CN116543017A - Single-target tracking method based on sparse optical flow motion enhancement - Google Patents
Single-target tracking method based on sparse optical flow motion enhancement Download PDFInfo
- Publication number
- CN116543017A CN116543017A CN202310495347.7A CN202310495347A CN116543017A CN 116543017 A CN116543017 A CN 116543017A CN 202310495347 A CN202310495347 A CN 202310495347A CN 116543017 A CN116543017 A CN 116543017A
- Authority
- CN
- China
- Prior art keywords
- target
- frame image
- tracking
- current frame
- resolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a single-target tracking method based on sparse optical flow motion enhancement, which belongs to the field of target tracking, and aims to improve the accuracy of motion modeling through a sparse optical flow algorithm, and reduce the parameter number of a tracking model by adopting a linear matrix when the tracking model is constructed, so as to avoid the problems of excessive model parameter number and reduced tracking efficiency caused by motion modeling error propagation; the method comprises the following steps: performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is positioned in the current frame image; extracting image features of the target candidate region image to obtain multi-level resolution features; and inputting the multi-level resolution characteristic into a preset tracking model, and estimating the position of the tracking target in the current frame image.
Description
Technical Field
The invention relates to the technical field of target tracking, in particular to a single target tracking method based on sparse optical flow motion enhancement.
Background
In the related art, since correlation filtering can process two signal correlations, it has been widely used in single-object tracking. The current correlation filtering method uses motion information of a tracking target in a tracking sequence to model, extracts gradient features, color features or depth features of an image from a candidate region of the tracking target obtained by modeling, and calculates correlation between a frame to be tracked and a template frame (usually a first frame) to predict target positioning. In order to cope with changes of the appearance scale of the target in the sequence, the conventional correlation filtering method generally adopts an online updating mode for the tracking model so as to avoid the influence of the changes of the appearance scale of the target.
However, the current related filtering method is adopted to track the target, so that the problems of low tracking efficiency and low tracking accuracy are solved.
Disclosure of Invention
In view of the above, the present invention aims to provide a sparse optical flow motion enhanced single-target tracking method, so as to solve the problems of low tracking efficiency and low tracking accuracy of the current target tracking method.
In a first aspect of the embodiment of the present invention, a single-target tracking method based on sparse optical flow and enhanced understanding is provided, where the method includes:
performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is positioned in the current frame image; wherein the current frame image is any frame image except the first frame image in the image sequence;
Extracting image features of the target candidate region image to obtain multi-level resolution features;
inputting the multi-level resolution characteristic into a preset tracking model, and estimating the position of the tracking target in the current frame image; the preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter.
Further, the motion modeling of the tracking target by using a sparse optical flow algorithm based on the previous frame image adjacent to the current frame image and the current frame image includes:
acquiring a previous frame image adjacent to the current frame image;
converting the previous frame image and the current frame image into gray level images, and determining angular point coordinates of corresponding pixel positions in the two frames of images by using a Harris angular point detection algorithm;
determining the optical flow moving pixel distance between the corresponding characteristic angular points in the previous frame image and the current frame image based on the angular point coordinates;
determining the moving distance of the tracking target in two frames according to the optical flow moving pixel distance;
determining the central position of the target candidate region in the current frame image according to the predicted target position of the previous frame image and the moving distance;
And cutting out the target candidate region image with a preset size from the current frame image based on the center position of the target candidate region.
Further, the determining, according to the predicted target position and the moving distance of the previous frame image, the center position of the target candidate region in the current frame image includes:
determining the central coordinate of the predicted position of the tracking target in the previous frame of image;
determining the center point coordinates of the target candidate region in the current frame image based on the center coordinates and the moving distance;
the clipping the current frame image into the target candidate region image with a preset size based on the center position of the target candidate region includes:
and cutting out the target candidate region image with a preset size from the current frame image based on the center point coordinates of the target candidate region and a preset cutting frame.
Further, the constructing step of the multi-resolution filter includes:
cutting out a candidate region with a preset size from a first frame image of the image sequence by taking a target truth value boundary frame as a center; the target truth value bounding box is the marked bounding box of the tracking target;
Extracting image features aiming at the candidate areas to obtain the multi-level resolution features;
introducing an interpolation model, and mapping the multi-level resolution features to continuous spaces with the same continuous period so as to continuously carry out the multi-level resolution features;
defining a group of continuous convolution filters for the continuous space to obtain continuous convolution operators through parameterization;
determining a primary convolution filter set from the continuous convolution filters based on the linear matrix;
constructing a dimensionality-reduced convolution filter through mapping based on the main convolution filter set, and taking the dimensionality-reduced convolution filter as the multi-resolution filter;
and optimizing the multi-resolution filter by taking the first frame image as a preset sample to obtain the multi-resolution filter with the optimal parameter value.
Further, the optimizing the multi-resolution filter by using the first frame image as a preset sample includes:
constructing a minimum loss function for the multi-resolution filter based on a preset sample pair; the preset sample pair is obtained by converting the first frame image; the minimization loss function employs L2 regularization to limit the values of the multi-resolution filter;
And iteratively optimizing the minimized loss function by adopting a Gaussian-Newton iteration and conjugate gradient method to obtain the optimal parameter value of the multi-resolution filter.
Further, after the inputting the multi-resolution feature into a preset tracking model to estimate a position of the tracking target in the current frame image, the method further includes:
updating the Gaussian mixture model by adopting the multi-level resolution features to add the multi-level resolution features into a training sample set; the Gaussian mixture model is used for carrying out feature modeling on the multi-level resolution features so as to express feature distribution of the multi-level resolution features by adopting Gaussian components; the multi-level resolution feature takes the target correlation output of the multi-level resolution feature as a label;
a preset frame number is spaced, and the Gaussian component is adopted to estimate the expected loss value of the multi-resolution filter;
and updating parameters of the multi-resolution filter so that the updated output of the multi-resolution filter meets the expected loss value.
