53results about How to "Accurate target tracking" patented technology

The present invention discloses a target tracking method. The method is implemented in a mobile device having the photographing function. The method comprises the steps of according to a user input, determining a target position within an initial frame, wherein the target position represents a target frame around the center of a target; based on the target position within the initial frame, training and generating a tracker and a detector, wherein the tracker is adapted to track the target in a photographed video in the frame-by-frame manner, and the detector is adapted to conduct the frame-by-frame detection for the target in the photographed video; for each subsequent image frame of the photographed video, tracking by the tracker to obtain the target position of the image frame and outputting a tracking response value; judging whether the tracking response value is greater than or equal to a threshold or not; if yes, continuing the target tracking for the next image frame; if not, starting the detector and outputting the target position of the corresponding image frame by the detector; keeping the detector to run continuously for a predetermined number of frames; and switching to the a tracker for continuously conducting the target tracking. The invention also discloses a corresponding mobile device.

The invention discloses a parallax-based three-dimensional trajectory tracking method of fish movement. The parallax-based three-dimensional trajectory tracking method comprises the following steps: performing histogram equalization and difference operation on two input camera capture images, to obtain a fish movement region in the image, quickly and effectively extracting the fish outline from the original image according to gray scale information of the movement region; secondly, automatically extracting a centroid from the outline image by using a centroid tracking algorithm; then, performing three-dimensional reconstruction on two centroids by a parallax method; finally, combining all three-dimensional points, to obtain a three-dimensional trajectory of fish in stereoscopic vision. The parallax-based three-dimensional trajectory tracking method is applicable to complex environment in biological water quality monitoring process, can track the target accurately and can observe the fish trajectory in a three-dimensional space, so as to reflect the reaction results of the change of water quality on fish movement in time.

The invention discloses a target tracking method based on triple twin hash network learning, and relates to the technical field of computer vision, target tracking and deep learning. According to themethod, firstly, a triple twin hash network is constructed, and the network is composed of a data input layer, a convolution feature extraction layer and a hash coding layer. In the initial training process of the network, a training data set and a random gradient descent back propagation algorithm are used for training the triple twinning Hash network, and after training is completed, the initialcapacity of target positioning can be obtained through the network. In the tracking process, firstly, an input image passes through a triple twin region recommendation network to obtain correspondingcandidate frames, then the candidate frames are input into a triple twin Hash network to be subjected to forward processing, the similarity between each candidate frame and a query sample is calculated, the candidate frame with the highest similarity is selected as a tracking target object, and therefore target tracking is achieved.

The invention discloses a target tracking method and system of a football robot. The target tracking method of the football robot comprises the following steps of (1) building a motion model of the football robot, (2) carrying out discretization on the motion model of the football robot to obtain a state equation of the football robot, (3) building a measurement model of the football robot, (4) building a filtering probability distribution model of a state variable xk-1 of the football robot at the time of k-1, (5) updating measurement, (6) updating prediction, (7) estimating the optimized estimated value of a state variable xk+1 of the football robot at the time of k+1, and (8) tracking a target. The target tracking system of the football robot comprises a data processor, a data storage card, a serial port communication module, a wireless communication module, an omni-directional visual module, an intelligent power module, a motion control card, a motor driving circuit module, a motor, a motor encoder and a football robot training machine. The target tracking method and system of the football robot are reasonable in design, simple and convenient to use and operate, convenient to realize, capable of rapidly and accurately tracking the target of the football robot, high in practicability, good in use effect, and high in popularization and application value.

The invention discloses a multi-sensor fusion greenhouse inspection robot tracking method based on EKF. The method comprises the steps: acquiring distance information from three base stations with known coordinates to a robot in real time through a UWB distance measuring module based on a DS-TWR distance measuring technology while acquiring the acceleration and the course angle of the robot in real time through MPU6050 carried by the robot; acquiring a robot coordinate initial observation value from the three distance values by using a trilateral positioning model based on a least square method; carrying out iterative correction at an observation point by utilizing a Taylor algorithm; simultaneously transmitting the corrected coordinate observation value, and the acceleration and the course angle which are measured in real time into a CTRV motion model; fusing the sensor data through extended Kalman filtering; and outputting the extended Kalman filtering (EKF) result as the current position of the robot so as to complete the greenhouse inspection robot tracking method based on multi-sensor fusion. According to the invention, accurate positioning of a greenhouse inspection robot inan indoor environment can be realized, and the cost of greenhouse management is reduced.

