Here Global B.v.

David Johnston Lawlor

Anish Mittal

Zhanwei Chen

Richard Kwant

Himaanshu Gupta

Nicholas Pojman

Jeff Connell

Mark David Tabb

Yelena Perelmutova

Niranjan Kotha

Krishna Kumar Balakrishnan

Peter Christian

Nick Pojman

Abinaya Manimaran

Alex Ung

A method, apparatus and computer program product are provided for establishing correspondences between images using through generation of a translation between the different perspectives. Methods may include: receiving first sensor data from a first image sensor, where the first sensor data includes a first image of an environment captured from a first perspective; receiving second sensor data from a second image sensor of a second image of the environment captured from a second perspective; identifying image correspondence points between the first sensor data and the second sensor data; computing pairwise vectors between corresponding pairs of first projected points of the first sensor data and second projected points of the second sensor data; clustering the pairwise vectors according to magnitude and orientation; and generating a translation vector from clusters of pairwise vectors, where the translation vector represents a shift of the second image sensor data to correspond to ground truth data.

Method, apparatus, and computer program product for point-to-point translation between images

Systems and methods for validating drive pose refinement are provided. In some aspects, a method includes receiving image data that depicts an area of interest, and receiving a plurality of virtual points generated using the image data. The method also includes selecting at least one drive in the area of interest that captures the plurality of virtual points, and generating a refined pose track for each of the at least one drive by applying a drive alignment process to drive data from the at least one drive using the plurality virtual points. The method further includes comparing the refined pose track to a control pose track generated using control repoints, and generating, based on the comparison, a report that validates the refined pose track.

Systems and methods for validating drive pose refinement

An approach is provided for an asymmetric evaluation of polygon similarity. The approach, for instance, involves receiving a first polygon representing an object depicted in an image. The approach also involves generating a transformation of the image comprising image elements whose values are based on a respective distance that each image element is from a nearest image element located on a first boundary of the first polygon. The approach further involves determining a subset of the plurality of image elements of the transformation that intersect with a second boundary of a second polygon. The approach further involves calculating a polygon similarity of the second polygon with respect the first polygon based on the values of the subset of image elements normalized to a length of the second boundary of the second polygon.

Method, apparatus, and system for determining polyline homogeneity

Systems and methods for correcting mapping information inaccuracies are described. In some aspects, the method includes receiving terrestrial data captured in an area of interest, and detecting features in the terrestrial data identifying ground points in the area of interest. The method also includes correlating the ground points with ground control points in the area of interest to determine a correspondence, and computing an aggregate of positional differences between corresponding points. The method further includes generating a report indicating a quality of the terrestrial data captured in the area of interest based on the aggregate.

Systems and methods for controlling mapping information inaccuracies

A method is provided for generating and revising map geometry based on a received image and probe data. A method may include: receiving probe data from a first period of time, where the probe data from a first period of time is from a plurality of probes within a predefined geographic region; generating a first image of the predefined geographic region based on the probe data from the first period of time; receiving probe data from a second period of time different from the first period of time, where the probe data from the second period of time is from a plurality of probes within the predefined geographic region; generating a second image based on the probe data from the second period of time; comparing the first image to the second image; and generating a revised route geometry based on changes detected between the first image and the second image.

Method and apparatus for updating road map geometry based on received probe data

An approach is provided for image labeling for cross view alignment. The approach, for example, involves determining camera pose data, camera trajectory data, or a combination thereof for a first image depicting an area from a first perspective view. The approach also involves processing the camera pose data, the camera trajectory data, or a combination thereof to generate meta data indicating a position, an orientation, or a combination thereof of the first perspective view of the area relative to a second image depicting the area from a second perspective view. The approach further involves providing data for presenting the meta data in a user interface as an overlay on the second perspective view.

Method, apparatus, and system for providing image labeling for cross view alignment

An approach is provided for a redundant feature detection engine. The approach, for instance, involves segmenting an input image into a plurality of grid cells for processing by the redundant feature detection engine. The redundant feature detection engine includes a neural network. The approach also involves, for each of the plurality of grid cells, initiating a prediction of an object code by the redundant feature detection engine. The object code is a predicted feature that uniquely identifies an object depicted in the input image. The approach further involves aggregating the plurality of grid cells into one or more clusters based on the object code predicted for said each grid cell. The approach further involves predicting one or more features of the object corresponding to a respective cluster of the one or more clusters by merging one or more feature prediction outputs of said each grid cell in the respective cluster.

