17 results about "Motion extraction" patented technology

The invention discloses an intelligent ankle rehabilitation exercise management system based on multi-modal data fusion, which belongs to the field of medical assistance and comprises an acquisition processing module, a feature compensation module, an action extraction module, a structure modeling module, a judgment and detection module, an error correction guide module, a management scheduling module, a decision execution module and a report generation module. The motion state of each patient can be comprehensively reflected, fine-grained detection of abnormal actions is achieved, the patient is guided to correct the actions through multi-modal feedback, the secondary injury risk caused by wrong action accumulation is reduced, and the training effect and the controllability of rehabilitation progress are improved; the training plan can be continuously adjusted along with the rehabilitation progress of the patient, dynamic evolution is achieved, meanwhile, insufficient training or potential risks can be rapidly found, doctors and the patient are helped to clearly know the rehabilitation progress, the rehabilitation scheme is continuously optimized, and the training process is more efficient.

A door control apparatus includes a controller that controls the opening and closing of a door provided at an opening of a vehicle, based on imaging data of a predetermined imaging range of an external area relative to the opening. The controller executes a process to extract a target of motion as at least one mobile object moving toward the opening by detecting the motion from the imaging data in a time series in a predetermined area included in the imaging range, and a process to vary the opening and closing timing of the door based on the extracted at least one mobile object. In first and second areas contained in the predetermined area, the controller extracts the target as the mobile object when the direction of the motion is included within predetermined first and second ranges, respectively.

The invention discloses a digital human video generation method and device, equipment, a storage medium and a product, and relates to the technical field of artificial intelligence. The method comprises the following steps: inputting an original face image into an appearance and motion extractor of a target video generation model to obtain an appearance feature map and a face motion parameter; inputting the original audio into a facial expression generator of the target video generation model to obtain facial expression parameters; inputting the original audio into a head posture generator of a target video generation model to obtain head posture parameters; and inputting the appearance feature map, the facial motion parameter, the facial expression parameter and the head posture parameter into a video generation network of the target video generation model to obtain a digital human video generated by the video generation network. Decoupling is carried out through the determination process of the facial expression parameter and the head posture parameter, and the synchronism of the facial expression and the audio content in the generated video is improved.

The application discloses a long-time-delay high-resolution remote sensing image ocean vortex motion extraction method and system, relates to the synthetic aperture radar signal processing field and the ocean remote sensing application field, and the method comprises the following steps: three high-resolution time series remote sensing image acquisition and pretreatment; vortex velocity fields are extracted from the first remote sensing image and the second remote sensing image; the prior velocity field between the second remote sensing image and the third remote sensing image is extracted from the vortex velocity field; feature extraction is carried out on the second remote sensing image and the third remote sensing image; parameters are set according to the prior velocity field, and the vortex velocity field is extracted by tracking the features in the second remote sensing image and the third remote sensing image. The application makes full use of the correlation information between time series remote sensing images, improves the time resolution of vortex motion extraction, and lays a foundation for long-time-series ocean vortex space-time evolution information extraction.

The invention discloses an internet platform for vending machine management and control, and relates to the technical field of computer information, and the internet platform comprises a time anchoring module, a frame slot buffer module, a frame segment reconstruction module, an event triggering module, a rhythm stabilization module and a linkage defense module, and an independent time nail is calibrated for each frame of image, a continuous frame sequence timeline is generated, and a time sequence reference for picture repair is provided. According to the method, video time sequence continuity is kept in a network fluctuation environment, action recognition integrity is guaranteed through time anchoring, frame gap buffering and slippage repairing, high-sensitivity action extraction and door lock illumination linkage are combined, rapid recognition and physical intervention are achieved, the recognition delay and the false report and missing report rate are remarkably reduced, and the safety prevention and control capacity of the vending machine is improved.

The application discloses an internet platform for vending machine management and control, and relates to the field of computer information technology, comprising a time anchoring module, a frame gap buffering module, a frame segment reconstruction module, an event triggering module, a rhythm stabilizing module and a linkage defense module: the time anchoring module establishes a time anchor between a camera device and a network, labels an independent time nail for each frame image, generates a continuous frame sequence timeline, and provides a time sequence reference for picture repair. The application maintains video time sequence continuity under network fluctuation environment, guarantees motion recognition integrity through time anchoring, frame gap buffering and slip repair, and realizes rapid recognition and physical intervention by combining high-sensitivity motion extraction and door lock lighting linkage, thereby significantly reducing recognition delay and false alarm and missing alarm rates, and improving vending machine safety prevention and control capability.