The invention discloses a target tracking method and a device. The method comprises the following steps: in an initialization frame, a mask matrix is established according to a target area, and a spatio-temporal regularization discriminant correlation filter model is established by using space and time constraints; the multi-scale samples of the target in the initial position and rotation angle are extracted, and the scale space discriminant correlation filter model is learnt; in the subsequent frame, the target position and rotation angle are obtained by using spatio-temporal regularization discriminant correlation filter; the scale space is used to discriminate the correlation filter to obtain the target scale; according to the tracking results, the correlation filter is updated online.By using this method, the tracking robustness under occlusion, deformation and rotation of the target can be improved, and the displacement, rotation and scale of the target can be estimated, so thatthe accurate target area can be obtained, and the target can be continuously tracked. The device for realizing the target tracking of the method comprises an image acquisition unit, an image processing unit, a learning unit, a tracking unit and a servo driving unit.

The invention provides a target tracking method based on online state learning and estimation and belongs to the computer vision, computer graphics and image technical field. According to the method, a target positioning and state estimation network is obtained; and the target positioning and state estimation network is composed of a feature extraction network and a regression network, wherein the feature extraction network is a pre-trained AlexNet network, and the regression network is a recurrent neural network (RNN). In an initial network training process, an initial training set and a stochastic gradient descent method are utilized to train the target positioning and state estimation network, after being trained, the target positioning and state estimation network obtains an initial ability to perform target positioning and state estimation. In a tracking process, the target positioning and state estimation network performs forward processing on an inputted image and directly outputs target related information corresponding to the image, wherein obtained target probability and state information decides whether the network to perform online learning, and target position and size information can realize the localization of a target, and therefore, the tracking of the target can be realized.

The invention relates to a target focus tracking method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: receiving a photographing instruction, and controlling a camera to collect a plurality of frames of images according to the photographing instruction; obtaining a current preview image, and performing target main body segmentation on the current preview image to obtain a target main body mask image; determining an external area of the target main body mask pattern according to the area parameters of the target main body mask pattern; and generating a focus tracking prediction frame of the current preview image according to the target main body mask image and the external area, and performing focus tracking processing on a nextframe of image by using the focus tracking prediction frame. By adopting the method, focus tracking processing can be carried out on the target main body according to the focus tracking prediction frame of each frame of image, so that continuous focus tracking can be accurately carried out on the target main body, and the focus tracking accuracy is effectively improved.

The invention discloses a target tracking method based on a structured output correlation filter. The method mainly solves the problem that the tracking is failed due to the change, the shading, the rotation and the like of the target illumination in the prior art. The method comprises the following steps of (1) preprocessing a first frame image; (2) constructing a structured output correlation filter; (3) figuring out an optimal structured output correlation filter; (4) preprocessing a current frame image; (5) determining the position of a to-be-tracked target in the current frame image; (6)optimizing the structured output correlation filter; (7) judging whether all frame images in a to-be-tracked video image sequence are selected or not; if yes, ending the process; otherwise, executingthe step (4). According to the invention, the information contained in a sample can be better described by constructing the structured output correlation filter. As a result, characteristics with highdistinguishing degree can be learned through the structured output correlation filter, and a target can be tracked stably and accurately.

The invention relates to a visual characteristic defogging and image stabilizing detection system including the following modules and functions. First, image data is input into a fog feature analysisand enhancing module and an enhanced image and a correction parameter 1 are acquired through calculation. Second, the enhanced image is input to a multi-level parameter feedback control based object detection system module which calculates out the position of a to-be-detected object through detection and recognition and outputs a target displacement and a correction parameter 2. Third, an image stabilizing displacement calculating module outputs image stabilizing displacement to an image stabilizing flyback integrated control platform. Fourth, the image stabilizing flyback integrated control platform implements an image stabilizing function according to the image stabilizing displacement and adjusts sensor integral time according to the correction parameter 1 and the correction parameter 2. According to the invention, the system implements functions of visual characteristic analysis based defogging enhancement, photoelectric mechanical integrated image stabilization, detection, tracking and the like, achieves good definition, good stability and good accuracy and can be embedded into target detection systems of different types.

