50 results about "Computational vision" patented technology

A method of increasing information content for content-based image retrieval (CBIR) systems includes the steps of providing a CBIR database, the database having an index for a plurality of stored digital images using a plurality of feature vectors, the feature vectors corresponding to distinct descriptive characteristics of the images. A visual similarity parameter value is calculated based on a degree of visual similarity between features vectors of an incoming image being considered for entry into the database and feature vectors associated with a most similar of the stored images. Based on said visual similarity parameter value it is determined whether to store or how long to store the feature vectors associated with the incoming image in the database.

A method, apparatus, and software for exercising human eyes with a monitor onto which is projected a plurality of shapes such that portions of the shapes have a contrast changing at a speed less than or equal to approximately 2.0 cycles/sec. The shapes comprise paired shapes of opposite colors (black/white, red/green, or blue/yellow, or combinations thereof), and the speed is preferably less than or equal to approximately 0.8 cycles/sec. Also a method, apparatus, and software projecting a plurality of symbols each comprising a plurality of bars one of which has a length different than that of others in the symbol. A visual efficiency is calculated based upon a number of identical symbols correctly located by a user and a time to locate the identical symbols.

It is described a 3D monocular tracking system, being robust, having low cost, easy to install and use, useful for the upper limbs rehabilitation in a patient in need thereof, as well as a home self-directed therapy method for patients having upper limbs' movement disability. The system comprising a) a handle or gripper; b) a computational vision system comprising a video camera; c) software comprising a set of games; d) a processor; and, e) a display apparatus.

The invention discloses a GPS-fused robot vision inertial navigation integrated positioning method. The method comprises the following steps: extracting and matching feature points of left and right images and front and back images of a binocular camera, and calculating three-dimensional coordinates of the feature points and relative poses of image frames; selecting a key frame in an image stream,creating a sliding window, and adding the key frame into the sliding window; calculating a visual reprojection error, an IMU pre-integration residual error and a zero offset residual error and combining the errors into a joint pose estimation residual error; carrying out nonlinear optimization on the joint pose estimation residual error by using an L-M method to obtain an optimized visual inertial navigation (VIO) robot pose; if the GPS data exist at the current moment, performing adaptive robust Kalman filtering on the GPS position data and the VIO pose estimation data to obtain a final robot pose; and if no GPS data exist, replacing the final pose data with the VIO pose data. According to the method, the positioning precision of the robot is improved, the calculation consumption is reduced, and the demands of large-range and long-time inspection are satisfied.

The invention provides a method and equipment for automatically adjusting the visual characteristic of a page. The method comprises the following steps of: calculating a visual characteristic reference value according to a visual characteristic value of at least one displayed page; calculating a visual characteristic value of the page to be displayed; comparing the visual characteristic value of the page to be displayed with the visual characteristic reference value; and when the compared result is greater than a threshold value, adjusting the visual characteristic of the page to be displayed. The method and the equipment for automatically adjusting the visual characteristic of the page have the advantages of preventing the page to be displayed from making user eyes suddenly impacted and further feeling uncomfortable during page jumping due to over-high page difference of the visual characteristic and the page to be displayed, particularly protecting eyesight of people suffering from vision diseases, reducing to display images with high contrast of the visual characteristic and effectively prolonging the service life of a display.

The invention discloses a road vehicle tracking method, belonging to the technical field of computational vision processing. As that image between frame of the binocular camera is more than that of asingle purpose, the invention proposes a multi-angle optical flow characteristic with strong description ability to replace the original multi-surface optical flow characteristic, effectively solves the technical problems of wrong heel, drift and the like when vehicles with similar appearance and close distance are occluded from each other.

