40results about How to "Improve tracking robustness" patented technology

The invention provides a multi-target tracking method integrating obvious characteristics and block division templates. A target motion area is detected by adoption of RGB component background difference and an iterative threshold, and the adaptive ability of a motion detection algorithm to scene illumination change is improved. Based on target area block division, a motion pixel color saliency weighted block centroid model, block centroid shifting fusion and a scale updating method, the calculation efficiency is high, the resistance to partial occlusion is high, and the similar color scene jamming ability is strong. The problem of multi-target measuring-tracking distribution is solved by adoption of two-level data association, and an occluded local area can be accurately positioned. Therefore, adaptive template updating is guided by an occlusion matrix, a reliable global centroid transfer vector is obtained by making use of effective colors and motion information of blocks, and finally, continuous, stable and fast multi-target tracking in complex scenes is realized. The multi-target tracking method integrating obvious characteristics and block division templates is applied to fields like intelligent video surveillance, in-air multi-target tracking and attacking, and multi-task tracking intelligent robots.

The invention provides a target tracking method based on a residual dense twin network. The target tracking method comprises the following steps of: firstly, extracting a template image of a to-be-tracked target from a first frame image of a video, inputting the template image into a residual dense network to obtain initial template features, further inputting the extracted features into a globalattention module to obtain template features, and completing tracker initialization; secondly, cutting the t-th frame of image to extract a search region image, and inputting the search region image into the residual dense network to obtain search region features; and finally, inputting the template features and the search region features into a candidate region generation network to obtain a foreground and background classification confidence coefficient and a bounding box regression estimation value, and further acquiring a t-th frame tracking result. By applying the target tracking method and the target tracking device, the problem that an existing target tracking method based on the twin network cannot effectively process background disorder and similar semantic interference is solved,and the problems that an existing target tracking method based on the twin network is low in tracking accuracy and poor in robustness are further solved.

The invention discloses a global positioning receiver tracking system and a method thereof, wherein the method comprises: estimating the Doppler shift of a target satellite, updating step length by shifting a plurality of frequencies centered on the estimated Doppler shift to acquire a plurality of local carrier wave Doppler, updating a plurality of identical tracking sub-channels, enabling each of the tracking sub-channels to track independently, detecting the signal intensity of each channel and taking the tracking Doppler of the tracking channel with the highest signal intensity as the Doppler shift of the target satellite to estimate, updating each of the tracking sun-channels again based on the result and realizing continual tracking. On the premise of acquiring a more accurate frequency initial estimation, the multi-channel combined tracking method can greatly improve the tracking sensitivity and tracking robustness without the assistance of other information and can acquire better performance by combining with the proven designed various carrier wave tracking loops. Moreover, the system has the advantages of simple structure and easy realization and transplantation.

The invention relates to a contour tracing method based on shape-transmitting united division and image-matching correction. The contour tracing method comprises the steps of in a shape-transmitting united division part, combining a shape prediction image with original image and video frames, by virtue of relocation of position coordinates, providing a new image model construction method, constructing a new energy optimization function to carry out united division on an image model, in an image-matching correction part, carrying out image matching on a division result of a current image with a division result of a previous frame image, when the result of the image matching can not meet a limiting condition, regulating weighting parameters of an energy function in the shape-transmitting united division, and correcting inaccuracy of the division result. According to the contour tracing method based on the shape-transmitting united division and the image-matching correction, an accuracy rate of contour tracing of a target object in a video can be effectively improved.

The invention discloses a TLD and KCF fused video target tracking method. The method comprises the following steps that the position and size of a target area in an initial target frame are determined, and the initial frame is input to a TLD algorithm module and a KCF algorithm module; the TLD algorithm module and the KCF algorithm module run in parallel, and if only one of the TLD algorithm module and the KCF algorithm module processing a present frame includes tracking object output, the output serves as a tracking result of the present frame; and if both modules include tracking object output, the similarities St and Sk between the tracking object output and a target model M are calculated, and a maximum between the St and Sk is selected as a target tracking result; and the method is used to process a next frame of video till tracking of the video frames ends. According to the method, the disadvantage when the TLD and KCF algorithms are used independently can be overcome, the methodis widely adapted to target tracking of a complex video scene, and the instantaneity of the target tracking algorithms is maintained.

