The present invention proposes an ensemble deep learning method for identifying unsafe behaviors of operators in maritime working environment. Firstly, extract features of maritime images with the You Only Look Once (YOLO) V3 model, and then enhance a multi-scale detection capability by introducing a feature pyramid structure. Secondly, obtain instance-level features and time memory features of the operators and devices in the maritime working environment with the Joint Learning of Detection and Embedding (JDE) paradigm. Thirdly, transfer spatial-temporal interaction information into a feature memory pool, and update the time memory features with the asynchronous memory updating algorithm. Finally, identify the interaction between the operators, the devices, and unsafe behaviors with an asynchronous interaction aggregation network. The proposed invention can accurately determine the unsafe behaviors of the operators, and thus provide operation decisions for maritime management relevant activities.

A method for action recognition in a video is described. The method includes inputting a plurality of consecutive clips divided from the video into a convolutional neural network (CNN), and obtaining a set of clip descriptors; processing the set of clip descriptors via a Bi-directional Attention mechanism, and obtaining a global representation of the video; and performing video-classification for the global representation of the video such that action recognition is achieved.

A method for action recognition in a video is described. The method includes inputting a plurality of consecutive clips divided from the video into a convolutional neural network (CNN), and obtaining a set of clip descriptors; processing the set of clip descriptors via a Bi-directional Attention mechanism, and obtaining a global representation of the video; and performing video-classification for the global representation of the video such that action recognition is achieved.

A method for tracking one or more objects in a specific space is provided. The method includes steps of: (a) inputting original images of the specific space taken from camera to an obfuscation network and instructing the obfuscation network to obfuscate the original images to generate obfuscated images such that the obfuscated images are not identifiable as the original images by a human but the obfuscated images are identifiable as the original images by a learning network; (b) inputting the obfuscated images into the learning network, and instructing the learning network to detect obfuscated target objects, corresponding to target objects to be tracked, in the obfuscated images, to thereby output information on the obfuscated target objects; and (c) tracking the obfuscated target objects in the specific space by referring to the information on the obfuscated target objects.

A method for generating a relighted image includes: obtaining a to-be-processed image and a guidance image corresponding to the to-be-processed image; obtaining a first intermediate image consistent with an illumination condition in the guidance image by performing relighting rendering on the to-be-processed image in a time domain based on the guidance image; obtaining a second intermediate image consistent with the illumination condition in the guidance image by performing relighting rendering on the to-be-processed image in a frequency domain based on the guidance image; and obtaining a target relighted image corresponding to the to-be-processed image based on the first intermediate image and the second intermediate image.

A method for generating a relighted image includes: obtaining a to-be-processed image and a guidance image corresponding to the to-be-processed image; obtaining a first intermediate image consistent with an illumination condition in the guidance image by performing relighting rendering on the to-be-processed image in a time domain based on the guidance image; obtaining a second intermediate image consistent with the illumination condition in the guidance image by performing relighting rendering on the to-be-processed image in a frequency domain based on the guidance image; and obtaining a target relighted image corresponding to the to-be-processed image based on the first intermediate image and the second intermediate image.

A tracking device that includes a tracking unit that extracts image frames from video data based on a time threshold value that is a reference for extracting the image frames to be collated with a verification frame that is a verification target, detects a tracking target in the image frames, and sets a collation range for the tracking target based on a space threshold value that is a reference of the collation range for the tracking target, a display information generation unit that generates display information including a tracking image in which the collation range is associated with the tracking target in the image frames and an operation image for setting the time threshold value and the space threshold value, and a threshold value update unit that updates the space threshold value and the time threshold value with values set by the user who refers to the display information.

Implementations are described herein for predicting soil organic carbon (“SOC”) content for agricultural fields detected in digital imagery. In various implementations, one or more digital images depicting portion(s) of one or more agricultural fields may be processed. The one or more digital images may have been acquired by a vision sensor carried through the field(s) by a ground-based vehicle. Based on the processing, one or more agricultural inferences indicating agricultural practices or conditions predicted to affect SOC content may be determined. Based on the agricultural inferences, one or more predicted SOC measurements for the field(s) may be determined.

Systems and methods are provided for detecting a flashing light on one or more traffic signal devices. The method includes capturing a series of images of one or more traffic signal elements in a traffic signal device over a length of time. The method further includes, for each traffic signal element, analyzing the series of images to determine one or more time periods at which the traffic signal element is in an on state or an off state, and analyzing the time periods to determine one or more distinct on states and one or more distinct off states. The method further includes identifying one or more cycles correlating to a distinct on state immediately followed by a distinct off state, or a distinct off state immediately followed by a distinct on state, and, upon identifying a threshold number adjacent cycles, classifying the traffic signal element as a flashing light.

This relates generally to a method and a system for spatio-temporal polarization video analysis. The spatio-temporal polarization data is analyzed for a computer vision application such as object detection, image classification, image captioning, image reconstruction or image inpainting, face recognition and action recognition. Numerous classical and deep learning methods have been applied on polarimetric data for polarimetric imaging analysis, however, the available pre-trained models may not be directly suitable on polarization data, as polarimetric data is more complex. Further compared to analysis of the polarimetric images, a significant number of actions can be detected by polarimetric videos, hence analyzing polarimetric videos is more efficient. The disclosure is a spatio-temporal analysis of polarization video. The disclosed techniques include configuring a set of parameters from the polarization video to train a spatio-temporal deep network architecture for analyzing polarimetric videos for computer vision applications.