Further, said estimating an expected loss value for said multi-resolution filter using said gaussian component comprises:
Transforming the loss function of the multi-resolution filter based on the labels corresponding to the multi-resolution features in the training sample set to obtain an error expectation function of the multi-resolution filter;
substituting the Gaussian component into the error expectation function to obtain the expected loss value.
Further, the preset tracking model includes: the system comprises a mapping module, an acquisition module, a fusion module and an estimation module; the step of inputting the multi-resolution features into a preset tracking model, and estimating the position of the tracking target in the current frame image includes:
inputting the multi-level resolution features into the mapping module to obtain continuous features;
inputting the continuous features to the acquisition module to acquire a target response graph at each resolution through the multi-resolution filter;
inputting the target response graphs under a plurality of resolutions to the fusion module to obtain a target response function;
the target response function is input to the estimation module to estimate a position of the tracking target in the current frame image based on a grid search method and conjugate gradient descent.
In a second aspect of embodiments of the present invention, there is provided a sparse optical flow motion enhancement-based single-target tracking device, the device comprising:
The motion modeling module is used for performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is positioned in the current frame image; wherein the current frame image is any frame image except the first frame image in the image sequence;
the feature extraction module is used for extracting image features of the target candidate region image to obtain multi-level resolution features;
the estimation module is used for inputting the multi-level resolution characteristics into a preset tracking model and estimating the position of the tracking target in the current frame image; the preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter.
A third aspect of an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the sparse optical flow motion based enhanced single target tracking method according to the first aspect above.
Compared with the prior art, the sparse optical flow motion enhancement-based single-target tracking method has the following advantages:
the invention provides a single-target tracking method based on sparse optical flow motion enhancement, which comprises the following steps: performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is positioned in the current frame image; wherein the current frame image is any frame image except the first frame image in the image sequence; extracting image features of the target candidate region image to obtain multi-level resolution features; inputting the multi-level resolution characteristic into a preset tracking model, and estimating the position of the tracking target in the current frame image; the preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter; therefore, the invention adopts the sparse optical flow algorithm to calculate the front and back two frames of images, avoids the problem of target loss caused by target tracking only according to the clipping region of the previous frame of image, and carries out motion modeling through the sparse optical flow algorithm, thereby improving the modeling accuracy; meanwhile, as the multi-resolution filter is constructed through the linear matrix, the parameter quantity of the multi-resolution filter is reduced, the parameter updating efficiency of the multi-resolution filter is improved, and the tracking efficiency is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention. In the drawings:
FIG. 1 is a flowchart showing steps of a sparse optical flow motion enhancement-based single-target tracking method according to an embodiment of the present invention;
FIG. 2 is a flow chart showing steps of a method for constructing a multi-resolution filter according to an embodiment of the present invention;
FIG. 3 is a flowchart showing steps of a single-target tracking method based on sparse optical flow motion enhancement according to still another embodiment of the present invention;
FIG. 4 is a flowchart showing steps of a sparse optical flow motion enhancement-based single-target tracking method according to a second embodiment of the present invention;
fig. 5 shows a schematic structural diagram of a single-target tracking device based on sparse optical flow motion enhancement according to a third embodiment of the present invention.
Detailed Description
It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other.
In the related art, a related filtering technology is often used to track a target object, where in order to cope with changes of appearance dimensions of a target in a sequence, a conventional related filtering method generally adopts an online updating mode for a tracking model, that is, updating target features with a latest target template every fixed frame to improve robustness of the tracking model. However, the current amount of parameters of the tracking model is millions, so using a target template of a few frames to update the model parameters not only reduces tracking of the tracking model, but may also cause overfitting of the model.
In addition, tracking the attribute related to the target in the sequence has the motion information of the target besides the appearance characteristic, and the traditional related filtering method models the motion information of the target of the current frame in the sequence in a mode that the predicted target is cut out from the predicted position of the image of the previous frame in the current frame appearance area. However, when the target predicted by the previous frame is unreliable, the region cut out according to the error target may not contain the tracking target at all, and the error is propagated, so that the model accuracy is greatly reduced.
In view of this, the present invention provides a sparse optical flow motion enhancement-based single-target tracking method to reduce the parameter amount of a model by adopting a linear matrix to reduce the dimension when constructing a multi-resolution filter; meanwhile, a sparse optical flow algorithm is adopted to compensate the camera motion so as to avoid error propagation caused by unreliable targets predicted in the previous frame.
The sparse optical flow motion enhancement-based single-target tracking method of the present invention will be described in detail below with reference to the accompanying drawings in conjunction with the embodiments.
Example 1
Referring to fig. 1, fig. 1 shows a flowchart of steps of a sparse optical flow motion enhancement-based single-target tracking method according to an embodiment of the present invention, as shown in fig. 1, including:
S101, performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is located in the current frame image.
Wherein the current frame image is any frame image except the first frame image in the image sequence. Specifically, in the object tracking process based on correlation filtering, the first frame image is used for initializing the filter so as to determine the position of the tracked object by correlating the image features of the object in the first frame image with images in subsequent frames, so that the subsequent images after the first frame determine that the object candidate region of the tracked object exists through motion modeling.
The image sequence may be an image sequence obtained by capturing a video stream with a tracking target, or may be a series of images with a tracking target captured by a camera, where the image sequences are all arranged in time sequence.
