41results about How to "Avoid Tracking Failures" patented technology

The invention discloses an unmanned aerial vehicle autonomous navigation landing visual target tracking method. The method comprises the steps that: firstly, movement amplitude of a target between two consecutive frame images is divided into a plurality of resolution levels according to an order of 'coarse-to-fine'; prior movement simulation of different levels through an off-line training mode and calculation of corresponding prior error Jacobian matrixes are carried out; and as computing of each level of Jacobian matrix is combined with prior knowledge of training, an algorithm in an iterative search target process can be guaranteed to skip from a local optimum effectively to avoid tracking failure. The target is described by using sparse features of a template image target area, namely a gray value of a FAST corner point portion. Compared with a traditional Lucas-Kanade algorithm, complexity of the algorithm provided by the invention is reduced greatly, wherein dense expression targets of all pixels of the target area are often used in the traditional algorithm.

The invention particularly discloses a contour gradient-based target tracking method. The method comprises the steps of segmenting candidate targets in an initial frame of an input video or image sequence, and extracting a reference image; extracting a contour of the reference image to serve as a standard feature template, and performing scale and angle conversion to obtain a multi-scale and multi-angle template sequence; from the next frame of the input video or image sequence, extracting gradient features of a target image; performing sliding window scanning based on a step length of 1 for the multi-scale and multi-angle template sequence on a feature image of the target image, and calculating the similarity of matching between the multi-scale and multi-angle template sequence and the feature image; according to information such as a matching position, a scale factor, an angle factor and the like, segmenting out a sub-region where a target is located in the target image to serve as areference image for the next detection to perform update; and repeating the steps until all the frames of the video or image sequence are detected. The invention aims to provide the contour gradient-based target tracking method for adapting to scale change and / or angle change of the target and improving target tracking accuracy.

The embodiment of the invention discloses a pose determining method and device and a storage medium, and belongs to the technical field of computers. The method comprises: based on the homography matrix of a marker image relative to a specified coordinate system, tracking the feature points of the marker image and acquiring the homography matrix of a specified image relative to the specified coordinate system; and decomposing the homography matrixes according to a constraint condition and a depth hypothesis condition to obtain a rotational displacement matrix of the specified image relative tothe marker image; and acquiring the pose parameters of the specified image relative to the marker image from the rotational displacement matrix. When the homography matrixes are decomposed, the influence of the depths of the feature points on a tracking process is taken into consideration, and the pose parameters of the specified image relative to the marker image are obtained under the same depth of the feature points on the horizontal plane of the specified coordinate system, thereby determining the position and the attitude of a camera and improving the tracking accuracy of the camera. Noerror is introduced even when switching the marker images.

The invention relates to a long-time sheltering robust tracking method based on convolutional features and global search detection. Through adopting a convolutional feature and multi-scale relative filtering method in a tracking module, the feature representation capacity of a tracking object appearance model is enhanced, so that a tracking result is very robust to factors such as illumination variation, object scale variation and object rotation. In addition, through the introduced global search detection mechanism, a detection module can detect the object again when the tracking fails because the object is sheltered for a long time, thus the tracking module can recover from error, and the object can be continuously tracked for a long time even in the case of object appearance changes.

The invention provides a face tracking method based on a multiple appearance models. The method comprises the steps of face tracker start-up, tracking detection, multiple appearance model building, appearance similarity score calculation and face location based on weighted fractional level fusion. During the process, a variety of appearance models are used for matching in different tracking scenarios; a long-term facial appearance feature update relationship diagram is used for learning; and sub-images of the same structure are used for face location. The invention provides a face tracker which uses multiple appearance models for face tracking, wherein a face detector assists the face tracker in face location during short-term appearance changes and reinitialization. A face position with the highest fusion score is selected as a candidate, and the final tracking output is acquired. The problems of obstacle obstruction, scale change and the like in the tracking process are effectively solved. The face tracking method makes further contributions to new design in engineering field and innovative solutions in monitoring field.

The invention discloses a deep learning hand portion detection method based on hand portion area prediction. The hand portion is divided into a left hand, a right hand and overlapped hands. The deep learning hand detection method based on hand area prediction includes the steps: training a deep convolutional network, using the trained network to detect the hand portion class and the area in the first frame of the video stream in the complicated background; according to the correlation, between the hand portion and the time and space, generated from movement inertia of the hand portion, using atracking algorithm to predict the hand portion area in the second frame, and combining with the adjacent frame difference method, acquiring the area shielded by the hand portion and a newly appearedhand portion area; constructing a mask by means of the areas obtained through the tracking algorithm and the adjacent frame difference method, increasing the interesting part in the image, and forminga frame image added with attention; inputting the image into the trained deep convolutional network to detect, so as to obtain the accurate hand portion class and area; and until the last frame, applying the detection method being identical to the detection method of the second frame, thus realizing video stream hand portion detection in the complicated background.