The present invention discloses a particle filter infrared tracking method with the fusion of a gradient feature and an adaptive template. The method comprises a step of taking a particle filter algorithm as the body frame of an infrared image tracking algorithm, a step of fusing the gradient histogram describing the gradient feature as a target observation model on the basis of the gradient feature, a step of fusing an adaptive template updating strategy in a tracking algorithm to update a target feature template, and a step of carrying out parallelization on a tracking program by using a sharing storage parallel programming library to improve the algorithm real-time performance. According to the method, the long-time stable and accurate tracking of an infrared target can be ensured, the operational efficiency is high, and a theoretical support is provided for infrared-guided engineering.

The invention belongs to the data processing technology field and discloses a vision tracking method based on a consistency predictor model. The method comprises steps that firstly, a two-input convolutional neural network model is constructed, a video frame sampling region and high-level characteristics of a target template are extracted synchronously, and the logistic regression method is utilized to distinguish the target from the background region; secondly, the convolutional neural network is embedded in the consensus predictor framework, the algorithm randomness test is utilized to evaluate reliability of the classification result, under the specified risk level, the classification result with the credibility index is outputted in the domain form; and lastly, the high-confidence region is selected as a candidate target region, through optimizing the space-time domain global energy function, the target trajectory is acquired. The method is advantaged in that the method can be adapted to complex situations such as target blocking, appearance change and background interference, and the method has strong robustness and accuracy compared with presently popular tracking algorithms.

The invention provides a vehicle tracking method and system based on FCOS, and belongs to the technical field of vehicle tracking. According to the method, a vehicle in a video is detected through anFCOS model; deep learning features and edge features are fused to serve as feature description of the vehicle; by comparing vehicle features, vehicles with the highest similarity of adjacent frame features are matched, small tracks corresponding to the vehicles are generated in a fixed time window, the similarity between the small tracks is measured through a convolutional neural network, and thetracks with the highest similarity are connected, a complete track is obtained, and the whole tracking process is completed. According to the method, the accuracy of vehicle detection can be effectively improved, the influence caused by shielding, camera movement and other factors is reduced, and the vehicle tracking accuracy is improved.

The invention provides a fractional differential-based multi-feature combined sparse representation tracking method. The method includes the following steps: in a frame of particle filtering, first, performing partitioning processing on a target image region, dividing the target region into 9 related and unequal subblocks according to the features of the target region, extracting the gray scale feature and HOG feature of each subblock, combining the two features to perform sparse representation on a target subblock, and also performing the same feature extraction and sparse representation on 8 adjacent regions around the target; then, adopting a nucleating accelerated neighbor gradient algorithm to jointly solve sparse coefficients of 9 candidate particles; and finally, regarding target blocks in different positions as different categories, utilizing a block of the same category as a candidate particle block and a representation coefficient in a dictionary to reconstruct the block, and building a likelihood function according to a reconstruction error to determine an optimal candidate particle, thereby realizing accurate tracking of a main target and 8 auxiliary targets.

The invention discloses a dynamic human face tracking method. The method comprises the steps of S1, tracking a target by using a feature-based tracking algorithm, and when the tracking of the target is lost, going to and executing the step S2; S2, predicting a human face region position after the tracking is lost, establishing a feature point template, matching pixel points of the predicted humanface region position with the current feature point template, initializing an optical flow tracking algorithm by successfully matched feature points, tracking the target, and according to a loss stateof the feature points, updating the feature point template; and S3, when a target position which can be tracked by the feature-based tracking algorithm is restored, returning to and executing the step S1. The method can be suitable for human face tracking under complex conditions of deflection, shielding and the like, and has the advantages of simple implementation method, high tracking efficiency and precision, capability of taking stability and timeliness into account, high environmental adaptability, good anti-interference capability and the like.