Specifically, the sparse optical flow algorithm refers to a method for performing image registration on sparse points on an image, namely, giving a plurality of points on a reference image, and finding out corresponding points in the current image; in the embodiment of the invention, as two modes of camera motion and target motion exist, under the condition of camera motion, feature points can be selected for the picture background; under the condition of target movement, feature points can be selected for the target; in some embodiments, feature points may also be selected for both the image background and the object to facilitate the motion modeling process.
The target moving distance between the previous frame image and the current frame image can be calculated through a sparse optical flow algorithm, the candidate position of the target in the current frame image can be determined according to the moving distance and the predicted target position of the previous frame image, and the image with the preset size is cut out as a target candidate area image for the candidate position, so that the input interference of irrelevant image features on the target tracking process is reduced or the tracking efficiency is reduced.
And S102, extracting image features of the target candidate region image to obtain multi-level resolution features.
The multi-level resolution features may be HOG (Histogram of Oriented Gradient, gradient color histogram), SIFT (Scale-invariant feature transform ), global feature information features, among others.
S103, inputting the multi-resolution features into a preset tracking model, and estimating the position of the tracking target in the current frame image.
The preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter.
Specifically, the preset tracking model comprises a mapping module, an acquisition module, a fusion module and an estimation module; the method comprises the steps that a multi-level resolution characteristic is input into a mapping module of a preset tracking model, and the multi-level resolution characteristic is mapped into a continuous space by the mapping module to obtain the continuous characteristic; and inputting the continuous characteristics into an acquisition module, acquiring a target response graph under each resolution through a multi-resolution filter in the acquisition module, and fusing a plurality of target response graphs to obtain a target response function.
The target response function is input to an estimation module, and the response value of the target response function at each position is estimated through a grid search method. Based on the response value of each position, taking the position with the largest response value in the grid search as the central position of the target; and applying standard Newton iterative optimization on Fourier series expansion, and further estimating to obtain the final target position.
Wherein, the steps of constructing the multi-resolution filter refer to fig. 2, fig. 2 shows a flowchart of steps of a method for constructing the multi-resolution filter according to an embodiment of the present invention, as shown in fig. 2, including:
s201, cutting out a candidate region with a preset size from the center of a target truth value boundary box for the first frame image of the image sequence.
The target truth value bounding box is the marked bounding box of the tracking target.
In the embodiment of the invention, the position of the tracking target in the first frame image is marked by a person, so that the marked boundary frame of the tracking target is a target truth value boundary frame, and a candidate region with a preset size is determined based on the center of the target truth value boundary frame, so that a candidate region containing the tracking target is obtained; the size of the candidate region is cut out for the purpose of facilitating feature extraction, and the invention is not particularly limited.
S202, introducing an interpolation model, and mapping the multi-level resolution characteristic to a continuous space with the same continuous period so as to continuously carry out the multi-level resolution characteristic.
Wherein, because the multi-level resolution characteristic is the multi-dimensional characteristic of discrete distribution, in order to facilitate the subsequent filter construction, an interpolation model is introduced, and the multi-level resolution characteristic is mapped into a continuous space. Specifically, an interpolation operator J is defined for each feature dimension d d ：L 2 (T) is a Hilbert space:
wherein N is d B for the resolution of each feature dimension d Representing the use of a cubic interpolation kernel function, the entire interpolation operator J d By interpolation function b d Is constructed by superposition of various translational forms, x d [n]Representing the weight of each translated function.
S203, defining a group of continuous convolution filters for the continuous space to acquire continuous convolution operators through parameterization.
S204, determining a main convolution filter set from the continuous convolution filters based on the linear matrix.
And S205, constructing a dimensionality-reduced convolution filter through mapping based on the main convolution filter set, and taking the dimensionality-reduced convolution filter as the multi-resolution filter.
Specifically, the continuous convolution operator S f ：χ→L 2 (T) mapping the samples xεχ into a space defined in continuous space tε [0 ]A target confidence function in T); wherein the continuous convolution operator consists of a series of convolution filters f= (f) 1 ，...，f D )∈L 2 (T) D Parameterization, where f d ∈L 2 (T) is a continuous filter for the characteristic channel d. At the same time, a linear mapping matrix is definedBased on the main convolution filter set f 1 ，…，f C C < D, by linear mapping +.>The convolution filter in f can be constructed:
wherein S is Pf { x } is the target confidence function, P is the linear matrix, f is the convolution filter, J d {x d And the interpolation operator of the feature dimension d.
S206, optimizing the multi-resolution filter by adopting the first frame image as a preset sample to obtain the multi-resolution filter with the optimal parameter value.
Wherein the optimizing the multi-resolution filter comprises:
constructing a minimum loss function for the multi-resolution filter based on a preset sample pair; the preset sample pair is obtained by converting the first frame image; the loss function employs L2 regularization to limit the values of the multi-resolution filter.
And iteratively optimizing the minimized loss function by adopting a Gaussian-Newton iteration and conjugate gradient method to obtain the optimal parameter value of the multi-resolution filter. The iteration times are not particularly limited, and the optimal parameter value can be obtained. Illustratively, the formula is iteratively optimized using 10 Gaussian-Newton iterations and 20 conjugate gradient methods to obtain the optimal values of the filter and the linear matrix.
Specifically, image enhancement strategies such as contrast enhancement, brightness enhancement, gaussian noise addition, laplace sharpening and the like are carried out on the first frame image, and the number of preset samples is expanded; wherein the filter f d Is performed on a given m training sample pairsThe minimum functional E (f, P) is expressed below:
wherein m represents the number of training samples, alpha j The weight of each training sample is controlled, and the weight can also be regarded as the influence of the training sample on the filter model;is an L2 regularization term, w is a weight matrix for spatial location; s is S Pf {x j Is sample x j Is a target confidence function of (1).
