A system for dense depth computation aided by sparse feature matching generates a first image using a first camera, a second image using a second camera, and a third image using a third camera. The system generates a sparse disparity map using the first image and the third image by (1) identifying a set of feature points within the first image and a set of corresponding feature points within the third image, and (2) identifying feature disparity values based on the set of feature points and the set of corresponding feature points. The system also applies the first image, the second image, and the sparse disparity map as inputs for generating a dense disparity map.

A system for eye-tracking according to an embodiment of the present invention includes a data collection unit that acquires face information of a user and location information of the user from an image captured by a photographing device installed at each of one or more points set within a three-dimensional space and an eye tracking unit that estimates a location of an area gazed at by the user in the three-dimensional space from the face information and the location information, and maps spatial coordinates corresponding to the location of the area to a three-dimensional map corresponding to the three-dimensional space.

A system for eye-tracking according to an embodiment of the present invention includes a data collection unit that acquires face information of a user and location information of the user from an image captured by a photographing device installed at each of one or more points set within a three-dimensional space and an eye tracking unit that estimates a location of an area gazed at by the user in the three-dimensional space from the face information and the location information, and maps spatial coordinates corresponding to the location of the area to a three-dimensional map corresponding to the three-dimensional space.

The present invention relates to a CPR posture monitoring model and system, the model being configured to: based on human skeleton point data extracted from CPR moves, at least compute arm-posture angle data and GC matching angle data related to the CPR moves, and determine whether the CPR moves are qualified by comparing arm-posture angle data and GC matching angle data to a CCP qualification threshold; wherein human skeleton point data are extracted from first move data collected by a first optical component and second move data collected by a second optical component simultaneously, wherein an included angle between collection directions of the two optical components ranges between 30 and 90 degrees. By constructing and applying a CPR posture monitoring model defined by data like GC matching angle, evaluation result has higher accuracy. Compared with the prior art where the equipment positions must be exactly the same as the equipment positions when the data are acquired for the AI algorithm to compute the CPR mistakes, the disclosed method does not require that the equipment positions must be exactly the same as the equipment positions when collecting the data and could tolerate inaccuracy or error in the equipment positions, besides, the disclosed method could realize harmless quality control of CPR moves.

The present invention relates to a CPR posture monitoring model and system, the model being configured to: based on human skeleton point data extracted from CPR moves, at least compute arm-posture angle data and GC matching angle data related to the CPR moves, and determine whether the CPR moves are qualified by comparing arm-posture angle data and GC matching angle data to a CCP qualification threshold; wherein human skeleton point data are extracted from first move data collected by a first optical component and second move data collected by a second optical component simultaneously, wherein an included angle between collection directions of the two optical components ranges between 30 and 90 degrees. By constructing and applying a CPR posture monitoring model defined by data like GC matching angle, evaluation result has higher accuracy. Compared with the prior art where the equipment positions must be exactly the same as the equipment positions when the data are acquired for the AI algorithm to compute the CPR mistakes, the disclosed method does not require that the equipment positions must be exactly the same as the equipment positions when collecting the data and could tolerate inaccuracy or error in the equipment positions, besides, the disclosed method could realize harmless quality control of CPR moves.

Described herein are systems and techniques for imaging inventory objects in an environment. A system can include a cart, a first fixed camera fixedly mounted on the cart at a first angle, a pan-tilt-zoom (PTZ) camera controllably mounted on the cart, a PTZ controller, and a cart controller. The PTZ controller can receive PTZ instructions from the cart controller and send engagement instructions to the PTZ camera. The cart controller can receive, from the first fixed camera, first image data that captures a first inventory object, determine, from the first image data, a spatial location of a first inventory object, generate PTZ instructions to cause the PTZ camera to capture the first inventory object, transmit the PTZ instructions to the PTZ controller, and receive PTZ image data that captures the first inventory object.

Described herein are systems and techniques for imaging inventory objects in an environment. A system can include a cart, a first fixed camera fixedly mounted on the cart at a first angle, a pan-tilt-zoom (PTZ) camera controllably mounted on the cart, a PTZ controller, and a cart controller. The PTZ controller can receive PTZ instructions from the cart controller and send engagement instructions to the PTZ camera. The cart controller can receive, from the first fixed camera, first image data that captures a first inventory object, determine, from the first image data, a spatial location of a first inventory object, generate PTZ instructions to cause the PTZ camera to capture the first inventory object, transmit the PTZ instructions to the PTZ controller, and receive PTZ image data that captures the first inventory object.

Mesh-tracking based dynamic 4D modeling for machine learning deformation training includes: using a volumetric capture system for high-quality 4D scanning, using mesh-tracking to establish temporal correspondences across a 4D scanned human face and full-body mesh sequence, using mesh registration to establish spatial correspondences between a 4D scanned human face and full-body mesh and a 3D CG physical simulator, and training surface deformation as a delta from the physical simulator using machine learning. The deformation for natural animation is able to be predicted and synthesized using the standard MoCAP animation workflow. Machine learning based deformation synthesis and animation using standard MoCAP animation workflow includes using single-view or multi-view 2D videos of MoCAP actors as input, solving 3D model parameters (3D solving) for animation (deformation not included), and given 3D model parameters solved by 3D solving, predicting 4D surface deformation from ML training.

Implementations generally perform pose reconstruction by tracking for video analysis. In some implementations, a method includes obtaining a plurality of videos of at least one subject performing at least one action in an environment. The method further includes tracking the at least one subject across at least two cameras. The method further includes reconstructing a 3-dimensional (3D) model of the at least one subject based on the plurality of videos and the tracking of the at least one subject.

Implementations generally perform pose reconstruction by tracking for video analysis. In some implementations, a method includes obtaining a plurality of videos of at least one subject performing at least one action in an environment. The method further includes tracking the at least one subject across at least two cameras. The method further includes reconstructing a 3-dimensional (3D) model of the at least one subject based on the plurality of videos and the tracking of the at least one subject.