An image processing apparatus includes a holding unit configured to hold a positional relationship of a plurality of image capturing units, an acquisition unit configured to detect objects from respective images captured by the plurality of image capturing units and acquire positions of the objects on the captured images and geometric attributes of the objects, an associating unit configured to associate the detected objects, based on the positional relationship, the positions and the geometric attributes, and an estimation unit configured to estimate a three-dimensional position of the objects based on the positional relationship and the positions of the detected objects.

The present invention discloses to a selective-laser-melting (SLM) printing defect detection and repair method and system based on a deep learning network and belongs to the technical field of additive manufacturing. The method includes: training a first neural network model through a defect dataset to obtain a defect recognition model; in a printing process of a part to be detected, performing online defect recognition through the defect recognition model; if a current layer has no defects, continuing to perform powder spreading and printing on a next layer; if a defect is recognized in the current layer, selecting whether to repair the defect according to a defect type; if defect repair is needed, after the current layer is printed, suspending powder spreading once, predicting laser remelting parameters by adopting a pre-trained second neural network model, and performing laser remelting repairing until the current layer has no defects; and repeating the processes of online defect recognition and laser remelting repair until the part to be detected is printed. In the present invention, the online defect recognition and repair are realized, and the real-time performance of the defect repair is improved.

A method is provided, which comprises generating at least three images of an area of interest from at least one imaging system, the generated images being provided from at least three different angles, establishing point correspondence between the provided images. The method further involves generating at least two sets of three-dimensional information based on the provided images, wherein the at least two sets of three-dimensional information are generated based on at least two different combinations of at least two of the at least three provided images of the area of interest. The method further includes comparing the at least two sets of three-dimensional information so as to determine discrepancies, and providing information related to the imaging system or errors in the images based on the determined discrepancies.

A computer-implemented method for semantic segmentation of an image determines unary energy of each pixel in an image using a first subnetwork, determines pairwise energy of at least some pairs of pixels of the image using a second subnetwork, and determines, using a third subnetwork, an inference on a Gaussian random field (GRF) minimizing an energy function including a combination of the unary energy and the pairwise energy. The GRF inference defining probabilities of semantic labels for each pixel in the image, and the method converts the image into a semantically segmented image by assigning to a pixel in the semantically segmented image a semantic label having the highest probability for a corresponding pixel in the image among the probabilities determined by the third subnetwork. The first subnetwork, the second subnetwork, and the third subnetwork are parts of a neural network.

A better basis for a further processing such as virtual view rendering, in form of a disparity map is achieved. To this end, the disparity map generation is done in two separate steps, namely the generation of two depth/disparity map estimates based on two different pairs of views of the scene in a manner independent from each other, with then comparing both depth/disparity map estimates so as to obtain a reliability measure for one or both of the depth/disparity map estimates.

In one embodiment, a device comprises interface circuitry and processing circuitry. The processing circuitry receives, via the interface circuitry, a video stream captured by a camera during performance of an industrial process, wherein the video stream comprises a sequence of frames; detects, based on analyzing the sequence of frames, a degree of particle scatter that occurs during performance of the industrial process; and determines, based on the degree of particle scatter, that an anomaly occurs during performance of the industrial process.

Systems and methods are disclosed for tracking objects in a physical environment using visual sensors onboard an autonomous unmanned aerial vehicle (UAV). In certain embodiments, images of the physical environment captured by the onboard visual sensors are processed to extract semantic information about detected objects. Processing of the captured images may involve applying machine learning techniques such as a deep convolutional neural network to extract semantic cues regarding objects detected in the images. The object tracking can be utilized, for example, to facilitate autonomous navigation by the UAV or to generate and display augmentative information regarding tracked objects to users.

The present invention discloses to a selective-laser-melting (SLM) printing defect detection and repair method and system based on a deep learning network and belongs to the technical field of additive manufacturing. The method includes: training a first neural network model through a defect dataset to obtain a defect recognition model; in a printing process of a part to be detected, performing online defect recognition through the defect recognition model; if a current layer has no defects, continuing to perform powder spreading and printing on a next layer; if a defect is recognized in the current layer, selecting whether to repair the defect according to a defect type; if defect repair is needed, after the current layer is printed, suspending powder spreading once, predicting laser remelting parameters by adopting a pre-trained second neural network model, and performing laser remelting repairing until the current layer has no defects; and repeating the processes of online defect recognition and laser remelting repair until the part to be detected is printed. In the present invention, the online defect recognition and repair are realized, and the real-time performance of the defect repair is improved.

Techniques are described herein for generating a virtual try-on augmented image. An example method can include A system can access a first two-dimensional model of a target garment. The system can determine a first plurality of shape keypoints for a first part of the first plurality of parts of the image. The system can access a second two-dimensional image of a first user. The system can determine a second plurality of keypoints for a second part of the second plurality of parts of the second two-dimensional image, a second number of the second plurality of keypoints based at least in part on a second predetermined number of keypoints associated with the second part. The system can generate a rendered augmented image of the target garment on the target user.

Systems and methods are disclosed for tracking objects in a physical environment using visual sensors onboard an autonomous unmanned aerial vehicle (UAV). In certain embodiments, images of the physical environment captured by the onboard visual sensors are processed to extract semantic information about detected objects. Processing of the captured images may involve applying machine learning techniques such as a deep convolutional neural network to extract semantic cues regarding objects detected in the images. The object tracking can be utilized, for example, to facilitate autonomous navigation by the UAV or to generate and display augmentative information regarding tracked objects to users.