Wherein the Fourier coefficients of the mapped successive features after feature extraction of the first frame image are used to train a filter with a minimum of E (f, P) in the Fourier domainThe fourier coefficients described above are substituted into the minimization functional E (f, P) according to the pasmodus theorem (energy conservation theorem) of fourier variation, as follows:
wherein,,fourier coefficients representing a continuous feature map z=j { x }, whereIs x d Expression of discrete fourier transform; the last term of the regularization term is the Buddha Luo Beini Usness norm, also called F-norm, for the linear matrix P, λ=2×10 -7 Is a control weight; and then, iteratively optimizing the minimized functional by using a Gaussian-Newton iteration and conjugate gradient method to obtain an optimal value of the filter and the linear matrix, thereby obtaining the multi-resolution filter with optimal parameters.
Referring to fig. 3, fig. 3 shows a flowchart of steps of a single-target tracking method based on sparse optical flow motion augmentation according to still another embodiment of the present invention, as shown in fig. 3, including:
s301, acquiring a previous frame image adjacent to the current frame image.
S302, converting the previous frame image and the current frame image into gray level images, and determining corner coordinates of corresponding pixel positions in the two frames of images by using a Harris corner detection algorithm.
S303, determining the optical flow moving pixel distance between the corresponding characteristic corner points in the previous frame image and the current frame image based on the corner point coordinates.
The sparse optical flow algorithm calculates the moving distance of the feature points between two frames, so that the previous frame image adjacent to the current frame image is acquired first, and calculation is performed based on the feature points between the two frames. Since sparse optical flow algorithms rely on sparse optical flow hypotheses: the gray value of the moving object on the picture is not changed in a short time; therefore, for the previous frame image and the current frame image, both the two frame images are converted into gray images, the positions of the characteristic points in the other image are determined through the gray values of the characteristic points in the images, and then the moving distance is determined according to the coordinates. Specifically, a Harris corner detection algorithm is adopted to determine corner coordinates H of corresponding pixel positions of the front frame and the rear frame t-1 (x，y)，H t (x+u (x, y), y+v (x, y)); based on the angular point coordinate positions, a moving distance calculation formula is constructed as follows:
I(x，y，t-1)＝I(x+u(x，y)，y+v(x，y)，t)
wherein I (x, y, t-1) represents the brightness of the pixel location in the t-1 frame image; i (x+u (x, y), y+v (x, y), t represents the brightness of the corresponding pixel position in the t frame image, u (x, y) represents the moving distance of the pixel position on the horizontal axis, and v (x, y) represents the moving distance of the pixel position on the vertical axis.
Solving the above calculation formula: i (x, y, t-1) ≡I (x, y, t) +I x u(x，y)+I y v (x, y) Taylor expansion, yielding:
solving the above equation yields specific values of u (x, y) and v (x, y).
S304, determining the moving distance of the tracking target in two frames according to the optical flow moving pixel distance.
The moving distance of the characteristic points, namely the pixel positions, in the horizontal axis and the vertical axis of the coordinates is obtained through the steps, and the moving distance of the tracking target in two frames can be obtained, wherein the target positions are changed in the background moving process, if the moving distance of the pixel positions is determined by selecting the characteristic points of the background, the moving distance of the tracking target in two frames of images is determined according to the moving distance of the pixel positions.
Specifically, as two moving modes of target movement and background movement, namely camera movement exist in the target tracking process, for the target movement, characteristic points are selected from target positions so as to determine the moving distance conveniently, and therefore, the characteristic points can be directly extracted from the target positions; for background movement, feature points can be selected from the background part of the picture so as to facilitate calculation of movement distance; in some embodiments, to improve the accuracy of calculation of the moving distance, the moving distance may be calculated and analyzed comprehensively by selecting feature points from the target position and the background position of the picture, respectively. The camera movement and the target movement are opposite, namely, the camera movement to the left can be regarded as the target movement to the right, the direction can be set for the movement distance, the positive direction indicates the target movement, the negative direction indicates the background movement, and the calculation of the subsequent movement distance and the determination of the center coordinates of the target are convenient.
S305, determining the center position of the target candidate area in the current frame image according to the predicted target position of the previous frame image and the moving distance.
S306, cutting out the target candidate region image with a preset size from the current frame image based on the center position of the target candidate region.
Specifically, after the moving distance is determined, the position coordinates (x t-1 ，y t-1 ) Then based on the distances u (x, y) and v (x, y) of the tracking target movement on the horizontal axis and the vertical axis, respectively, the center coordinates (x, y) = (x) of the target position of the current frame image can be obtained t-1 +u(x，y)，y t-1 +v (x, y)), the image region to be cropped can be determined based on the center position coordinates, and the target candidate region is cropped by adopting a cropping frame with a preset size, so that the subsequent feature extraction process is facilitated.
In the embodiment of the invention, although the center point coordinates of the target in the current frame image are determined, the essence is rough estimation of the target position, so that the target candidate region is obtained by image clipping only by adopting the center point coordinates, the clipped image position is prevented from losing the target, and then the target position is redetermined based on the clipped region.
S307, extracting image features of the target candidate region image to obtain multi-level resolution features.
S308, inputting the multi-resolution features into a preset tracking model, and estimating the position of the tracking target in the current frame image.
And S309, updating the Gaussian mixture model by adopting the multi-level resolution features to add the multi-level resolution features into a training sample set.