The invention discloses a target tracking method and a device based on targeted candidate areas, wherein the target tracking method comprises: determining the current frame image containing a tracked target; obtaining the tracked target area from the current frame image; obtaining the frame image following the current frame; obtaining a plurality of target candidate areas from the following frame image; calculating the similarities between the tracked target area and each target candidate area; based on the similarities, determining the target tracking area from the multiple target candidate areas. According to the invention, through the target candidate manner, the specific position of a tracked target can be determined; and in the tracking process, the tracked target object can be detected accurately, therefore, effectively increasing the stability of tracking a target object and avoiding the tracking failure.

The invention relates to a multi-camera combined initiative object tracking method for teleoperation of a mechanical arm, which belongs to the field of the teleoperation of the mechanical arm. The method comprises the steps of installing a plurality of cameras in different angles on the basis of the teleoperation, calibrating each camera, realizing the combined initiative tracking by adopting the matching of particle filter and sift local characteristics, enabling a to-be-tracked target to always stay in the center of a view field, and monitoring each angle of the mechanical arm, so that the tracking failure caused by factors such as the shielding can be avoided, and the tracking robustness can be improved by adding the human-machine interaction. According to the method, in the teleoperation process, a new target is marked in a human-machine interaction way, so that the target always stays in an observation area of an operator, or the target is updated, so that a subsequent control strategy can be made by the operator.

The invention provides a real-time detection tracking frame and tracking method based on compressed sensing feature selection. According to the tracking frame and tracking method, compressed features can be selected, and only the sample characteristic with a high distinction degree is used for classification. Real-time tracking can be achieved through the frame and method, the tracking failure phenomenon caused by selection of wrong features is avoided, the influence of bad features on the tracking result is effectively restrained, the tracking speed is obviously increased, and the tracking precision is obviously improved.

The invention discloses a target tracking method based on a conditional adversarial generation twinning network. The method comprises the following steps: (1) acquiring real-time video data for data preprocessing; (2) inputting the data preprocessed in the step (1) into a conditional adversarial generation twin tracking network consisting of a conditional adversarial generation deblurring network module and a full-convolution twin tracking network; and (3) embedding the conditional adversarial generation deblurring network module subjected to adversarial optimization into the full-convolution twinning tracking network framework subjected to feedback updating, and performing target tracking on the preprocessed real-time video data by using a conditional adversarial generation twinning tracking network subjected to online combination to obtain a target position. The dynamic adjustment capability of the tracking network is enhanced, the robustness of a tracker is improved, the tracking network is subjected to separation training in a transfer learning mode, the over-fitting phenomenon is prevented, the training time is shortened, and the generalization capability of the tracking network is improved.

The invention relates to a robot positioning and mapping method and device based on point-line feature fusion, a computer device and a storage medium. The method comprises the following steps; using an RGB-D camera for carrying out surrounding environment detection; acquiring an RGB image and a depth image; continuous image frames are determined; extracting point features from continuous image frames; using a self-adaptive line segment detection method; extracting line features from the continuous image frames; using a feature matching algorithm and a screening mechanism; performing feature matching on the line features; obtaining an inter-frame initial pose; and finally, carrying out minimum error processing on the inter-frame initial pose by adopting a dot-line error model, to obtain theinter-frame pose and the map point, and generating the local map based on the inter-frame pose and the map point, so that the tracking robustness is improved through point-line feature fusion. The problem of tracking failure in scenes such as low texture and motion blur is avoided, and the positioning and mapping accuracy of the robot is effectively improved.

The invention discloses a model drift suppression method based on similarity measurement and a device thereof. The method comprises the following steps: step S100: obtaining parameters in a correlation filter tracking algorithm according to an initialization frame; Step S200: sampling the target position and the target scale in the initialization frame, extracting the initial target feature, extracting the feature of the frame selected target in the i-th frame image after the tracking time, calculating the similarity between the frame selected target and the initial target, and judging the similarity; Step S300: If the target set is updated, the training model is updated according to the update threshold beta. The method sets a fixed number of target sets, and determines whether to updatethe target set and model after similarity measurement, so as to avoid the model drift phenomenon caused by template updating. Another aspect of the present invention also provides an apparatus for model drift suppression based on similarity measures.

