generated
Variations
Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.
Other field of use
Overhead guidance for container yard tractors
A yard tractor carries forward and side cameras, and the yard supplies an orthorectified aerial mosaic of the container lanes. An image encoder trained on yard imagery produces embeddings for lane markings, container corners, and temporary barriers in both views. The processor projects camera detections onto the yard surface using camera calibration and tractor pose. It inserts the projected features into the overhead map and supplies the updated lane boundaries to the tractor route planner.
Component substitution
Laser intensity images for ground sensing
A scanning lidar replaces the ground camera and records range and return intensity for each beam. The processor compensates for vehicle motion during each sweep and bins the returns into a top-down image with intensity and height channels. Separate encoders trained on labeled lidar images and aerial imagery produce embeddings for curbs, lane stripes, and raised islands. The processor positions the detected features in the aerial view using their lidar coordinates and the vehicle pose.
Parameter or scale variation
Two grid resolutions for local planning
The processor maintains a 40 by 40 meter grid at 0.10 meter per cell and a 200 by 200 meter grid at 0.50 meter per cell. Both grids share a map coordinate frame and stay centered on the vehicle. Embeddings from the ground sensors and aerial data supply road features to the inner grid at 20 hertz and to the outer grid at 2 hertz. The outer grid aggregates overlapping inner cells by retaining the highest occupancy probability and rasterizing lane boundaries separately.
Combination with existing systems
Overhead map input to parking control
A low-speed parking controller receives an overhead map built from aerial imagery of a parking facility and live images from vehicle surround cameras. Image encoders produce embeddings for stall lines, wheel stops, and vehicles, and the fusion processor places the detected features in a common ground-plane coordinate frame. Ultrasonic range measurements update occupied cells near the bumpers. The parking controller checks the vehicle footprint along candidate paths against the updated grid before issuing steering and speed commands.
Wheel slip monitoring in an overhead grid
A traction monitoring system supplies wheel speeds, steering angle, and an independently estimated vehicle ground speed to the overhead-map processor. The processor uses road-region embeddings from ground cameras and aerial imagery to define the drivable grid, then projects each wheel contact location into that grid at the measurement timestamp. It adjusts the expected wheel speeds for turning geometry and assigns a slip estimate to each observed contact cell. The speed planner uses recent slip estimates along candidate paths to set local speed limits.
Automated or data-driven variant
Visibility based updates for overhead features
A fusion processor stores an aerial feature layer and a live observation layer for the same road segment. A segmentation model assigns class probabilities to features represented by ground-view embeddings, and a depth image identifies the cells visible to the ground sensors. The processor updates a visible cell after three successive ground observations assign the same class a probability of at least 0.85. It holds the last fused value in occluded cells and publishes the observation age with each cell.
Specific version of a part
Training on rendered road scenes
A renderer generates ground and overhead images from procedural lane layouts and varies camera pose, lighting, and pavement texture. Shared road element identifiers pair the two views during training so the model learns feature vectors that associate the same element across viewpoints.
Synchronized forward stereo cameras
Two forward-facing cameras capture synchronized frames at 30 frames per second with a 20-centimeter baseline. The processor rectifies each image pair and estimates depth from pixel disparity in the overlapping field of view.
Georeferenced aerial image tiles
A local cache stores orthorectified aerial image tiles with a ground sampling distance of 10 centimeters per pixel. The processor uses vehicle position, heading, and camera field of view to retrieve the tiles that intersect the visible roadway.
Paired road element embeddings
A two-branch image encoder maps ground and aerial image patches to 64-component vectors. Training pairs from collected driving scenes pull vectors for the same crosswalk or lane marking closer together and push vectors for different elements farther apart.
Ground verified lane marking overlay
The processor matches a live ground image embedding to a lane marking template and projects the detected marking endpoints onto the road plane using camera calibration and vehicle pose. It draws the resulting polyline in the overhead view when the marking is absent within 30 centimeters of the projected position.
Section 3 · quoted from USPTO records
Other work
Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.
1.1obtaining ground view data being sensed in real-time by at least one sensor in a region of view of a ground vehicle;
For example, classifying can be performed based on an auxiliary metric in combination with the steering activity metric (e.g., speed extracted from GPS time series, parameters of the vehicle surroundings extracted from external imagery using computer-vision techniques, etc.).
A · US 2020/0198645 A1 · ¶ 52
One or more non-transitory computer-readable media storing instructions that, when executed by a system comprising one or more processors, cause the one or more processors to perform operations comprising: collecting, using a smartphone, a plurality of ground-level images capturing all sides of a single building; determining a location for the plurality of ground-level images; generating an outline of a plurality of facades associated with all sides of the single building sho
B · US 2022/0284668 A1 · claim 12
1.2obtaining air based data pertaining to a driving environment of the ground vehicle that corresponds to the region of view;
1.3generating, by using a machine learning process, embeddings of the ground view data and the air based data, wherein the machine learning process is trained across environmental driving related data of collected images to produce embeddings for road elements in the collected images;
1.4producing an augmented air based view with reality checks in the air based data for autonomous driving, by adding road elements that are represented by at least one of the embeddings generated based on the ground view data into the augmented air based view;
An empty cell means only that no passage was reproduced there.
Statements in these documents
Thus, there is a need in the automotive field for a system and method for accurately determining driver distraction.
A · US 2020/0198645 A1 · ¶ 4
Additionally, the system and method can quantify driver behavior (e.g., driver risk) on the basis of non-localized driving context and/or driving maneuver, to enable mapless comparisons between driver behaviors, wherein the geographic contexts are disparate and/or unknown.
A · US 2020/0198645 A1 · ¶ 7
The technology described herein relates generally to a system and method for estimating dimensions of buildings, and in particular to a system and method for estimating facade elements of buildings using ground-level and orthogonal imagery.
B · US 2022/0284668 A1 · ¶ 3
However, these 3D maps have limited texture resolution, geometry quality, accurate geo-referencing and are expensive, time consuming and difficult to update and provide no robust real-time image data analytics for various consumer and commercial use cases.
B · US 2022/0284668 A1 · ¶ 5
Cite aspatent.fyi, US 2026/0356389 A1, claim 1, first published 2026-11-20T01:05:31-05:00, https://patent.fyi/US20260356389/, SHA-256 d3683936ac46e33df07db38b7f9babc1dfb58bc388ec0e1cb62908a5424048d8