Method, apparatus, and system for providing a redundant feature detection engine

Systems and methods for verifying mapping information are provided. In some aspects, a method includes receiving control data acquired in an area of interest, the control data comprising a plurality of control points, and receiving mapping data associated with the area of interest, the mapping data comprising a plurality of mapping points that correspond to the plurality control points. The method also includes applying a localization algorithm to the control data to generate a control track, and applying the localization algorithm to the mapping data to generate a mapping track. The method further includes comparing the control track and the mapping track to determine a difference.

Systems and methods for verifying mapping information

An approach is provided for a location-aware evaluation of a machine learning model. The approach, for example, involves designating a geographic area for creating an evaluation dataset for the machine learning model. The approach also involves separating a plurality of observation data records into the evaluation dataset and a training dataset based on a comparison of a respective data collection location of each of the plurality of observation data records to the geographic area. The training dataset is then used to train the machine learning model, and the evaluation dataset is used to evaluate the trained machine learning model.

Method, apparatus, and system for providing a location-aware evaluation of a machine learning model

The disclosure provides methods, apparatus, and products for attention guidance and labeling. In one aspect, a method comprises receiving metadata associated with a perspective image captured by an image capture device located at the image position and having the image pose, wherein the metadata comprises the image position and the image pose. The method comprises defining a field-of-view indicator having a first indicator position located at the image position and having the image pose; receiving ground control point (GCP) information identifying a GCP, wherein the GCP information comprises a GCP position; defining a GCP indicator having a second indicator position located at the GCP position; and causing display of an indicator layer comprising the field-of-view indicator and the GCP indicator in a second portion of an interactive user interface (IUI) of a labeling tool, wherein at least the perspective image is displayed in a first portion of the IUI.

Attention guidance for ground control labeling in street view imagery

An approach is provided for determining a polygon of a geographic database that overlaps a candidate polygon or candidate point. The geographic database represents stored polygons as respective polygon points with zero area. The approach involves determining proximate polygon points from among the respective polygon points with zero area that are within a distance threshold of the candidate polygon or the candidate point. The approach also involves retrieving one or more proximate polygons from the geographic database that correspond to the one or more proximate polygon points. The approach further involves determining an intersection between the one or more proximate polygons and the candidate polygon or the candidate point. The approach then involves selecting the polygon that overlaps the candidate polygon or the candidate point based on the determined intersection.

Method for determining polygons that overlap with a candidate polygon or point

An approach is provided for generating a super-resolution image as a higher resolution version of an input image. The approach, for example, involves determining a set of tasks to be performed on the input image to facilitate generating the super-resolution image. The approach also involves selecting a combination of loss functions, wherein each loss function of the combination of loss functions is respectively a task-specific neural network pre-trained to perform a corresponding one of the set of tasks. The approach also involves training the super resolution neural network using the combination of loss functions as one or more layers of the super resolution neural network. The approach also involves using the trained super resolution neural network to generate the super-resolution image as a higher resolution version of the input image.

Method, apparatus, and system for task driven approaches to super resolution

An approach is provided for predicting a pose error for a sensor system based on a trained machine learning model. The approach, for example, involves receiving images depicting a survey point with a known physical location. The approach also involves determining meta-data associated with the sensor system used to capture the images. The approach further involves generating a ray from the capture location through a pixel location of the survey point on an image plane of each image. The approach further involves calculating an error between the ray generated for the image and the known physical location. The approach further involves training a machine learning model to predict a pose error from image data captured using the sensor system based on the error in combination with features extracted from the image and the meta-data for the image. The approach further involves providing the trained machined learning as an output.

Method, apparatus, and system for predicting a pose error for a sensor system

An approach is provided for location correction using feature point correspondences between images or image sources. The approach, for example, involves processing a first image and a second image to identify a feature that is visible in both the first image and the second image. The first image has an image-to-ground correspondence between the feature in the first image and a geographic location. The approach also involves transferring the image-to-ground correspondence from the first image to the second image. The approach further involves using the transferred image-to-ground correspondence to reference the feature in the second image to the geographic location.

Method, apparatus, and system for location correction based on feature point correspondence

An approach is provided for determining a ground control point from image data using machine learning. The approach, for example, involves selecting an feature based determining that the feature meets one or more properties for classification as a machine learnable feature. The approach also involves retrieving a plurality of ground truth images depicting the feature. The plurality of ground truth images is labeled with known pixel location data of the feature as respectively depicted in each of the plurality of ground truth images. The approach further involves training a machine learning model using the plurality of ground truth images to identify predicted pixel location data of the ground control point as depicted in an input image.

Growing community of inventors

Evanston, IL, United States of America

David Johnston Lawlor