The invention is suitable for the technical field of computers, and provides a target tracking method, device and equipment and a storage medium, and the target tracking method comprises the steps of obtaining a target video frame of a video containing a to-be-tracked target; filtering the target video frame according to the mark information of the to-be-tracked target to obtain confidence coefficient data of the to-be-tracked target at each position in the target video frame; predicting a moving target position of the to-be-tracked target in the target video frame according to the historical position information of the to-be-tracked target in the historical video frame; constructing a Gaussian matrix taking the position of the moving target as the center under the condition that the confidence coefficient data meets the non-occlusion condition, and determining a search center of a to-be-tracked target in the target video frame according to the Gaussian matrix and the confidence coefficient data; and intercepting a search area from the target video frame according to the search center and a preset search area size, and extracting position information of the to-be-tracked target in the search area. According to the invention, the tracking performance can be improved.

The invention discloses a high-precision follow-up system, and belongs to the technical field of automation. According to the system, an image processing module is matched with an infrared detector module to convert an analog video into a digital signal, then the digital signal is sent to the image processing module for preprocessing, then target detection and dynamic tracking are carried out, andfinally the offset is transmitted to a servo controller. The servo controller takes an offset signal as a given value; and a speed instruction is calculated according to the actual position fed backby the absolute value encoder, the deviation value of the speed instruction and the actual rotating speed is calculated through a pulse width modulation power amplifier to generate corresponding PWM waveform output, and then the driving module drives the rotary table double-shaft direct-current torque motor, so that the infrared detector module is aligned to a target, and automatic following of the target is realized. In order to prevent shafting shaking from causing physical damage to the rotary table in a shutdown state, an electromagnetic lock is controlled by a relay to lock the two shafting. The high-precision follow-up system has the beneficial effects that the steady-state position following error is reduced, and the positioning precision and the servo rigidity are improved.

The invention discloses a false track elimination method based on distance differential. The problems of high false alarm probability and high false alarm law of a flight path in a complex environmentare solved. According to the method, track distance differential Doppler speed and trace point Doppler speed consistency statistical discrimination processing are added on the basis of an original classical processing algorithm, so that the false alarm probability is greatly reduced, and effective tracking of a real target is ensured. Compared with the prior art, the method has the advantages that a statistical method instead of real-time judgment is adopted, so that the problem of track tracking failure caused by occasional one-time Doppler velocity jump is avoided; according to the method,the channel number difference between a track Doppler filter and a plot Doppler filter is calculated, so that the track Doppler speed and the plot Doppler speed are prevented from being fuzzy, and therelated correctness is improved; compared with a traditional processing method, the method has the advantages that the false alarm probability of the flight path can be greatly reduced, the number offalse flight paths can be reduced by one order of magnitude through testing, and the effect is obvious.

The invention provides a panoramic video data processing method and device, and the method comprises the steps of determining a reference polar coordinate of a central position of a target object in a spherical polar coordinate system under the condition of receiving a frame selection operation for the target object of which a current video frame is located at the center of a playing plane image; generating a tracking plane image of the next video frame by taking the reference polar coordinate as a center according to the panoramic image of the next video frame in the spherical polar coordinate system; determining an updated polar coordinate of the central position of the target object in the tracking plane image of the next video frame in the spherical polar coordinate system according to the object features of the target object; and taking the updated polar coordinate as a reference polar coordinate, and returning to execute the operation step of generating the plane image of the next video frame by taking the reference polar coordinate as a center according to the panoramic image of the next video frame in the spherical polar coordinate system until a tracking stop condition is reached, thereby obtaining a reference polar coordinate sequence corresponding to the current video frame to the target video frame.

The invention discloses an integrated environment perception and multi-target tracking system. The system comprises an image acquisition system, which is used for acquiring images around a maneuvering target in real time for preprocessing and sending the images to an image splicing projection system and a target identification tracking system; an image splicing projection system, which is used for splicing the received images and performing 3D projection to form a 3D panoramic image and sending the 3D panoramic image to the display control system; a target recognition and tracking system, which is used for recognizing sensitive targets from the received images, carrying out frame selection on the recognized sensitive targets and outputting the selected sensitive targets to the 3D panoramic image in a superposed manner, and is used for carrying out continuous recognition and tracking on the recognized sensitive targets, locking the selected specific targets and carrying out continuous recognition and tracking; and a display control system which is used for displaying the 3D panoramic image and used for selecting a specific target in a touch control mode. The system environment sensing distance can be effectively improved, the visual field of maneuvering target passengers is expanded, and the capacity of the passengers for recognizing the field complex environment is comprehensively improved.