This application discloses a video motion extraction method, storage medium, and electronic device. The method involves identifying the detection region information, body posture features, and initial acquisition device parameters of a target object in a target video. Based on these parameters, body motion is extracted to obtain the target object's body motion information. Hand motion is extracted from the target video, yielding hand motion information and rotation information of the target joints associated with the hand. The body motion information is then updated based on the joint rotation information. The updated body motion information is then physically optimized. Finally, based on the optimized body motion and hand motion information, the full-body motion information of the target object and the corresponding camera pose are obtained. This improves the accuracy and efficiency of motion extraction from videos.

The invention discloses an object grabbing method based on RGB image incremental three-dimensional reconstruction. Comprising the following steps: S0, initializing the motion of a mechanical arm; s1, extracting a key frame of the RGB image; s2, performing depth recovery on the image based on the key frame to obtain a preliminary recovery depth map and a confidence map Ci; s3, performing depth optimization on the preliminarily recovered depth map to obtain an optimized depth map; S4, performing incremental updating on a global reconstruction result by using the optimized depth map to obtain a depth map fused with multi-view information; S5, calculating a normal vector of the depth map to obtain a normal map Ni; s6, the grabbing pose is expressed; s7, estimating the captured key points; s8, estimating the grabbing approaching direction and the width of the clamping jaw; and S9, the mechanical arm executes the grabbing action according to the current motion state and the updated target pose. The mechanical arm has the universal grabbing capacity on various objects, the grabbing success rate is high, reconstruction and grabbing pose optimization are synchronously carried out in the process that the mechanical arm approaches the objects, and extra time consumption is avoided.

The present application relates to a door control device, a vehicle control system, a door control method, and a non-transitory computer-readable medium. The opening and closing of a door of a vehicle are more appropriately controlled. A control section controls the opening and closing of a door provided to an opening section of a vehicle based on captured data of a predetermined captured range with respect to an external region of the opening section, and performs processing of detecting a motion from the captured data in a predetermined region included in the captured range according to a time series, extracting an object of the motion as at least one moving body moving toward the opening section, and changing an opening and closing timing of the door based on the at least one extracted moving body, in a first region included in the predetermined region, the control section extracts the object as the moving body in a case where a direction of the motion is included in a predetermined first range, and in a second region included in the predetermined region, extracts a region where a direction of the motion is included in a predetermined second range as the moving body.

The invention discloses a random point tracking method based on an event camera, and the method comprises the steps: 1, obtaining an initial position of a to-be-tracked target point, and constructing an event representation suitable for a neural network; 2, the space encoder extracts geometric structure features in the event representation; 3, calculating time sequence motion characteristics of the target point from the event data by using a motion extraction module; 4, a variable speed sensing module is adopted, and a feature correlation graph is generated under the guidance of object time sequence motion features; and 5, predicting the coordinates of any point according to the feature correlation graph and the time sequence motion features of the object so as to track the position of any point. The method effectively improves the tracking precision of any point of the event camera in a complex scene, is suitable for the fields of intelligent, virtual reality, man-machine interaction and the like, and has a good actual deployment value.

This invention discloses a cross-modal learning method for motion features in small object detection, applied to the field of small object detection technology, addressing the problem of poor performance in existing small object detection technologies. The method of this invention includes two key parts: motion pattern mining and a motion visual adapter. The former mines motion patterns from a time-varying visual representation space, while the latter is designed to associate motion patterns with visual semantics. Then, their cross-modal interactions are explored to guide the motion extractor to effectively capture the implicit motion modalities. With the cooperation of motion modalities, even using a simple detector, state-of-the-art detection performance can be achieved with almost no increase in the inference cost of small moving object detection.