The invention relates to the field of multimedia information (pictures and videos), machine vision and computational vision, in particular to a method for quickly checking a video. The method for quickly checking the video is characterized in that a system automatically distributes parameters to finish lens boundary extraction or similarity filtering, then, extracted boundary frames are combined into a short video, and finally, a checking person or an automatic system checks the short video. The method comprises the following steps that: S1: inputting a checking type, and uniformly inputting the checking type of a video to be checked into a system preprocessing system; S2: extracting a boundary lens, and carrying out similar filtering; S3: carrying out video synthesis, utilizing an open source tool ffmpeg (for example: ffmpeg-fimage2-i%d.jpg-vcodec lix264-rNtest.mp4) command to generate a video of which the frame rate is N frame/s; and S4: checking and outputting the video, and outputting the video of which the frame rate is N frame/s so as to bring convenience for manual checking or system checking.

The invention discloses a 3D road vehicle tracking method. The invention belongs to the technical field of computational vision processing. A 3D spatial feature and 2D image feature of an object are combined with an MDP model. The evaluation function is reconstructed, and a multi-target tracking method based on 2D and 3D joint features is proposed. The original 2D image domain tracking is extendedto 3D spatial domain tracking, which effectively solves the technical problems such as mistracking and drift when vehicles with similar appearance and close distance occlude each other.

The invention discloses a method for solving a polymorphic statement video positioning task through a space-time graph reasoning network, and belongs to the field of natural language visual positioning. According to the method, firstly, a video is parsed into a space-time region graph, and the space-time region graph not only has implicit and explicit space sub-graphs of each frame, but also has across-frame time dynamic sub-graph; next, a text clue is added into the space-time region graph, and multi-step cross-modal graph reasoning is established; the multi-step process may support multi-order relational modeling; and thereafter, a temporal boundary of the pipeline is determined using a temporal locator, then the object is located in each frame using a spatial locator having a dynamic selection method, and a smooth pipeline is generated. According to the method, the video does not need to be trimmed when the natural language is positioned, so that the video positioning cost is reduced; and question sentences and declaration sentences can be effectively processed, technical support is provided for higher-level natural language processing and computational vision combined research(such as video questions and answers), and the application prospect is wide.

The invention discloses a three-dimensional reconstruction precision testing method and device and electronic equipment, and relates to the technical field of high-precision maps. The method comprisesthe steps that three-dimensional coordinates of a first object and a second object in an image sequence in visual point cloud are acquired, the visual point cloud is obtained through three-dimensional reconstruction, and the first object and the second object are objects existing in a high-precision map; calculating a similarity transformation matrix between the visual point cloud and the high-precision map according to the three-dimensional coordinates of the first object in the visual point cloud and the three-dimensional coordinates of the first object in the high-precision map; performingcoordinate transformation on the three-dimensional coordinates of the second object in the visual point cloud according to the similarity transformation matrix to obtain first three-dimensional coordinates of the second object; and calculating the precision of three-dimensional reconstruction according to the error between the first three-dimensional coordinate and the three-dimensional coordinate of the second object in the high-precision map. According to the method, existing high-precision map data are fully utilized, only a small amount of calibration needs to be carried out on the imagesequence, and the three-dimensional reconstruction precision test efficiency can be improved.

The invention provides a robustness control method and system for multi-source fusion localization, and the method comprises the steps: carrying out the visual feature point extraction of the same frame of image data of the same scene, obtaining a first pose through visual inertial navigation fusion, and calculating a visual feature point error and a first re-projection error according to the first pose; carrying out inertial navigation and laser radar feature point extraction on the point cloud data of the same frame in the same scene, and carrying out inertial navigation and laser radar fusion to obtain a second pose; calculating a point cloud feature point error and a second re-projection error according to the second pose; calculating a third pose of the laser radar and the camera, and calculating a camera position feature point error and a third reprojection error according to the third pose; and comparing each re-projection error, carrying out weight ranking on each feature point error, and carrying out quality grade division. Based on the method, the invention further provides a robustness control system. According to the method, the resolving speed of the system can be increased, the environmental adaptability and precision of the system are improved, and the robustness of the system is improved.