According to the semi-offline deep target tracking method based on deep learning provided by the invention, when the semi-offline deep target tracking method is used for target tracking, the problem of target tracking failure caused by false samples in online updating can be avoided, and the target tracking recognition performance can be improved. The method comprises the following steps: S1, constructing a tracker network model, and constructing a structure of a shared layer of the tracker network model based on an MDNet, wherein the specific domain layer is included and comprises a Dropout layer, a full connection layer and a classification function which are connected in sequence, and the feature map output by the last full connection layer in the sharing layer is input into a specificdomain layer after being processed by an activation function; s2, selecting a training set to obtain a trained tracker network model; and S3, obtaining a to-be-tracked feature sequence from the video,inputting the to-be-tracked feature sequence into the trained tracker network model, carrying out subsequent target tracking operation, training the specific domain layer through a sample collected on a first frame in the to-be-tracked feature sequence before the target tracking operation is started, and defining the network weight of the specific domain layer.

The invention belongs to the field of computer vision, and particularly relates to a pedestrian tracking method based on a twin neural network, which comprises the following steps: inputting a video;performing pedestrian marking; acquiring a pedestrian space-time group; establishing and training a twin neural network, and storing the trained twin neural network; obtaining a pedestrian short track; and obtaining pedestrian long trajectories. By adopting the pedestrian tracking method to track pedestrians, the accuracy of pedestrian tracking is effectively improved.

The invention relates to a low-voltage ride-through control method and apparatus. A total current instruction value is generated by summation of a current instruction value which is obtained by difference between a high-voltage side voltage reference value and a high-voltage side actual positive sequence voltage of a power grid, and a current instruction value which is obtained by difference between an SVG side voltage reference value and an SVG side actual positive sequence voltage; then a difference value between the total current instruction value and the actual positive sequence current is obtained; and PWM modulation wave is generated through a regulator. When the power grid falls off, a control strategy in combination with high-voltage side voltage control and SVG side voltage control is rapidly switched, so as to realize rapid reactive compensation; and therefore, it is ensured that transient state impact current is low at the falling-off moment of the voltage, and SVG side voltage tracking failure also can be avoided.

The invention provides a panoramic video data processing method and device. The method comprises the following steps: generating a tracking plane image of a next video frame according to a panoramic image of the next video frame in a spherical polar coordinate system with a reference polar coordinate of a central position of a playing plane image of a current video frame as a center; determining an updated polar coordinate of the central position of a target object in the tracking plane image of the next video frame in the spherical polar coordinate system according to object features of the target object; and taking the updated polar coordinate as the reference polar coordinate, and returning to execute an operation step of generating a tracking plane image of the next video frame by taking the reference polar coordinate as a center according to the panoramic image of the next video frame in the spherical polar coordinate system until tracking stop conditions are reached, and obtaining a reference polar coordinate sequence corresponding to the current video frame to the target video frame. The playing experience is improved.

The invention relates to a long-time sheltering robust tracking method based on convolutional features and global search detection. Through adopting a convolutional feature and multi-scale relative filtering method in a tracking module, the feature representation capacity of a tracking object appearance model is enhanced, so that a tracking result is very robust to factors such as illumination variation, object scale variation and object rotation. In addition, through the introduced global search detection mechanism, a detection module can detect the object again when the tracking fails because the object is sheltered for a long time, thus the tracking module can recover from error, and the object can be continuously tracked for a long time even in the case of object appearance changes.

The invention provides a method and device for tracking the head part of strip steel. The method comprises the steps: if it is determined that a current identity value corresponding to a shearing mark position of a flying shear is a valid identity value, acquiring first actual distance and first tracking distance of operation of the strip steel; if it is determined that a first difference value between the first actual distance and the first tracking distance meets a preset distance interval, generating a first strip steel head part tracking signal; when it is determined that a detection signal of a metal detector changes from an invalid identity value to the valid identity value, acquiring second actual distance and second tracking distance of the operation of the strip steel; if it is determined that a second difference value between the second actual distance and the second tracking distance does not meet the preset distance interval, giving up generation of a second strip steel head part tracking signal; and tracking the head part of the strip steel according to the first strip steel head part tracking signal. In such a manner, even if the head part of the strip steel cannot be tracked through the detection signal, the head part of the strip steel is triggered and tracked according to the second strip steel head part tracking signal, and the head part of the strip steel is accurately identified, so that a steel-heaping accident is avoided, and the production efficiency is guaranteed.