The Gaussian mixture model is used for carrying out feature modeling on the multi-level resolution features so as to express feature distribution of the multi-level resolution features by adopting Gaussian components; the multi-resolution feature is tagged with a target correlation output of the multi-resolution feature.
S310, estimating the expected loss value of the multi-resolution filter by adopting the Gaussian component at intervals of preset frames.
And S311, carrying out parameter updating on the multi-resolution filter so that the output of the updated multi-resolution filter meets the expected loss value.
In the embodiment of the invention, each frame is adopted as a training sample for updating, and the interval between the frames of continuous pictures is smaller, so that the picture similarity is high, and the problem of over fitting of a tracking model can be caused; thus, the samples of N frames are used for updating after an interval of N frames.
Specifically, the input multi-level resolution characteristic x is used as a training sample to be added into a training sample set, and the output response diagram y is used as a label of the multi-level resolution characteristic x. Since the multi-resolution filter is constructed and optimized on the multi-resolution feature x and the corresponding output response graph y, i.e., given p (x, y), and the target correlation output y on the multi-resolution feature x is a gaussian function, the difference between the target correlation output y corresponding to different training samples is represented by the difference between the peak values of the gaussian function, it can be assumed that the outputs corresponding to all training sample features are the same, i.e., y=y 0 The shift of different y peaks is accomplished by the multi-level resolution feature x, so the distribution of training samples can be decomposed into:thereby(s)>Characterizing the offset of the y peak as a fixed value; p (x) represents the distribution of training sample features; further, the optimization of the multi-resolution filter can be performed by estimating only p (x).
Thus, the mixed Gaussian model is used for feature modeling of training sample features: order the
Wherein L representsGaussian component N (x; μ) in Gaussian mixture model l The method comprises the steps of carrying out a first treatment on the surface of the I) Number of pi l A priori weights, μ representing the first gaussian component l E chi is the mean of the gaussian components and the identity matrix I is the covariance of the gaussian components. Thus, the Gaussian components in the Gaussian mixture model are used for representing the characteristics of the training sample, then the Gaussian components are substituted into the training loss function, so that the expected loss of the output of the multi-resolution filter is estimated through the corresponding Gaussian mixture model of the training sample, and the multi-resolution filter is updated online based on the expected loss:
Wherein pi l A priori weights, μ representing the first gaussian component l E chi is the mean of the Gaussian components, y 0 Outputting a value for a fixed response of the current hypothesis; s is S f {μ l [ mu ] is adopted l And outputting target correlation of the features. The training loss function of the multiresolution filter may employ a minimization of the target correlation expectation expression:thereby, the Gaussian component and the fixed response output value are brought into the expected expression, and the expected loss calculation formula is obtained; wherein, expect->Is estimated over the distribution p (x, y).
In the embodiment of the invention, a training sample is added into a training sample set realized by a Gaussian mixture model, and the Gaussian mixture model is required to be updated through the characteristics of the current frame; the specific updating method comprises the following steps:
give a new training sample x j First, initialize a pi m ＝γ，μ m ＝x j Wherein the learning rate γ=0.012. If the total Gaussian component exceeds L, the two peaks μ are combined k Sum mu l The close gaussian components k and l to update the gaussian model:
wherein pi l A priori weights, μ representing the first gaussian component l E chi is the mean value of the Gaussian components, pi k Representing the a priori weight of the kth gaussian component.
According to the embodiment of the invention, a filter is initialized based on a first frame image in an image sequence, and for a subsequent frame image in the image sequence, a previous frame image and a current frame image are adopted to calculate the moving distance of a target between two frames by adopting a sparse optical flow algorithm; predicting the center point position of a target in the current frame image through the moving distance and the target position in the previous frame image, further extracting the characteristics of the image at the center point position, and estimating the target position in the current frame image through a preset tracking model; therefore, the characteristic dimension reduction is carried out by adopting the linear matrix in the initialization process of the multi-resolution filter, so that the parameter quantity of the multi-resolution filter is reduced, and the problem that the tracking efficiency is influenced due to slower updating caused by more parameters in the optimization process is further avoided; by adopting the sparse optical flow algorithm to calculate the target moving distance between two frames of images so as to perform motion modeling on the tracking target, the adopted images are the current frame image and the previous frame image, rather than only the target candidate region of the previous frame image, so that the problem that the target is lost due to the fact that the candidate region is adopted to perform motion modeling is avoided, and the candidate region containing the tracking target can be cut out from the current frame image more accurately.
Example two
Referring to fig. 4, fig. 4 shows a flowchart of steps of a sparse optical flow motion enhancement-based single-target tracking method according to an embodiment of the present invention, as shown in fig. 4, where the method includes:
s401, constructing a multi-resolution filter of a preset tracking model based on the first frame image.
Specifically, based on a given target bounding box true value in the first frame image, the image is cropped to extract image features of the cropped portion, and a convolution filter of the relevant filtering tracking model is initialized through the extracted multi-level resolution features and a linear matrix of factorization dimension reduction is utilized.
In some embodiments, the weight matrix of the main convolution filter is set to model the importance of several composite indicators, i.e., the weights of the composite indicators, with a linear matrix, due to the difference in the target aspects of interest of the different filters in a set of convolution filters.
Specifically, after constructing the multi-resolution filter, a weight mapping matrix is definedFor the main convolution filter set f 1 ，…，f C Weighting:
wherein W is C Representing each principal convolution filter f c Is a weight of (2).
Thus, in optimizing the filter, the influence of the weights is also considered: a minimization function E (f, P) optimization filter is defined. Wave filter f d Is performed on a given m training sample pairsThe minimum functional E (f, P, W) is expressed below:
wherein m represents the number of training samples, alpha j The representation of the weight controlling each training sample can also be seen as the influence of the training samples on the filter model.Is the L2 regularization term, betaIs a weight matrix for spatial locations, W is a weight matrix for convolution filters.