The invention provides a multiwindow-based target tracking method, belonging to the computer vision technology. The method comprises the following steps: 1) reading in the image of the initial position of the target, constructing N-numbered tracking windows according to the target characteristics and measuring the compensation di between each tracker and the initial position of the target X0=(x0,y0); 2) reading in the next frame image and computing the new position Yi of each tracker on the current image frame; 3) computing the weight Wi of all the trackers; 4) compensating the new positionsYi of all the trackers on the current image frame by the compensation di and obtaining the value shown in the specification; 5) computing the target position parameter E(X) with the weight Wi of the trackers and the value shown in the specification; and 6) initializing all the trackers according to the target position parameter, updating the reference windows of the trackers and continuously reading in the next frame image, wherein each window corresponds to one tracker. The tracking method can improve the tracking performance of the system under the complex conditions of shading, lighting variation, etc and has strong robustness, and the target tracking result is free from control by the result of a certain tracking window.

The invention relates to the technical field of computers, in particular to a target tracking method, device and equipment based on a mutual supervision twin network. The method comprises the following steps: acquiring a first twinning network similarity response graph in a twinning A network; obtaining a 90-degree rotation twinning network similarity response graph in the twinning B network, andthen performing reverse 90-degree rotation to obtain a second twinning network similarity response graph; performing network training on the obtained first and second twin network similarity responsegraphs to obtain an optimal network model; obtaining a fusion response graph from the first twin network similarity response graph and the second twin network similarity response graph through a meanvalue fusion method; according to the method, more visual information can be better fused from multiple perspectives of a homologous image, the problem of rotation invariance of a convolutional neuralnetwork can be effectively solved, the tracking robustness of a tracker in target rotation is improved, and the tracking precision of the tracker in target rotation is improved. And meanwhile, the problems of tracking drift and tracking failure caused by tracking error accumulation and tracking target rotation can be solved.

The invention relates to a texture-free three-dimensional object tracking method based on confidence coefficient and feature fusion. The tracking method comprises the following steps: (1) establishinga color model; (2) dividing the pixel points into contour points and region points by using a cluster structure; (3) determining the weight alpha i of the edge item, the weight beta i of the color item and the cluster weight omega i according to the confidence coefficient of the contour point and the confidence coefficient of the region point; (4) solving an optimal pose according to the total energy equation corresponding to all the clusters, and rendering the three-dimensional model of the object to obtain an object area on the current frame image; and (5) repeating the steps until the tracking is finished. According to the method, a clustering structure is used, contour points and regional points are re-unified into one energy function, and the problem of non-uniform sampling points issolved; the confidence coefficients of the edge points and the region points are calculated respectively, the edge points and the region points are normalized automatically, the weight of each energyitem is calculated according to the confidence coefficients, and the problem of non-uniform error measurement of different features is solved.

The invention discloses a multi-pig motion trail extraction and behavior analysis method in a group environment, and the method comprises the following steps: improving a known single-target trackingalgorithm, and proposing an algorithm for simultaneously tracking multiple live pigs in the group environment; according to a tracking result, extracting a central point coordinate of each frame of each pig, and drawing a motion trail diagram and an instantaneous speed diagram of each pig; extracting a region of interest according to a target frame obtained by tracking the pig body, segmenting a target contour of each pig, and judging whether the current morphological characteristic of the pig body is standing or lying; judging the motion behaviors of each pig in each time period by combiningthe motion trail, the motion speed and the contour form of the pig body and combining the motion law of the pig body and expert suggestions, and the motion behaviors mainly comprising pig body standing, lying, slow walking, rapid running and abnormal restlessness. According to the invention, how to monitor behaviors of pigs in a group environment by using a precise and robust tracking algorithm isdisplayed, abnormal movement of the pigs can be detected in real time, and judgment of health conditions of pig bodies is assisted.

The invention discloses a clustering subdomain association-based stable characteristic mining and target tracking method. The method comprises the steps of a) detecting a target motion region in the self-adaptive manner, extracting the histogram peak profile of a V-color component in the target motion region, and acquiring the number of clusters according to the energy of a candidate peak, the energy of a region peak and the energy of residual peaks; b) constructing a sample grayscale matrix for an S component and a V component in the target motion region, and conducting the class-number-adaptive K-means clustering operation; c) marking class-based connected sub-regions and establishing the sub-region template, observation model and incremental model description; d) establishing the relationship between a target template and a current observed model sub-region, mining the characteristics of a template-observation stable sub-region and the change rate of template characteristics; e) fusing the displacements in all stable sub-regions of the template, the center of a target detection region and the track of a previous frame in the weighted manner so as to locate the current track of a target, and updating the target template according to the weighted average increment of stable characteristics and the change rate thereof in the frame-by-frame manner.

