General Electric Company

Eric Michael Gros

Peter Henry Tu

Arpit Jain

Ming-Ching Chang

Junli Ping

Charles Burton Theurer

Huan Tan

Mauricio Castillo-Effen

David Erik Chevalier

Jiang Hsieh

John Michael Lizzi

Romano Patrick

John Howard Londt

John Irvin Jackson

Roy A Nilsen

Jie Tang

Ana Dasilva

Dave Chevalier

The present approach relates to an automatic and efficient motion plan for a drone to collect and save a qualified dataset that may be used to improve reconstruction of 3D models using the acquired data. The present architecture provides an automatic image processing context, eliminating low quality images and providing improved image data for point cloud generation and texture mapping.

Image and video capture architecture for three-dimensional reconstruction

The present approach relates to the use of a point cloud of an object to initialize or seed a space carving technique used to generate a 3D model of the object. In one implementation, feature matching is performed on 2D images, with matched features constituting the points of a point cloud model. The point cloud generated in this manner, is one input of a foreground/background segmentation algorithm, which generates a set of segmented 2D images used by a space carving routine to generate a 3D model of the object.

Three-dimensional modeling of an object

Systems and methods for improved image denoising using a deep learning network model are disclosed. An example system includes an input data processor to process a first patient image of a first patient to add a first noise to the first patient image to form a noisy image input. The example system includes an image data denoiser to process the noisy image input using a first deep learning network to identify the first noise. The image data denoiser is to train the first deep learning network using the noisy image input. When the first deep learning network is trained to identify the first noise, the image data denoiser is to deploy the first deep learning network as a second deep learning network model to be applied to a second patient image of the first patient to identify a second noise in the second patient image.

Patient-specific deep learning image denoising methods and systems

Methods and systems are provided for generating deep learning training data with an imaging system. In one embodiment, a method for an imaging system comprises performing a scan of a subject to acquire imaging data, training a deep neural network on the imaging data to obtain updates to the deep neural network, and transmitting the updates to a server for training a central deep neural network. In this way, imaging data may be leveraged for training and developing global deep learning models without transmitting the imaging data itself, thereby preserving patient privacy.

Systems and methods for capturing deep learning training data from imaging systems

Modifying a motion plan for an autonomously-operated inspection platform (AIP) includes obtaining sensor data for an industrial asset area of interest, analyzing the obtained sensor data during execution of an initial motion plan to determine if modification of the initial motion plan is required. If modification is required then performing a pose estimation on a first group of potential targets and a second group of potential targets, optimizing the results of the pose estimation to determine a modification to the initial motion plan, performing reactive planning to the initial motion plan to include the modification, the reactive planning providing a modified motion plan that includes a series of waypoints defining a modified path, and autonomously controlling motion of the AIP along the modified path. The analysis, pose estimation, optimization, and reactive planning occurring during movement of the AIP along a motion plan. A system and computer-readable medium are disclosed.

Autonomously-controlled inspection platform with model-based active adaptive data collection

Methods and systems are provided for capturing deep learning training data from imaging systems. In one embodiment, a method for an imaging system comprises performing a scan of a subject to acquire imaging data, inputting the imaging data to a deep neural network, displaying an output of the deep neural network and an image reconstructed from the imaging data, and transmitting an intermediate representation of the imaging data generated by the deep neural network to a server for training a central deep neural network. In this way, imaging data may be leveraged for training and developing global deep learning models without transmitting the imaging data itself, thereby preserving patient privacy.

A robotic system includes a processing system comprising at least one processor. The processor generates a plan to monitor the asset. The plan comprises one or more tasks to be performed by the at least one robot. The processor receives sensor data from at least one sensor indicating one or more characteristics of the asset. The processor adjusts the plan to monitor the asset by adjusting or adding one or more tasks to the plan based on one or both of the quality of the acquired data or a potential defect of the asset. The adjusted plan causes the at least one robot to acquire additional data related to the asset when executed.

Robot system path planning for asset health management

An imaging system determines a heart rate of a patient, a cardiac disease state, and/or a target cardiac anatomy of the patient. The system calculates an image acquisition time range from a patient population model using the heart rate, the cardiac disease state, and/or the target anatomy. The model represents relationships between cardiac motion of other patients and time or cardiac phases of the other patients. The system also determines imaging settings to acquire image data of the target anatomy during the image acquisition time range that is calculated. Imaging the target anatomy of the patient according to the imaging settings generates image data of the target anatomy having less cardiac motion and/or a reduced image acquisition time range relative to determining the imaging settings without using the patient population model.

Growing community of inventors

Waukesha, WI, United States of America

Eric Michael Gros