The invention discloses a target tracking method applied to videos. According to the target tracking method, an FAST algorithm and a KLT algorithm are effectively combined together, so that problems in target tracking in videos can be solved; the FAST algorithm and the KLT algorithm are correspondingly improved, so that target tracking can be more accurate. With the target tracking method of the invention adopted, appropriate feature information is efficiently selected so as to be tracked, so that a target can be accurately tracked. The target tracking method provided by the invention has the advantages of high accuracy, high robustness, higher simplicity and short operation time.

The invention discloses a target tracking method based on objectivity and a correlation filter, comprising: expanding a target region in a frame, constructing a training sample and training a kernelized correlation filter, calculating a forward optical flow of a search region in the current frame when each new frame of image comes, obtaining boundaries of a moving object through an edge detectionalgorithm in combination with the color information of the search region, generating a target candidate by an objectivity method according to an edge map generated by edge detection, and adjusting thefour boundaries of a candidate window to fit the true edges of the object. In the search region of the current frame, the initial target position corresponding to the maximum response value is detected by the kernelized correlation filter, the candidate response value is obtained by calculating the objectivity, the target position and size of the current frame are obtained by comparing the response values and the model is updated. The method can effectively deal with the problems such as fast motion and scale change, thereby achieving robust tracking.

The invention discloses a space-time channel constraint correlation filtering tracking method based on suppressible abnormity. The method comprises the steps of S1, extracting HOG features, first depth features and second depth features of a t-th frame; S2, performing fusion processing on the HOG feature, the first depth feature and the second depth feature to obtain a first fusion feature X, anddetermining the position and the scale of a target in the t-th frame of image based on the first fusion feature X and a filter; S3, according to the feature map of the t-th frame, updating the filterbased on a space-time channel constraint related filtering model capable of suppressing abnormality; and S4, repeating the step S2S4 until all frames are tracked, and finally obtaining a tracking result. According to the method, the feature representation capability of the target template is remarkably improved by combining manual features with depth features, self-adaptive channel feature selection is realized through l2, 1 norm, and the problems of boundary effect and background clutter are effectively solved.

The invention discloses a visual tracking algorithm taking paper as an object, and the algorithm comprises the steps: carrying out the classification of positive and negative samples in a first image frame; calculating the feature values of all positive and negative samples, and determining a first target area; carrying out the sampling nearby the target region of the former frame on each image frame after the second frame through a sampling block; calculating the scores of all sampling blocks through combining with the feature values of the positive and negative samples of the former frame; employing the sampling block with the highest score, and carrying out the detection of each sampling block through an LSD linear detection algorithm; finding the sampling blocks forming an inclined angle in the section [70 degrees, 110 degrees], wherein the intersection of two straight lines is located in the sampling block; further selecting one sampling blocks with the intersection of two straight lines being the nearest to the center as the target sampling block. The method gives consideration to a problem of a photographing angle of a camera when paper moves, and enables a finally obtained tracking target to be more accurate.

The invention provides a target tracking method based on a transfer learning regression network, which relates to the technical field of computer vision. A to-be-tracked target object is selected anddetermined from an initial image; a target position regression network is constructed based on block prediction; generation of a training data set for tracking and network training are carried out; image input is carried out, in a real-time processing condition, a video image acquired by a camera and saved in a storage area is extracted as a to-be-tracked input image; and target positioning is carried out, the acquired image is inputted to the position regression network, after network forward processing, a network output layer obtains 8*8*8 relative position data. Network updating is carriedout, according to the obtained target position, 8*8*8 relative positions between 8*8 image blocks divided by the whole image and the target are calculated, and together with the current input image, agroup of training data can be formed.