The invention discloses a target tracking method based on correlation of space-time-domain edge and color feature. The target tracking method based on correlation of space-time-domain edge and color feature comprises the following steps: (1) selecting a tracked target area; (2) extracting the edge outline of the target and calculating the direction angle of the edge; (3) along the two orthogonal directions of horizontal direction and vertical direction, conducting statistics of edge-color symbiosis character pairs, and building a target edge-color correlation centroid model; (4) selecting the centroids of the edge-color pairs with high confidence coefficient to conduct probability weighting, so as to gain a transfer vector of a target centroids in a current frame; (5) conducting statistics of histograms of target edge distances between adjacent frames, conducting probability weighting of the successfully matched distance change rates between the adjacent frames so as to gain a target dimension scaling parameter. By means of the target tracking method based on correlation of space-time-domain edge and color feature, a target tracking in a crowded scene, a shelter, and a condition that the target dimension changes is achieved, and robustness, accuracy and instantaneity of the tracking are improved. The target tracking method based on correlation of space-time-domain edge and color feature has a wide application prospect in the video image processing field, and can be applied to the fields such as intelligent video monitoring, enterprise production automation and intelligent robot.

The invention discloses an aircraft target tracking method and system under an aerospace background, and a computer program product, relates to the technical field of target tracking, and solves the problems of large prediction error, low tracking precision and track matching failure caused by strong maneuverability of an aircraft target under the aerospace background. Acquiring a detection frame set; outputting a prediction trajectory set based on the IMM model; based on first-stage association of a Hyper-Hungarian trajectory association strategy, a hypergraph model is used for matching a high-resolution detection frame set and a prediction trajectory set, for a successfully matched trajectory, the trajectory is updated, and for a detection frame which is not successfully matched, the trajectory is initialized as a new trajectory; on the basis of second-stage association of the Hyper-Hungarian trajectory association strategy, re-matching the trajectory which is not successfully matched with the low-score detection box set, updating the trajectory for the trajectory which is successfully matched, and marking the trajectory which is not successfully matched as a lost state.

A video target tracking method combining particle filtering and metric learning belongs to the target tracking field, and comprises the following steps: offline training a convolutional neural networkcapable of effectively obtaining high-level abstract features of a target; then, learning a weighted distance metric matrix based on a kernel regression metric learning method to minimize a kernel regression prediction error, and solving an obtained optimization problem by utilizing a gradient descent method so as to obtain a distance metric matrix representing an optimal candidate target; calculating a reconstruction error based on the obtained optimal candidate target prediction value so as to construct a target observation model; finally, introducing an updating strategy combining short-time stable updating and long-term stable updating, achieving effective target tracking based on a particle filter tracking framework. The method has high target tracking precision and good robustness.

The invention discloses a pair of augmented reality glasses based on MicroLED display. An inner surface of each lens is provided with a MicroLED display unit, each MicroLED display unit comprises a transparent substrate and pixel units arranged on the transparent substrate in an array, each pixel unit comprises a driving unit and a MicroLED chip, the driving unit is arranged between the transparent substrate and the MicroLED chips, the MicroLED chips comprise visible light MicroLED chips and infrared light MicroLED chips, light emitted by the visible light MicroLED chips is in a visible light wave band, light emitted by the infrared light MicroLED chips is in an infrared wave band, and top surfaces of the MicroLED chips are transparent light emitting layers. The light transmittance of the MicroLED display units is higher than 65%, the light transmittance of the glasses is prevented from being affected, the visible light MicroLED chips can improve the display brightness, meanwhile, the infrared light MicroLED chips are integrated in the MicroLED display unit array, uniform and structured infrared illumination can be provided for eyeballs, and eyeball tracking robustness is effectively improved.