The invention discloses a method for objectively evaluating the quality of stereoscopic video without reference based on sparse representation: the stereoscopic video is downsampled to obtain stereoscopic video pair. The left and right viewpoints of stereoscopic video pair are respectively used to obtain the monocular energy magnitude map, and the weighting factors of the left and right views areobtained. Stereoscopic video weighted the two viewpoints to get the visual perception map. Visual perception saliency map is obtained by calculating ROI on visual perception map; dictionary learning.The sparse representation of visual perception saliency graph is used to obtain the coefficient matrix and its entropy. The coefficient matrix is obtained by means of mean, variance and two-norm operation. Choosing the coefficient matrix of video pair and MOS in the video database for training, the training model is obtained. The training model is used to predict the quality of any stereo video, and the final objective prediction value is obtained. The invention makes more comprehensive and accurate objective evaluation of stereoscopic video quality according to the visual perception image andthe sparse representation as a tool.

The invention discloses a visual inertia real-time initialization alignment method and system. The visual inertia real-time initialization alignment method comprises the steps of: acquiring a currentvisual frame, and acquiring a plurality of continuous visual frames of the current visual frame and the measurement data of a corresponding inertia measurement unit; calculating visual frame poses, and performing pre-integration on the measurement data of the inertial measurement unit between two adjacent visual frames to obtain integral data; acquiring an ontology frame pose corresponding to thevisual frame, establishing a linear equation set by utilizing the integral data and the ontology frame pose, and iteratively updating a coefficient matrix and a constant term of the linear equation set; solving the linear equation set to obtain a current ontology frame speed and initial visual frame gravitational acceleration; and if an absolute error between a module value of an initial visual frame gravitational acceleration value and a local gravitational acceleration value is smaller than a preset value, and the value of n is larger than the preset value, judging that visual inertia initial alignment succeeds. The dimension of linear equation set is small, the amount of calculation for solving unknown quantities is smaller, the efficiency is high, and the real-time performance is improved.

The invention discloses an image processing method and device, electronic equipment and a storage medium, belongs to the field of computers, and particularly relates to computing vision, image processing, face recognition and deep learning technologies in artificial intelligence. The implementation scheme comprises the steps of performing skin color recognition on a to-be-processed face image, anddetermining a target skin color of a face in the face image; if the style migration model set does not contain the style migration model corresponding to the target skin color, processing the face image by using any style migration model to obtain a reference transformation image corresponding to the face image, and adjusting a hue value, a saturation value and a brightness value of each pixel point in a target area in the reference image according to the target skin color, and obtaining a target transformation image matched with the target skin color. Therefore, when the target skin color ofthe human face in the to-be-processed human face image does not have the corresponding style migration model, style transformation processing can be carried out on various skin color users by utilizing the existing style migration model, and the application range is expanded.

The invention discloses a multi-modal fused unsupervised pedestrian re-identification rearrangement method. The method comprises the following steps of collecting multi-modal information of pedestrians in a walking process; extracting pedestrian features by using a convolutional neural network model, and calculating visual similarity; constructing image space-time distribution by utilizing the image space-time information; constructing WiFi space-time distribution by utilizing the WiFi information; and the visual similarity, fusing the image space-time distribution and the WiFi space-time distribution, and arranging a pedestrian re-identification sorting result. Multi-modal information is integrated for secondary rearrangement, the method is an effective measure for reducing the search space, and the defect that a traditional pedestrian re-identification method based on visual features is sensitive to the monitoring environment is effectively overcome.

The invention relates to the technical field of artificial intelligence, in particular to a rail-mounted robot anti-jamming detection method and device based on artificial intelligence. The method comprises the following steps of acquiring an ROI area image of a photovoltaic cell panel; obtaining a segmentation result graph of a ROI area by using a semantic segmentation network model; obtaining abinary image of an edge frame of the photovoltaic cell panel from the segmentation result image, and judging a first longitudinal edge frame, a first transverse edge frame and a second transverse edgeframe in the binary image; calculating the maximum gap width between the first transverse edge frame and the second transverse edge frame, and averaging the maximum gap width and the measured gap width to obtain the gap width; calculating a visual fall height, and averaging the visual fall height and the measured fall height to obtain a fall height; and when the gap width or the fall height is larger than a set threshold value, returning the rail-installed cleaning robot. According to the method and device, the problems that the robot is stuck due to too large gaps or too high fall, and the distance measurement of an infrared distance measurement sensor is inaccurate are solved, and the accuracy and reliability of a judgment result are improved.