The filter is trained in the fourier domain to minimize E (f, P, W). Fourier coefficients of continuous features using post-feature extraction mappingWherein (1)>Is x d Expression of discrete fourier transform.
Substituting the above formula into the minimized functional E (f) of the fourth step according to the pasmodus theorem (energy conservation theorem) of fourier variation, the following formula:
wherein,,the fourier coefficients representing the continuous feature map z=j { x } and the last term of the regularization term is the phor Luo Beini us norm, also called F norm, for the linear matrix P, λ=2×10 -7 Is the control weight.
The formula is iteratively optimized by using 10 Gaussian-Newton iterations and 20 conjugate gradient methods to obtain the optimal values of the filter and the linear matrix for subsequent tracking.
S402, performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is located in the current frame image.
Wherein the current frame image is any frame image except the first frame image in the image sequence.
Specifically, converting the two frames of images into gray level images, and determining angular point coordinates of corresponding pixel positions in the two frames of images by using a Harris angular point detection algorithm. Calculating the moving distance corresponding to the pixel position in the two frames of images according to the angular point coordinates, namely the optical flow moving distance; and then determining the center position of the current frame image according to the optical flow moving distance and the target position of the previous frame image, and cutting out a target candidate region with a preset size based on the center position.
S403, extracting image features of the target candidate region image to obtain multi-level resolution features.
S404, inputting the multi-resolution features into a preset tracking model, and estimating the position of the tracking target in the current frame image.
The preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter.
In the embodiment of the present invention, according to the filter f, the linear matrix P and the weight matrix W initialized in the step S401, the samples x e χ in the continuous feature space are mapped into a target confidence function S (t) =s in the continuous interval t e [0, t) by using a convolution filter based on factorization dimension reduction Pf {x}(t)：
Wherein s is d Representing a target response plot at a d-th dimensional resolution byThe objective response functions at different resolutions are fused.
And acquiring response values of all positions of the fused target response function by adopting a grid search method, taking the position with the largest response value as a target center position, and further estimating the final target position based on the target center position.
And S405, updating the Gaussian mixture model by adopting the multi-level resolution features to add the multi-level resolution features into a training sample set.
The Gaussian mixture model is used for carrying out feature modeling on the multi-level resolution features so as to express feature distribution of the multi-level resolution features by adopting Gaussian components; the multi-resolution feature is tagged with a target correlation output of the multi-resolution feature.
S406, spacing preset frames, and estimating the expected loss value of the multi-resolution filter by adopting the Gaussian component.
And S407, carrying out parameter updating on the multi-resolution filter so that the output of the updated multi-resolution filter meets the expected loss value.
In the embodiment of the invention, as the weight matrix is added when the multi-resolution filter is constructed, the influence of the weight is also added in the expected loss estimation method adopted when the parameter of the multi-resolution filter is optimized:
Wherein pi l A priori weights, μ representing the first gaussian component l E chi is the mean of the Gaussian components, y 0 Outputting a value for a fixed response of the current hypothesis; s is S f {μ l [ mu ] is adopted l Outputting target correlation of the features; beta is a weight matrix for spatial position and W is a weight matrix for convolution filter.
According to the embodiment of the invention, a filter is initialized based on a first frame image in an image sequence, and for a subsequent frame image in the image sequence, a previous frame image and a current frame image are adopted to calculate the moving distance of a target between two frames by adopting a sparse optical flow algorithm; predicting the center point position of a target in the current frame image through the moving distance and the target position in the previous frame image, further extracting the characteristics of the image at the center point position, and estimating the target position in the current frame image through a preset tracking model; therefore, the characteristic dimension reduction is carried out by adopting the linear matrix in the initialization process of the multi-resolution filter, so that the parameter quantity of the multi-resolution filter is reduced, and the problem that the tracking efficiency is influenced due to slower updating caused by more parameters in the optimization process is further avoided; by adopting the sparse optical flow algorithm to calculate the target moving distance between two frames of images so as to perform motion modeling on the tracking target, the adopted images are the current frame image and the previous frame image, rather than only the target candidate region of the previous frame image, so that the problem that the target is lost due to the fact that the candidate region is adopted to perform motion modeling is avoided, and the candidate region containing the tracking target can be cut out from the current frame image more accurately.
Example III
Referring to fig. 5, fig. 5 shows a schematic structural diagram of a single-target tracking device based on sparse optical flow motion enhancement according to an embodiment of the present invention, as shown in fig. 5, where the device includes:
the motion modeling module 501 is configured to perform motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image, so as to determine a target candidate region image in the current frame image where the tracking target is located; wherein the current frame image is any frame image except the first frame image in the image sequence;
the feature extraction module 502 is configured to perform image feature extraction on the target candidate region image to obtain a multi-level resolution feature;
an estimating module 503, configured to input the multi-resolution feature into a preset tracking model, and estimate a position of the tracking target in the current frame image; the preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter.
In some embodiments, the motion modeling module 501 includes:
An acquisition sub-module, configured to acquire a previous frame image adjacent to the current frame image;
the first determining submodule is used for converting the previous frame image and the current frame image into gray level images, and determining angular point coordinates of corresponding pixel positions in the two frames of images by utilizing a Harris angular point detection algorithm;
the second determining submodule is used for determining the optical flow moving pixel distance between the corresponding characteristic angular points in the previous frame image and the current frame image based on the angular point coordinates;
a third determining submodule, configured to move a pixel distance according to the optical flow, and obtain a movement distance of the tracking target in two frames;
a fourth determining sub-module, configured to determine a center position of the target candidate region in the current frame image according to the predicted target position and the moving distance of the previous frame image;
and the clipping module is used for clipping the target candidate region image with a preset size from the current frame image based on the central position of the target candidate region.