The invention discloses a particle filter infrared tracking method that combines gradient features and self-adaptive templates. The steps are as follows: first, the particle filter algorithm is used as the main frame of the infrared image tracking algorithm; The gradient histogram of the target is used as the target observation model; again, the adaptive template update strategy is integrated in the tracking algorithm to update the target feature template; finally, in order to improve the real-time performance of the algorithm, the tracking program is parallelized with the help of a shared memory parallel programming library. The invention can ensure long-term stable and precise tracking of infrared targets, has high computing efficiency, and can provide theoretical support for infrared guidance engineering.

The invention discloses a multi-target tracking method and system and a computer storage medium, and the method comprises the steps: inputting a current frame image and a preorder frame image into a twin network, and extracting a corresponding feature map; determining a prediction frame according to the feature map by using a region suggestion network; screening the prediction frames for the first time to obtain matching pairs with successful matching and failed matching; performing secondary screening on matching pairs which are successfully matched and failed by adopting appearance feature matching threshold values which are higher than and lower than the optimal appearance feature matching threshold value respectively; and inputting the target features which fail in matching after the second screening into the re-identification branches to obtain a tracking result. According to the method, the regional suggestion network is introduced to serve as a prediction module, a distance matrix in a data association module is analyzed, a simple adaptive threshold determination method is adopted, and a difference matching strategy is combined, so that the adaptability of a multi-target tracking algorithm to nonlinear, high-speed and other complex and diverse application scenes is enhanced, and the target tracking accuracy is improved. And the tracking robustness of the shielded target is improved.

The invention discloses a stable feature mining and target tracking method based on clustering sub-area association. Residual peak energy to obtain the number of clusters; b) construct the S and V component sample grayscale matrix of the target motion area, and perform class-adaptive K-means clustering; c) label and establish the class-based connected sub-regions Sub-region template, observation model and incremental model description; d) Establish the correlation between the target template and the current observation model sub-region, and mine the "template-observation" stable sub-region feature pair and template feature change rate; e) Weighted fusion template for each stable Sub-area displacement, target detection area center and previous frame trajectory to locate the target current trajectory, and update target template frame by frame according to the weighted average increment of stable features and their rate of change.

The invention provides a multi-target tracking method integrating obvious characteristics and block division templates. A target motion area is detected by adoption of RGB component background difference and an iterative threshold, and the adaptive ability of a motion detection algorithm to scene illumination change is improved. Based on target area block division, a motion pixel color saliency weighted block centroid model, block centroid shifting fusion and a scale updating method, the calculation efficiency is high, the resistance to partial occlusion is high, and the similar color scene jamming ability is strong. The problem of multi-target measuring-tracking distribution is solved by adoption of two-level data association, and an occluded local area can be accurately positioned. Therefore, adaptive template updating is guided by an occlusion matrix, a reliable global centroid transfer vector is obtained by making use of effective colors and motion information of blocks, and finally, continuous, stable and fast multi-target tracking in complex scenes is realized. The multi-target tracking method integrating obvious characteristics and block division templates is applied to fields like intelligent video surveillance, in-air multi-target tracking and attacking, and multi-task tracking intelligent robots.

The application belongs to the technical field of computer vision, and discloses a three-dimensional point cloud registration method based on point-line feature fusion. The three-dimensional point cloud registration method firstly uses SAC-IA to complete static registration of three-dimensional point clouds on the surface of a target object, and establishes an initial motion transformation relationship from the three-dimensional point clouds of the 3D model of the target object to the three-dimensional point clouds on the surface of the target object. Then, in the dynamic tracking stage, point features extracted by ORB are matched with line segment features extracted by LSD-LBD, coarse registration is realized by combining an EPnPL algorithm, and the robustness of pose estimation is improved by eliminating false matching through RANSAC. Finally, in the fine registration stage, point-to-plane ICP is used to further optimize pose estimation. The method does not need to rely on artificial marking, and can complete stable three-dimensional point cloud registration based on natural features, and can be widely applied to complex scenes such as mechanical product maintenance and industrial assembly.

