Abstract: Methods, systems, and apparatuses, among other things, may detect and store activity in videos based on a spatiotemporal graph representation. Spatiotemporal proximity graphs may be built based on one or more received tracks and may include one or more nodes and each node may include one or more attributes associated with a corresponding entity. One or more spatiotemporal relationships may be identified between the entities based on each spatiotemporal proximity graph one or more activities of the entities may be identified based on the spatiotemporal relationships.

Abstract: A method of supporting an aircraft approaching a runway is provided. Example implementations involve receiving a sequence of images, captured by a camera onboard the aircraft. Example implementations involve applying a received image to a machine learning model trained to detect the runway or a runway marking in the image, and to produce a mask that includes a segment of pixels of the image assigned to an object class for the runway or marking. Example implementations may also involve applying the mask to a corner detector to detect interest points on the mask and match the interest points to corresponding points on the runway or the marking that have known runway-framed local coordinates. Example implementations may also involve performing a perspective-n-point estimation to determine a current pose estimate of the aircraft relative to the runway or the marking and outputting the current pose estimate for use during final approach.

Abstract: A method of machine learning model development includes building an autoencoder including an encoder trained to map an input into a latent representation, and a decoder trained to map the latent representation to a reconstruction of the input. The method includes building an artificial neural network classifier including the encoder, and a classification layer partially trained to perform a classification in which a class to which the input belongs is predicted based on the latent representation. Neural network inversion is applied to the classification layer to find inverted latent representations within a decision boundary between classes in which a result of the classification is ambiguous, and inverted inputs are obtained from the inverted latent representations. Each inverted input is labeled with a class that is its ground truth, and thereby producing added training data for the classification, and the classification layer is further trained using the added training data.

Abstract: Methods, systems, and apparatuses, among other things, may detect and store activity in videos based on a spatiotemporal graph representation. Spatiotemporal proximity graphs may be built based on one or more received tracks and may include one or more nodes and each node may include one or more attributes associated with a corresponding entity. One or more spatiotemporal relationships may be identified between the entities based on each spatiotemporal proximity graph one or more activities of the entities may be identified based on the spatiotemporal relationships.

Abstract: A method of supporting an aircraft approaching a runway is provided. Example implementations involve receiving a sequence of images, captured by a camera onboard the aircraft. Example implementations involve applying a received image to a machine learning model trained to detect the runway or a runway marking in the image, and to produce a mask that includes a segment of pixels of the image assigned to an object class for the runway or marking. Example implementations may also involve applying the mask to a corner detector to detect interest points on the mask and match the interest points to corresponding points on the runway or the marking that have known runway-framed local coordinates. Example implementations may also involve performing a perspective-n-point estimation to determine a current pose estimate of the aircraft relative to the runway or the marking and outputting the current pose estimate for use during final approach.

Abstract: A system may be configured to perform label recollection, e.g., by automatically snapping, via a trained ML model, a set of vector labels by aligning one or more of the labels to an image, the alignment being performed at a quality that satisfies a criterion. Before this automatic snapping or matching of vectorized labels with reference imagery, this ML model may obtain training data from an output of another trained ML model. In another context, a computer-implemented method is disclosed for creating training data that better aligns labels with corresponding image features. This training data, created with reduced effort yet increased quality, may then be fed into to existing models, resulting in an automated pipeline.

Abstract: An artificial intelligence (AI) system may be configured to efficiently annotate most if not all unlabeled image data. Some embodiments may: provide, to an object-detection, machine-learning (ML) model, a plurality of unlabeled data such that the object-detection model predicts a plurality of regions; correct at least one vertex of bounds of at least one of the regions such that the bounds fit tighter around an object; convert the regions to first subregions by cropping the first subregions from the unlabeled data; and provide the first subregions to an embedding, ML model configured to output feature vectors for each of the first subregions.

Abstract: Methods, systems, and apparatuses, among other things, may perform persistent object tracking and reidentification through detection and continuous feature comparison. For example, video frames may be received (e.g., from a camera, an application, or a data storage device) and an object of interest may be detected at a first position in a video frame and the object of interest may be detected at a second position in another video frame. A track associated with the object of interest may be generated based on the detected first and second positions of the object of interest.

Abstract: A system may be configured to detect an unseen object. Some embodiments may: train a machine learning (ML) model, with training data and with both a positive-support content item and a negative-support content item; and predict, via the trained ML model, presence, within a region, of an object in a newly-obtained content item. The object may (i) not have previously been used to train the ML model and (ii) be among a background and a candidate object present in the newly-obtained content item.

Abstract: Systems, methods, and computer-readable media storing instructions for determining cross-track error of an aircraft on a taxiway are disclosed herein. The disclosed techniques capture electronic images of a portion of the taxiway using cameras or other electronic imaging devices mounted on the aircraft, pre-process the electronic images to generate regularized image data, apply a trained multichannel neural network model to the regularized image data to generate a preliminary estimate of cross-track error relative to the centerline of the taxiway, and post-process the preliminary estimate to generate an estimate of cross-track error of the aircraft. Further embodiments adjust a GPS-based location estimate of the aircraft using the estimate of cross-track error or adjust the heading of the aircraft based upon the estimate of cross-track error.

Abstract: A method of machine learning model development includes receiving a plurality of images of a scene, and performing an unsupervised image selection. This includes applying the images to a pre-trained model to extract and embed the images with respective feature vectors, and performing a cluster analysis to group the images in a clusters based on correlations among the respective feature vectors. The unsupervised image selection also includes selecting at least some but not all images in each of the clusters, and any images considered outliers that belong to none of the clusters, for a subset of the images that includes fewer than all of the images. And the method includes receiving user input to label or labeling objects depicted in the subset of the images to produce a training set of images, and building a machine learning model for object detection using the training set of images.

Abstract: A method of machine learning model development includes building an autoencoder including an encoder trained to map an input into a latent representation, and a decoder trained to map the latent representation to a reconstruction of the input. The method includes building an artificial neural network classifier including the encoder, and a classification layer partially trained to perform a classification in which a class to which the input belongs is predicted based on the latent representation. Neural network inversion is applied to the classification layer to find inverted latent representations within a decision boundary between classes in which a result of the classification is ambiguous, and inverted inputs are obtained from the inverted latent representations. Each inverted input is labeled with a class that is its ground truth, and thereby producing added training data for the classification, and the classification layer is further trained using the added training data.

Abstract: A method of machine learning model development includes receiving a plurality of images of a scene, and performing an unsupervised image selection. This includes applying the images to a pre-trained model to extract and embed the images with respective feature vectors, and performing a cluster analysis to group the images in a clusters based on correlations among the respective feature vectors. The unsupervised image selection also includes selecting at least some but not all images in each of the clusters, and any images considered outliers that belong to none of the clusters, for a subset of the images that includes fewer than all of the images. And the method includes receiving user input to label or labeling objects depicted in the subset of the images to produce a training set of images, and building a machine learning model for object detection using the training set of images.