In some embodiments, the apparatus further comprises:
a first updating module, configured to update the gaussian mixture model using the multi-resolution feature, so as to add the multi-resolution feature to a training sample set; the Gaussian mixture model is used for carrying out feature modeling on the multi-level resolution features so as to express feature distribution of the multi-level resolution features by adopting Gaussian components; the multi-level resolution feature takes the target correlation output of the multi-level resolution feature as a label;
A second estimating module, configured to estimate an expected loss value of the multi-resolution filter using the gaussian component at intervals of a preset number of frames;
and the second updating module is used for updating parameters of the multi-resolution filter so that the output of the updated multi-resolution filter meets the expected loss value.
Based on the same inventive concept, the embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor performs the steps in the sparse optical flow motion based single target tracking method according to any one of the embodiments.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described by differences from other embodiments, and identical and similar parts between the embodiments are all enough to be referred to each other.
It will be apparent to those skilled in the art that embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the invention may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.
The sparse optical flow motion enhancement-based single-target tracking method provided by the invention is described in detail above, and specific examples are applied to illustrate the principles and the implementation of the invention, and the description of the above examples is only used for helping to understand the method and the core idea of the invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.
Claims (8)
1. A sparse optical flow motion enhancement-based single-target tracking method, the method comprising:
performing motion modeling on a tracking target by using a sparse optical flow algorithm on a current frame image based on a previous frame image adjacent to the current frame image and the current frame image so as to determine a target candidate region image in which the tracking target is positioned in the current frame image; wherein the current frame image is any frame image except the first frame image in the image sequence;
extracting image features of the target candidate region image to obtain multi-level resolution features;
Inputting the multi-level resolution characteristic into a preset tracking model, and estimating the position of the tracking target in the current frame image; the preset tracking model comprises a multi-resolution filter and a linear matrix, wherein the linear matrix is used for reducing the parameter number of the multi-resolution filter.
2. The sparse optical flow motion enhanced single-target tracking method of claim 1, wherein the motion modeling of the tracked target using a sparse optical flow algorithm based on a previous frame image adjacent to the current frame image and the current frame image comprises:
acquiring a previous frame image adjacent to the current frame image;
converting the previous frame image and the current frame image into gray level images, and determining angular point coordinates of corresponding pixel positions in the two frames of images by using a Harris angular point detection algorithm;
determining the optical flow moving pixel distance between the corresponding characteristic angular points in the previous frame image and the current frame image based on the angular point coordinates;
determining the moving distance of the tracking target in two frames according to the optical flow moving pixel distance;
determining the central position of the target candidate region in the current frame image according to the predicted target position of the previous frame image and the moving distance;
And cutting out the target candidate region image with a preset size from the current frame image based on the center position of the target candidate region.
3. The sparse optical flow motion enhanced single-target tracking method of claim 2, wherein the determining the center position of the target candidate region in the current frame image from the predicted target position of the previous frame image and the distance of movement comprises:
determining the central coordinate of the predicted position of the tracking target in the previous frame of image;
determining the center point coordinates of the target candidate region in the current frame image based on the center coordinates and the moving distance;
the clipping the current frame image into the target candidate region image with a preset size based on the center position of the target candidate region includes:
and cutting out the target candidate region image with a preset size from the current frame image based on the center point coordinates of the target candidate region and a preset cutting frame.
4. The sparse optical flow motion enhanced-based single-target tracking method of claim 1, wherein the step of constructing the multi-resolution filter comprises:
Cutting out a candidate region with a preset size from a first frame image of the image sequence by taking a target truth value boundary frame as a center; the target truth value bounding box is the marked bounding box of the tracking target;
extracting image features aiming at the candidate areas to obtain the multi-level resolution features;
introducing an interpolation model, and mapping the multi-level resolution features to continuous spaces with the same continuous period so as to continuously carry out the multi-level resolution features;
defining a group of continuous convolution filters for the continuous space to obtain continuous convolution operators through parameterization;
determining a primary convolution filter set from the continuous convolution filters based on the linear matrix;
constructing a dimensionality-reduced convolution filter through mapping based on the main convolution filter set, and taking the dimensionality-reduced convolution filter as the multi-resolution filter;
and optimizing the multi-resolution filter by taking the first frame image as a preset sample to obtain the multi-resolution filter with the optimal parameter value.
5. The sparse optical flow motion enhanced-based single-target tracking method of claim 4, wherein optimizing the multi-resolution filter using the first frame image as a preset sample comprises:
Constructing a minimum loss function for the multi-resolution filter based on a preset sample pair; the preset sample pair is obtained by converting the first frame image; the minimization loss function employs L2 regularization to limit the values of the multi-resolution filter;
and iteratively optimizing the minimized loss function by adopting a Gaussian-Newton iteration and conjugate gradient method to obtain the optimal parameter value of the multi-resolution filter.
6. The sparse optical flow motion enhanced single-target tracking method of claim 1, wherein after said inputting the multi-resolution features into a preset tracking model to estimate the position of the tracking target in the current frame image, the method further comprises:
updating the Gaussian mixture model by adopting the multi-level resolution features to add the multi-level resolution features into a training sample set; the Gaussian mixture model is used for carrying out feature modeling on the multi-level resolution features so as to express feature distribution of the multi-level resolution features by adopting Gaussian components; the multi-level resolution feature takes the target correlation output of the multi-level resolution feature as a label;
A preset frame number is spaced, and the Gaussian component is adopted to estimate the expected loss value of the multi-resolution filter;
and updating parameters of the multi-resolution filter so that the updated output of the multi-resolution filter meets the expected loss value.