The invention discloses a traffic scene-oriented hyperspectral hybrid expert adapter target tracking method and system, and belongs to the technical field of hyperspectral visual target tracking. The method comprises the following steps: firstly, preprocessing a hyperspectral vehicle tracking video sequence, dividing a hyperspectral channel into multi-stream groups, performing dynamic cutting and patch embedding, and constructing a multi-stream image token sequence; then, a hyperspectral interaction hybrid expert tracking network comprising a multi-stream Transform extraction module, a hybrid expert adapter and a time-space hidden state token evolution mechanism is introduced, and joint modeling is carried out on hyperspectral features of a multi-stream template image and a search area image; further, adaptive enhancement is performed on the multi-dimensional features and background interference is suppressed through a mixed expert module consisting of a spectrum expert, a space expert, a time expert and a visual expert; and finally, target positioning prediction is executed based on the enhanced features, and a target tracking result is output. The method can effectively mine interaction information between hyperspectral bands, realizes robust target tracking in combination with a spatio-temporal context, and is suitable for vehicle target tracking application in a complex traffic scene.

The invention provides a multiwindow-based target tracking method, belonging to the computer vision technology. The method comprises the following steps: 1) reading in the image of the initial position of the target, constructing N-numbered tracking windows according to the target characteristics and measuring the compensation di between each tracker and the initial position of the target X0=(x0, y0); 2) reading in the next frame image and computing the new position Yi of each tracker on the current image frame; 3) computing the weight Wi of all the trackers; 4) compensating the new positions Yi of all the trackers on the current image frame by the compensation di and obtaining the value shown in the specification; 5) computing the target position parameter E(X) with the weight Wi of the trackers and the value shown in the specification; and 6) initializing all the trackers according to the target position parameter, updating the reference windows of the trackers and continuously reading in the next frame image, wherein each window corresponds to one tracker. The tracking method can improve the tracking performance of the system under the complex conditions of shading, lighting variation, etc and has strong robustness, and the target tracking result is free from control by the result of a certain tracking window.

The invention discloses a time-space perception-based correlation filtering target tracking method, which relates to the field of image processing target tracking. This method follows the idea of ​​reading the target image, extracting target features, training filter templates, extracting target multi-scale features, and determining the scale and position in the new image of the target to track the target. The structural framework and target features of the correlation filtering target tracking algorithm The extraction method has been optimized and improved, which makes this method have obvious advantages compared with traditional algorithms in terms of tracking robustness and accuracy. This method can enhance the robustness and accuracy of correlation filtering target tracking, solve the model drift problem caused by linear update of the template, improve the long-term tracking effect of correlation filtering target tracking, and at the same time ensure the real-time performance of target tracking. An important improvement in the.

The invention discloses an online target tracking method combining with feature point matching, and the method comprises the steps: firstly calculating a salient region of a tracked region, thereby obtaining an optimal target tracking region; secondly employing a median flow method for tracking, employing a method combining with the feature point matching for estimating the scale change of a target, and deleting interference points through employing a layer clustering method, and obtaining a final position of the target; finally feeding back the scale information estimated by a tracker to a detector, and increasing the speed of the detector through reducing the size of a detection scale space. The method employs a target tracking method based on TLD, carries out the target tracking through employing the significance of the tracked target, and combines with an optical flow method and the feature point matching method so as to improve the target tracking robustness of a tracking algorithm in different scenes. The method is simple and easy, and is high in implementation speed.

The invention relates to a texture-free three-dimensional object tracking method based on confidence coefficient and feature fusion. The tracking method comprises the following steps: (1) establishinga color model; (2) dividing the pixel points into contour points and region points by using a cluster structure; (3) determining the weight alpha i of the edge item, the weight beta i of the color item and the cluster weight omega i according to the confidence coefficient of the contour point and the confidence coefficient of the region point; (4) solving an optimal pose according to the total energy equation corresponding to all the clusters, and rendering the three-dimensional model of the object to obtain an object area on the current frame image; and (5) repeating the steps until the tracking is finished. According to the method, a clustering structure is used, contour points and regional points are re-unified into one energy function, and the problem of non-uniform sampling points issolved; the confidence coefficients of the edge points and the region points are calculated respectively, the edge points and the region points are normalized automatically, the weight of each energyitem is calculated according to the confidence coefficients, and the problem of non-uniform error measurement of different features is solved.