The invention discloses a WSN robot-based indoor tracking system. The system comprises a tracking system, wherein the tracking system comprises a power supply module, a video collection module, a positioning module, a tracking module, a control module and a video management module; the power supply module is respectively connected with the video collection module and the positioning module; the power supply module is connected with a lithium battery; the lithium battery is arranged outside the tracking system; the other end of the lithium battery is connected with a charge circuit; the chargecircuit is arranged in the tracking system; the charging circuit is connected with the power supply module; the video collection module consists of a camera, a camera collection chip and a video coding chip which are connected in sequence; the positioning module comprises an infrared sensor and a positioning device; and the tracking device comprises a SIM card device and a motion tracking device.The WSN robot-based indoor tracking system can be used for tracking a target according to target positioning and can solve the problems of poor indoor positioning precision, small positioning range and the like.

The invention provides a real-time detection tracking frame and tracking method based on compressed sensing feature selection. According to the tracking frame and tracking method, compressed features can be selected, and only the sample characteristic with a high distinction degree is used for classification. Real-time tracking can be achieved through the frame and method, the tracking failure phenomenon caused by selection of wrong features is avoided, the influence of bad features on the tracking result is effectively restrained, the tracking speed is obviously increased, and the tracking precision is obviously improved.

A carrier tracking system is applied to a predetermined space. Carriers with optical recognizable marks are operating in the predetermined space, in which each of the optical recognizable mark is associated with information of materials carried by the carrier. The carrier tracking system includes at least one optical capturing device and a computing unit. The at least one optical capturing device is configured to capture images of the predetermined space. The computing unit is configured to determine an optical recognized coordinate of the optical recognizable mark in the images. The computing unit converts the optical recognized coordinate to a real-space coordinate in the predetermined space. In order to locate the carrier, the computing unit further associates the real-space coordinate with the information of materials carried by the carrier.

A carrier tracking system is applied to a predetermined space. Carriers with optical recognizable marks are operating in the predetermined space, in which each of the optical recognizable mark is associated with information of materials carried by the carrier. The carrier tracking system includes at least one optical capturing device and a computing unit. The at least one optical capturing device is configured to capture images of the predetermined space. The computing unit is configured to determine an optical recognized coordinate of the optical recognizable mark in the images. The computing unit converts the optical recognized coordinate to a real-space coordinate in the predetermined space. In order to locate the carrier, the computing unit further associates the real-space coordinate with the information of materials carried by the carrier.

The invention discloses a scanning method for searching a maximal power point of a solar energy photovoltaic panel, which comprises the following steps of: (1) determining a power region needing scanning according to a solar energy photovoltaic power and voltage characteristic curve and determining a voltage range (A-B) needing scanning according to the power region; (2) determining two golden partition points (X1and X2) in the voltage range (A-B) needing scanning and calculating practical powers corresponding to the two golden partition points; (3) determining the voltage range of next scanning; (4) determining a new golden partition point in anew scanning voltage range, calculating the power corresponding to the new golden partition point and turning to the step (3), and repeating such the process till the maximal power point is found out; and (5) after the maximal power point is found out, tracing the maximal power point. The method can find out the maximal power point in short time and reduce energy loss.

The invention provides a target tracking method of a vehicle-mounted radar, the vehicle-mounted radar and a storage medium. The method comprises the following steps: acquiring measurement of a target detected by a vehicle radar under a vehicle coordinate system at the current moment; if the vehicle is in the turning state at the current moment, the measurement of the target under the vehicle coordinate system at the current moment is converted into a world coordinate system, the world coordinate system is the vehicle coordinate system at the first moment, and the first moment is the moment when the vehicle enters the turning state; then measuring and calculating a state estimation quantity of the target at the current moment under the world coordinate system according to the quantity of the target at the current moment under the world coordinate system; and converting the state estimator of the target in the world coordinate system at the current moment into a vehicle coordinate system at the current moment. Through the process, tracking failure caused by the fact that the relative motion of the radar and the target does not conform to the linear relation when the vehicle turns can be avoided, and therefore the target tracking accuracy of the vehicle-mounted radar is improved.