Provided are methods for semantic annotation of sensor data using unreliable map annotation inputs, which can include training a machine learning model to accept inputs including images representing sensor data for a geographic area and unreliable semantic annotations for the geographic area. The machine learning model can be trained against validated semantic annotations for the geographic area, such that subsequent to training, additional images representing sensor data and additional unreliable semantic annotations can be passed through the neural network to provide predicted semantic annotations for the additional images. Systems and computer program products are also provided.

In the field of video processing, a detection and tracking method and apparatus, and a storage medium, are provided. The method includes: performing feature point analysis on a video frame sequence, to obtain feature points on each video frame thereof; performing target detection on an extracted frame through a first thread based on the feature points, to obtain a target box in the extracted frame; performing target box tracking in a current frame through a second thread based on the feature points and the target box in the extracted frame, to obtain a result target box in the current frame; and outputting the result target box. As the target detection and the target tracking are divided into two threads, a tracking frame rate is unaffected by a detection algorithm, and the target box of the video frame can be outputted in real time, improving real-time performance and stability.

In the field of video processing, a detection and tracking method and apparatus, and a storage medium, are provided. The method includes: performing feature point analysis on a video frame sequence, to obtain feature points on each video frame thereof; performing target detection on an extracted frame through a first thread based on the feature points, to obtain a target box in the extracted frame; performing target box tracking in a current frame through a second thread based on the feature points and the target box in the extracted frame, to obtain a result target box in the current frame; and outputting the result target box. As the target detection and the target tracking are divided into two threads, a tracking frame rate is unaffected by a detection algorithm, and the target box of the video frame can be outputted in real time, improving real-time performance and stability.

A camera with a field of view toward an external environment of an aircraft is disposed within an aircraft door such that a ground surface is within the field of view of the camera during taxiing of the aircraft. A display device is disposed within an interior of the aircraft. A processor is operatively coupled to the camera and to the display device. The processor analyzes image data captured by the camera for docking guidance by identifying, within the captured image data, a region on the ground surface corresponding to an alignment fiducial indicating a parking location for the aircraft, determining, based on the region of the captured image data corresponding to the alignment fiducial indicating the parking location, a relative location of the aircraft with respect to the alignment fiducial, and outputting an indication of the relative location of the aircraft to the alignment fiducial.

An example method includes receiving a 2D image of a 3D space from an optical camera, identifying, in the 2D image. A virtual image generated by an optical instrument refracting and/or reflecting the light is identified. The example method further includes identifying, in the 2D image, a first object depicting a subject disposed in the 3D space from a first direction extending from the optical camera to the subject and identifying, in the virtual image, a second object depicting the subject disposed in the 3D space from a second direction extending from the optical camera to the subject via the optical instrument, the second direction being different than the first direction. A 3D image depicting the subject based on the first object and the second object is generated. Alternatively, a location of the subject in the 3D space is determined based on the first object and the second object.

An image processing device includes a hardware processor. The hardware processor designates, from one frame image of a dynamic image acquired by dynamic imaging of a movement of a locomotorium, a plurality of regions or points on a structure included in the locomotorium, sets an alignment reference based on the designated regions or points, tracks the designated regions or points in a plurality of frame images of the dynamic image, aligns a line segment connecting the regions or the points to each other in the plurality of frame images based on the alignment reference, and causes a display to display the line segment so as to be superimposed on a representative frame image of the dynamic image.

An image processing device includes a hardware processor. The hardware processor designates, from one frame image of a dynamic image acquired by dynamic imaging of a movement of a locomotorium, a plurality of regions or points on a structure included in the locomotorium, sets an alignment reference based on the designated regions or points, tracks the designated regions or points in a plurality of frame images of the dynamic image, aligns a line segment connecting the regions or the points to each other in the plurality of frame images based on the alignment reference, and causes a display to display the line segment so as to be superimposed on a representative frame image of the dynamic image.

The present disclosure describes techniques for facial expression recognition. A first loss function may be determined based on a first set of feature vectors associated with a first set of images depicting facial expressions and a first set of labels indicative of the facial expressions. A second loss function may be determined based on a second set of feature vectors associated with a second set of images depicting asymmetric facial expressions and a second set of labels indicative of the asymmetric facial expressions. The first loss function and the second loss function may be used to determine a maximum loss function. The maximum loss function may be applied during training of a model. The trained model may be configured to predict at least one asymmetric facial expression in a subsequently received image.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for monitoring events using a Virtual Inductive Loop system. In some implementations, image data is obtained from cameras. A region depicted in the obtained image data is identified, the region comprising lines spaced by a distance that satisfies a distance threshold. For each line included in the region: an object depicted crossing the line is determined whether to satisfy a height criteria indicating that the line is activated. In response to determining that an object depicted crossing the line satisfies the height criteria, an event is determined to have likely occurred using data indicating (i) which lines of the lines were activated and (ii) an order in which each of the lines were activated. In response to determining that an event likely occurred, actions are performed using at least some of the data.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for monitoring events using a Virtual Inductive Loop system. In some implementations, image data is obtained from cameras. A region depicted in the obtained image data is identified, the region comprising lines spaced by a distance that satisfies a distance threshold. For each line included in the region: an object depicted crossing the line is determined whether to satisfy a height criteria indicating that the line is activated. In response to determining that an object depicted crossing the line satisfies the height criteria, an event is determined to have likely occurred using data indicating (i) which lines of the lines were activated and (ii) an order in which each of the lines were activated. In response to determining that an event likely occurred, actions are performed using at least some of the data.