The invention belongs to the technical field of digital image processing, and particularly relates to a target tracking method based on an improved double-center particle swarm optimization algorithm.Firstly, a target position of a first frame image in an image sequence is selected; according to the selected target, Hu invariant moment for the image of the target area to describe the shape feature of the target; meanwhile, calculating an HSV color histogram for the image of the target area is calculated according to the framed target to describe the color characteristics of the target; then,the calculated shape feature vector H and color feature vector G are connected in series to obtain a target feature vector [H, G] after feature fusion, namely target template features; then, a double-center particle swarm optimization algorithm is applied to subsequent image frames, and the positions of globally optimal particles in the image frames are obtained; and finally, the proposed anti-occlusion target template updating strategy is utilized to obtain updated new target template features. The method provided by the invention has good tracking accuracy for the target, and has good real-time performance, shielding resistance and robustness for target tracking.

The embodiment of the invention discloses a multi-target tracking method and a multi-target tracking device based on a TSK fuzzy system, and a storage medium. The method comprises the steps of firstlyjudging whether the number of targets with stable tracks is greater than 0 or not; if yes, constructing a TSK fuzzy classifier, inputting the observation set into the TSK fuzzy classifier to obtain alabel vector matrix, and then performing data association on the label vector matrix; if not, calculating the feature similarity between the target object in the target set and the observation objectin the observation set, inputting the feature similarity into the TSK fuzzy model to obtain a membership matrix, and then performing data association on the membership matrix; and finally performingtrajectory management based on a data association result. Through the implementation of the invention, the TSK fuzzy classifier is established to associate the stable flight path with the observation.The TSK fuzzy model is utilized to perform simple data association on the new observation, so that the data association between the target and the observation can be accurately completed. The accurate tracking of multiple targets in the video is realized.

A multi-feature and self-adaptive standard hedging combined target tracking method is characterized by comprising the following steps that (1) using multiple features of a tracking target to constructa dictionary, wherein the tracking target is expressed as a vector in a dictionary space; (2) designing self-adaptive standard hedging based on the standard hedging; and (3) in a self-adaptive standard hedging tracking framework, estimating the current tracking target. The multi-feature and self-adaptive standard hedging combined target tracking method still has good accuracy and robustness in acomplex environment.

The invention relates to the technical field of unmanned aerial vehicle tracking, in particular to a target tracking method and device of an unmanned aerial vehicle and an image processing chip. The method comprises the following steps of identifying a target image returned by an unmanned aerial vehicle, calculating a first identification result from the target image, conducting score comparison on the identification boxes in the second identification result and all the identification boxes in the first identification result, and constructing and screening out a target list with the highest conformity, calculating a relative offset direction of each identification box in the target list, determining a target device between the target image and the previous frame of target image as an optimal matching relationship, and tracking a plurality of target devices in the target image. By applying the target tracking method provided by the embodiment of the invention, real-time tracking of the specific target device can be kept in the inspection process and in the process that the unmanned aerial vehicle pan-tilt camera turns to the specific target device, and photographing is performed when the specific target device appears in the center of the picture, so that the imaging quality of the inspection picture is improved, and the imaging quality of the inspection picture is improved.

The invention discloses an energy saving-type Internet of things target tracking method, which belongs to the technical field of Internet of things target tracking methods. The method particularly comprises steps: sparse representation for moving target sensing information changes by a node in a sensing area is built, observation and sampling on the moving target sensing information by the node are used, and sensing differential information is reconstructed; a sensing differential method converted by a background differential method in a video tracking technology is then used for processing, the sensing area of the Internet of things is seen as a video monitoring area, moving of the target in the sensing area is seen as moving of a monitored target in the video, the sensing differential information and the initial sensing information are used to determine current effective tracking information, and a moving target is detected. A compressed sensing method is adopted to sample and reconstruct the sensing information, the sensing differential method is adopted for target tracking, and an energy saving-type and high-precision Internet of things target tracking method is provided. The method can be applied to various industries such as security and theft prevention, environmental monitoring, military tracking, and cargo warehousing, and has a wide application prospect.