The invention discloses a pan-tilt visual control system and control method, and the control method comprises the steps: obtaining environment information, and obtaining a visual image; identifying ato-be-detected target from the visual image; calculating three-dimensional coordinates of the to-be-detected target in a camera coordinate system; calculating a visual control angle according to the three-dimensional coordinates of the to-be-detected target in the camera coordinate system; calculating three-dimensional coordinates of the to-be-detected target in a ground coordinate system; calculating a prediction compensation angle according to the three-dimensional coordinates of the to-be-detected target in the ground coordinate system; and according to the visual control angle and the prediction compensation angle, a holder motor is controlled based on a fuzzy control algorithm. According to the technical scheme, when the pan-tilt motor is controlled, the yaw angle prediction compensation value and the pitch angle prediction compensation value of the to-be-detected target are obtained by referring to the three-dimensional coordinates of the to-be-detected target in the ground coordinate system and the gyroscope detection data, so that the response speed of an operation system and the control accuracy are improved.

The invention provides automatic visual sorting equipment applied to crayfish. The automatic visual sorting equipment comprises a feeding device, an automatic weighing device, an automatic recognition device, a product sorting device and a master control device. Crayfish is placed in the feeding device and conveyed to the automatic weighing device through the feeding device, the weight of each single product is automatically recorded and fed back to the master control device, the weight is conveyed to the automatic recognition device through the automatic weighing device, and photographing recognition of the size, color and the like of the product is conducted in the automatic recognition device. And the master control system carries out internal algorithm calculation according to parameters such as weight parameters in the automatic weighing device, the size and color of the automatic identification device and the like, so that the products are automatically sorted in the sorting device according to requirements. The whole device is stable in transmission, all products are automatically sorted according to requirements on the premise that the product quality is not damaged, and the production efficiency is remarkably improved. Under the condition that the product quality is not damaged, the products are automatically sorted according to different parameters such as weight, size and color by utilizing technologies such as computational vision, and the production efficiency of the whole production line is improved.

The invention discloses an image processing method and device and a medium, and relates to the technical field of image processing, and the method comprises the steps: obtaining a plurality of original cabin images at different visual angles; splicing the plurality of original vehicle cabin images to obtain a vehicle cabin panoramic image; determining a target object in the plurality of original cabin images based on computer visual perception; determining a target virtual image corresponding to the target object, and obtaining a target layer containing the target virtual image; and superposing the vehicle cabin panoramic image and the target image layer to obtain a target panoramic image containing the target virtual image. According to the method, the target object in the vehicle cabin image is perceived by using the computational vision technology, and the detected target object is represented by using the target virtual image, so that the perceiving effect of the synthesized target panoramic image on the vehicle cabin can be effectively improved.

The invention relates to the technical field of artificial intelligence, in particular to an autonomous robot anti-jamming prediction method and device based on artificial intelligence. The method comprises the following steps: acquiring an image of the surface of a photovoltaic cell panel; carrying out edge detection on the image by adopting a semantic segmentation network model to obtain edge frame information; adopting a grid key point detection network model to detect grid key points of the photovoltaic cell panel; determining a first grid and a second grid according to the distances fromthe grid key points to the edge lines and the perpendicular bisectors; calculating a visual height difference according to the areas of the first grid and the second grid; and obtaining the height difference through weighted fusion with the infrared height difference obtained through the infrared distance measuring sensor, and when the height difference is larger than a threshold value, stopping the autonomous robot from advancing and giving an alarm. According to the method, the problem that the autonomous robot is stuck due to overlarge height difference and the problem that the distance measurement of the infrared distance measurement sensor is inaccurate due to the anti-reflection characteristic of the photovoltaic cell panel are solved, and the accuracy and reliability of a judgment result are improved.