A high-definition map system receives sensor data from vehicles traveling along routes and combines the data to generate a high definition map for use in driving vehicles, for example, for guiding autonomous vehicles. A pose graph is built from the collected data, each pose representing location and orientation of a vehicle. The pose graph is optimized to minimize constraints between poses. Points associated with surface are assigned a confidence measure determined using a measure of hardness/softness of the surface. A machine-learning-based result filter detects bad alignment results and prevents them from being entered in the subsequent global pose optimization. The alignment framework is parallelizable for execution using a parallel/distributed architecture. Alignment hot spots are detected for further verification and improvement. The system supports incremental updates, thereby allowing refinements of subgraphs for incrementally improving the high-definition map for keeping it up to date.

A high-definition map system receives sensor data from vehicles traveling along routes and combines the data to generate a high definition map for use in driving vehicles, for example, for guiding autonomous vehicles. A pose graph is built from the collected data, each pose representing location and orientation of a vehicle. The pose graph is optimized to minimize constraints between poses. Points associated with surface are assigned a confidence measure determined using a measure of hardness/softness of the surface. A machine-learning-based result filter detects bad alignment results and prevents them from being entered in the subsequent global pose optimization. The alignment framework is parallelizable for execution using a parallel/distributed architecture. Alignment hot spots are detected for further verification and improvement. The system supports incremental updates, thereby allowing refinements of subgraphs for incrementally improving the high-definition map for keeping it up to date.

A displacement measurement device includes: an obtainer that obtains a first image which contains a subject and a second image which contains the subject; a generator that generates M template images which contain the subject and which have noise from the first image and generates M target images which contain the subject and which have noise from the second image, M being an integer of 2 or higher; a hypothetical displacement calculator that calculates M hypothetical displacements of the subject from the M template images and the M target images; and a displacement calculator that calculates a displacement of the subject by performing statistical processing on the M hypothetical displacements.

A method (100) of controlling a portable device comprising a first camera and a second camera facing in the same direction. The method comprises: selecting (110) one of the first camera and the second camera as a visualization camera; initializing (120) a localization algorithm having as an input image data representing images captured by one of the first camera and the second camera; determining (130) a respective focus score for at least one of the first camera and the second camera, said focus score indicating a focus quality of features identified from images captured by one of the respective camera; selecting (140,) one of the first camera and the second camera as an enabled camera based on the at least one focus score; and generating a control signal configured to cause the selected camera to be enabled such that the image data representing images captured by the enabled camera are provided as the input to the localization algorithm.

A method (100) of controlling a portable device comprising a first camera and a second camera facing in the same direction. The method comprises: selecting (110) one of the first camera and the second camera as a visualization camera; initializing (120) a localization algorithm having as an input image data representing images captured by one of the first camera and the second camera; determining (130) a respective focus score for at least one of the first camera and the second camera, said focus score indicating a focus quality of features identified from images captured by one of the respective camera; selecting (140,) one of the first camera and the second camera as an enabled camera based on the at least one focus score; and generating a control signal configured to cause the selected camera to be enabled such that the image data representing images captured by the enabled camera are provided as the input to the localization algorithm.

A high-definition map system receives sensor data from vehicles travelling along routes and combines the data to generate a high definition map for use in driving vehicles, for example, for guiding autonomous vehicles. A pose graph is built from the collected data, each pose representing location and orientation of a vehicle. The pose graph is optimized to minimize constraints between poses. Points associated with surface are assigned a confidence measure determined using a measure of hardness/softness of the surface. A machine-learning-based result filter detects bad alignment results and prevents them from being entered in the subsequent global pose optimization. The alignment framework is parallelizable for execution using a parallel/distributed architecture. Alignment hot spots are detected for further verification and improvement. The system supports incremental updates, thereby allowing refinements of subgraphs for incrementally improving the high-definition map for keeping it up to date.

A high-definition map system receives sensor data from vehicles travelling along routes and combines the data to generate a high definition map for use in driving vehicles, for example, for guiding autonomous vehicles. A pose graph is built from the collected data, each pose representing location and orientation of a vehicle. The pose graph is optimized to minimize constraints between poses. Points associated with surface are assigned a confidence measure determined using a measure of hardness/softness of the surface. A machine-learning-based result filter detects bad alignment results and prevents them from being entered in the subsequent global pose optimization. The alignment framework is parallelizable for execution using a parallel/distributed architecture. Alignment hot spots are detected for further verification and improvement. The system supports incremental updates, thereby allowing refinements of subgraphs for incrementally improving the high-definition map for keeping it up to date.

A method for improved image acquisition for photogrammetry includes focusing a camera on one end of an object, capturing one or more images of the object, incrementally adjusting the focal length of the camera toward the opposite end of the object, and capturing images at each new focal length. Once the object has been photographed at varying focal lengths that run the entire length of the object, the multitude of images are then combined using focus stacking to create a singular image that is more in focus for the entire length of the object. A method for utilizing thermographic cameras to aid in the acquisition of images for photogrammetry includes applying thermal textures to the object and isolating an object from the background due to thermal differences.

A method includes generating calibration data by geometrically calibrating first image data from a first camera unit relative to second image data from a second camera unit based on first descriptor data and second descriptor data. The first descriptor data is based on the first image data. The second descriptor data is based on the second image data. The calibration data is generated based on first position data corresponding to the first camera unit and second position data corresponding to the second camera unit. The method includes identifying, based on the calibration data, a target location relative to the first image data. The method further includes generating an output image that includes the first image data and an indication of where the target location is relative to a scene depicted in the first image data.

A method includes generating calibration data by geometrically calibrating first image data from a first camera unit relative to second image data from a second camera unit based on first descriptor data and second descriptor data. The first descriptor data is based on the first image data. The second descriptor data is based on the second image data. The calibration data is generated based on first position data corresponding to the first camera unit and second position data corresponding to the second camera unit. The method includes identifying, based on the calibration data, a target location relative to the first image data. The method further includes generating an output image that includes the first image data and an indication of where the target location is relative to a scene depicted in the first image data.