This invention discloses a method for robustly constructing visual odometry based on deep learning. The visual odometry includes a motion extraction network, an optical flow estimation network, and a pose estimation network. The motion extraction network outputs motion vectors of pixels based on the blurring of the image. The optical flow estimation network uses these pixel motion vectors as prior information to construct a matching cost volume layer, resulting in a more accurate optical flow map. The pose estimation network uses the estimated optical flow map and a confidence information layer as inputs to calculate the six-DOF pose information of the camera. This invention utilizes the motion information contained in the blurred image as prior information, constrains the distribution of the matching cost region, and adds confidence information to the optical flow map, thus achieving robust estimation of camera pose in blurred scenes.

The present application relates to the technical field of display, especially to a self-adaptive display screen quality optimization method and system. The method comprises: collecting ambient light, color temperature, viewing angle and frame data to be displayed, performing space-time synchronization and partition mapping, generating scene mode encoding and effective image data; performing brightness analysis, color statistics, motion extraction and scene recognition to generate content perception metadata; performing strategy mapping and adjustment sequence processing to obtain quality strategy data; generating quality optimization parameter target value according to the strategy data; then generating pixel brightness adjustment signal, partition backlight control signal and display driving signal to obtain image display data; finally collecting actual backlight brightness, calculating error and performing smooth transition and feedback correction to output corrected image display data. The present application realizes self-adaptive partition quality optimization based on content type and viewing angle, and significantly improves image quality and viewing experience under complex display scenes.

The invention belongs to the technical field of motion analysis in computer vision, and discloses a human body local motion extraction method, system and device and a storage medium, and the method comprises the steps: providing a frame image of a video sequence, and generating a human body mask; generating a superpixel region; performing optical flow estimation on the frame image of the video sequence to generate a full-image dense optical flow field; generating a motion salient region mask based on the superpixel region and the full-image dense optical flow field; and obtaining an intersection of the motion salient region mask and a human body mask, and outputting a human body local motion mask. According to the method, the superpixel segmentation is only executed in the human body mask region, so that the calculation overhead of a non-target region is reduced; the super-pixel region is used as a calculation unit, so that the calculation complexity of optical flow aggregation and motion analysis is greatly reduced; constructing a progressive constraint system in combination with the superpixel region, the full-image dense optical flow field and the human body mask; through a self-adaptive threshold strategy, a significant motion area is accurately identified; in a complex background, a moving human body and a static background can be effectively distinguished, and the false detection rate is reduced.

[Problem] To provide an information processing device, an imaging device, and a program that enable compatibility between low power consumption and high detection accuracy. [Solution] An information processing device according to the present disclosure comprises: a motion detection unit that performs motion extraction and motion determination of a subject on the basis of a plurality of first pixel signals generated by a plurality of first pixels; a distance detection unit that performs distance determination to the subject based on a plurality of second pixel signals generated by a plurality of second pixels; and a mode control unit that converts a moving body detection mode in which low-resolution image information is acquired to a feature extraction mode in which medium-resolution image information is acquired or an imaging mode in which high-resolution image information is acquired, based on a result of the motion determination and a result of the distance determination.

A method and system for quantifying movement patterns in a subject includes receiving motion data representing voluntary and involuntary movements of the subject from a sensor device worn or held by the subject. A signal indicative of involuntary movement deviations occurring in a plane diverging from a voluntary movement trajectory is extracted and associated to a current position of the voluntary movement trajectory. Peaks within the extracted signal are identified, wherein each peak corresponds to a local maximum of involuntary movement magnitude. A probability density function (PDF) of the identified peaks is estimated and a symptom severity score is computed based on one or more quantiles of the estimated PDF. Feedback is provided to a user via a user interface based on the symptom severity score to adjust a parameter of a therapeutic device.

The application belongs to the technical field of array signal processing, and particularly relates to a towed array target direction estimation method and system based on multi-dimensional null steering. The application can extract cross-aperture phase drift information according to the towed array movement, perform spatial registration and sequential fusion on non-repeated effective aperture data, construct a multi-dimensional null cancellation discriminant based on the phase cancellation relationship between local aperture projection components, extract target pointing parameters, and obtain target direction results, thereby further improving the direction estimation performance. Meanwhile, the de-coherence algorithm is added to more accurately distinguish multiple targets. The application can accurately estimate the directions of coherent multiple targets in a complex engineering environment with limited towed array size.