7. The sparse optical flow motion enhanced-based single-target tracking method of claim 6, wherein the estimating the expected loss value of the multi-resolution filter using the gaussian component comprises:
transforming the loss function of the multi-resolution filter based on the labels corresponding to the multi-resolution features in the training sample set to obtain an error expectation function of the multi-resolution filter;
substituting the Gaussian component into the error expectation function to obtain the expected loss value.
8. The sparse optical flow motion enhanced-based single-target tracking method of claim 1, wherein the preset tracking model comprises: the system comprises a mapping module, an acquisition module, a fusion module and an estimation module; the step of inputting the multi-resolution features into a preset tracking model, and estimating the position of the tracking target in the current frame image includes:
Inputting the multi-level resolution features into the mapping module to obtain continuous features;
inputting the continuous features to the acquisition module to acquire a target response graph at each resolution through the multi-resolution filter;
inputting the target response graphs under a plurality of resolutions to the fusion module to obtain a target response function;
the target response function is input to the estimation module to estimate a position of the tracking target in the current frame image based on a grid search method and conjugate gradient descent.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310495347.7A CN116543017A (en) | 2023-05-05 | 2023-05-05 | Single-target tracking method based on sparse optical flow motion enhancement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310495347.7A CN116543017A (en) | 2023-05-05 | 2023-05-05 | Single-target tracking method based on sparse optical flow motion enhancement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116543017A true CN116543017A (en) | 2023-08-04 |
Family
ID=87450020
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310495347.7A Pending CN116543017A (en) | 2023-05-05 | 2023-05-05 | Single-target tracking method based on sparse optical flow motion enhancement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116543017A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120370537A (en) * | 2025-06-23 | 2025-07-25 | 湖南省华芯医疗器械有限公司 | Method and device for determining rotation angle of endoscope, endoscope system and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160239982A1 (en) * | 2014-08-22 | 2016-08-18 | Zhejiang Shenghui Lighting Co., Ltd | High-speed automatic multi-object tracking method and system with kernelized correlation filters |
| CN115170621A (en) * | 2022-08-02 | 2022-10-11 | 西安奇维科技有限公司 | Target tracking method and system under dynamic background based on relevant filtering framework |
-
2023
- 2023-05-05 CN CN202310495347.7A patent/CN116543017A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160239982A1 (en) * | 2014-08-22 | 2016-08-18 | Zhejiang Shenghui Lighting Co., Ltd | High-speed automatic multi-object tracking method and system with kernelized correlation filters |
| CN115170621A (en) * | 2022-08-02 | 2022-10-11 | 西安奇维科技有限公司 | Target tracking method and system under dynamic background based on relevant filtering framework |
Non-Patent Citations (1)
| Title |
|---|
| 刘志青;李鹏程;陈小卫;张保明;郭海涛;: "基于信息向量机的机载激光雷达点云数据分类", 光学精密工程, no. 01, 15 January 2016 (2016-01-15) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120370537A (en) * | 2025-06-23 | 2025-07-25 | 湖南省华芯医疗器械有限公司 | Method and device for determining rotation angle of endoscope, endoscope system and storage medium |
| CN120370537B (en) * | 2025-06-23 | 2025-09-05 | 湖南省华芯医疗器械有限公司 | Method and device for determining rotation angle of endoscope, endoscope system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111080675B (en) | Target tracking method based on space-time constraint correlation filtering | |
| CN107680120B (en) | Infrared small target tracking method based on sparse representation and transfer restricted particle filter | |
| CN101398934B (en) | Method and system for tracking objects in images | |
| Bugeau et al. | Detection and segmentation of moving objects in complex scenes | |
| CN108198201A (en) | A kind of multi-object tracking method, terminal device and storage medium | |
| CN103559684B (en) | Based on the image recovery method of smooth correction | |
| CN110827262B (en) | Weak and small target detection method based on continuous limited frame infrared image | |
| CN110349186B (en) | Large-displacement motion optical flow calculation method based on depth matching | |
| CN118071932A (en) | A three-dimensional static scene image reconstruction method and system | |
| CN102087747B (en) | Object Tracking Method Based on Simplex Method | |
| CN115170621B (en) | A method and system for target tracking in dynamic background based on correlation filtering framework | |
| CN112991390B (en) | Multi-tracker fusion target tracking method based on background perception | |
| CN116543017A (en) | Single-target tracking method based on sparse optical flow motion enhancement | |
| CN114066942B (en) | A correlation filtering target tracking method based on manifold background perception | |
| CN113446998B (en) | Hyperspectral target detection data-based dynamic unmixing method | |
| CN115345902A (en) | Infrared image dim target detection tracking method and system based on machine learning | |
| CN118071793B (en) | An online target matching and tracking method based on local clipping | |
| CN111462181B (en) | Video single-target tracking method based on rectangular asymmetric inverse layout model | |
| CN120635109A (en) | A dynamic point cloud removal method, device and medium based on point cloud and image segmentation | |
| Istenic et al. | Mission-time 3D reconstruction with quality estimation | |
| CN118129692A (en) | A method and system for tracking the position of a target moving object | |
| CN118485690A (en) | Infrared small target variable-scale tracking method based on twin network | |
| CN110910425A (en) | Target tracking method for approaching flight process | |
| CN116993782B (en) | A target tracking method, device and medium based on asymmetric background perception | |
| CN115222771B (en) | A target tracking method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |