[go: up one dir, main page]

CN113033356B - A scale-adaptive long-term correlation target tracking method - Google Patents

A scale-adaptive long-term correlation target tracking method Download PDF

Info

Publication number
CN113033356B
CN113033356B CN202110265773.2A CN202110265773A CN113033356B CN 113033356 B CN113033356 B CN 113033356B CN 202110265773 A CN202110265773 A CN 202110265773A CN 113033356 B CN113033356 B CN 113033356B
Authority
CN
China
Prior art keywords
scale
target
frame
correlation
tracking method
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.)
Expired - Fee Related
Application number
CN202110265773.2A
Other languages
Chinese (zh)
Other versions
CN113033356A (en
Inventor
索继东
王思鹏
张伟红
柳晓鸣
陈晓楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian Maritime University
Original Assignee
Dalian Maritime University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Dalian Maritime University filed Critical Dalian Maritime University
Priority to CN202110265773.2A priority Critical patent/CN113033356B/en
Publication of CN113033356A publication Critical patent/CN113033356A/en
Application granted granted Critical
Publication of CN113033356B publication Critical patent/CN113033356B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种尺度自适应的长期相关性目标跟踪方法。首先，预处理第一帧图像，得到时间上下文回归模型Rc、目标外观回归模型Rt和检测器Drf。对于后续帧的跟踪，根据前一帧的目标位置创建搜索区域，并提取HOG特征，用以训练相关滤波器模板；进行平移估计，估计出当前帧的目标位置；然后，构建尺度池，自适应的估计出预测目标的最佳尺度，得到当前帧的目标状态；如果最大响应ys小于阈值τr，使用Drf执行重检测，更新目标的位置；接下来，更新Rc；如果最大响应ys大于阈值τa，更新Rt；然后，更新Drf；最后，得到当前帧预测的目标状态、Rc、Rt和Drf。重复以上步骤直到视频图像序列结束。本发明相比长期相关滤波(LCT)等算法提升了目标跟踪的性能，在多种复杂环境下鲁棒性更好。

The invention discloses a scale-adaptive long-term correlation target tracking method. First, preprocess the first frame image to obtain the temporal context regression model R c , the target appearance regression model R t and the detector D rf . For tracking of subsequent frames, a search area is created based on the target position of the previous frame, and HOG features are extracted to train the relevant filter template; translation estimation is performed to estimate the target position of the current frame; then, a scale pool is constructed to adaptively Estimate the best scale of the predicted target and obtain the target state of the current frame; if the maximum response y s is less than the threshold τ r , use D rf to perform re-detection and update the position of the target; next, update R c ; if the maximum response y When s is greater than the threshold τ a , R t is updated; then, D rf is updated; finally, the target state, R c , R t and D rf predicted by the current frame are obtained. Repeat the above steps until the video image sequence ends. Compared with algorithms such as long-term correlation filtering (LCT), the present invention improves the performance of target tracking and has better robustness in various complex environments.

Description

Scale-adaptive long-term correlation target tracking method
Technical Field
The invention belongs to the field of visual target tracking, and particularly relates to a scale self-adaptive long-term correlation target tracking method.
Background
Target tracking pertains to the content of video analysis, i.e., processing a sequence of video images. The task of target tracking is to determine the position and the size of the target in each subsequent frame by analyzing the group of video image sequences after the information such as the position and the size of the target in the first frame is given, and accurately frame the target. The target tracking technology integrates the knowledge of mathematics, physics, image processing and the like, and has wide application and development prospect in the aspects of military national defense, intelligent transportation and the like. For example, in the military field, for missile defense, guidance systems, air traffic control, and the like; the intelligent traffic system is used for real-time monitoring of traffic flow, traffic accident detection, pedestrian counting and the like in the intelligent traffic field.
The correlation filtering-based tracking algorithm sees the tracking process as a process of template matching and ridge regression. Nuclear correlation filtering (KCF) is a related filtering target tracking algorithm added with a kernel function, the algorithm extracts characteristics by using a multi-channel direction gradient Histogram (HOG), and positive and negative samples are constructed by cyclic shift, but the KCF cannot cope with the problem of target scale change.
In order to solve the problem, a Discriminant Scale Space (DSST) target tracking algorithm adopts a mode of jointly tracking a three-dimensional scale space related filter, firstly, position information of a target in a video sequence is determined by utilizing a two-dimensional discriminant position filter, and then, a tracking target output by the position filter is detected by utilizing a one-dimensional scale filter, so that the optimal scale of the current target is output. The long-term correlation filtering (LCT) algorithm adds a correlation filter for detecting confidence on the basis of a DSST algorithm position filtering and scale filtering framework, and the added detection mechanism enables the tracking performance of the algorithm in videos with attributes such as shielding and exceeding the field of view to be good, but the tracking performance of the LCT algorithm in environments with attributes such as overlarge target scale change, low resolution, rapid motion, illumination change and the like is required to be improved.
Disclosure of Invention
In view of this, the present invention provides a scale-adaptive long-term correlation (LCSA) target tracking algorithm to improve the accuracy of the existing target tracking algorithm, so that the target tracking algorithm can overcome the interference of multiple environmental factors.
For this purpose, the invention provides the following technical scheme:
a scale-adaptive long-term correlation target tracking algorithm, comprising the steps of:
(1) Initializing a target detection frame; extracting the characteristics of the target according to the detection frame in the first frame, and initializing a time context regression model R c Regression model R of target appearance t And detector D rf； wherein Rc Responsible for translation estimation, R t Responsible for scale estimation, D rf Responsible for re-detection;
(2) For the t-th frame, the target position (x t-1 ，y t-1 ) Cutting a search window in a t frame, extracting HOG characteristics, and training a relevant filter template;
(3) Performing translation estimation using R c And calculating a response based on the correlation filter score obtained from the correlation filter templateAnd estimates the current frame position +.>
(4) Constructing a scale pool by using a multi-scale search strategy and mapping y by correlation s and Rt Adaptive estimation of optimal dimensionsObtaining an initial predicted target state of the t-th frame +.>
(5) If it isUsing D rf Performing a re-detection to find a candidate set of states X, for each state X 'in X' i Calculating confidence score y' i If max (y' i )＞τ t Then-> wherein ,τr Is a first threshold value τ t Is a second threshold, ++>A correlation map representing the predictions; obtaining the final predicted target state of the t-th frame +.>
(6) Updating R c ；
(7) If it isUpdating R t； wherein ,τa Is a third threshold;
(8) Update D rf ；
(9) Repeating (2) - (8) for the t+1st frame until the video sequence ends.
Further, the relevant filter templates trained in the step (2) specifically include:
wherein w represents the correlation filter, x m,n Representing an image block x having m x n pixels, y (m, n) representing x m,n The gaussian sample tab generated as a training sample,represents the mapping to kernel space, λ represents the regularization parameters.
Further, the relevant filter templates trained in the step (2) specifically include:
wherein the coefficient a is defined by the following formula:f represents a discrete Fourier operator, x m,n The image block x is represented by m×n pixels, and x and y represent pixel coordinates.
Further, the tracking task is to calculate the correlation map by the image block z of the new frame in the image frames with the size of m×n, the response of the step (3)Determined according to the following formula:
where f represents the learned target appearance model, +.Finding the predicted position of the target.
Further, the scale search strategy of step (4) is:
the template scale is fixed as S T ＝(s x ,s y ) The scale pool is set to s= { t 1 ,t 2 ,…t k For the current frame, at { t } i s t |t i Sampling k dimensions in S to find the appropriate target scale, and then employing bilinear interpolation so that samples of each scale become equal to S T The samples are of uniform size;
the final target scale response value is calculated as follows:
is the ith scale sample in the scale pool, with the size of t i S t 。
Further, R is updated c Update R t Comprising the following steps:
for R c and Rt The coefficients f and a in the model are updated frame by frame at the learning rate α as:
further, D rf A support vector machine detector;
update D rf Comprising the following steps:
in each frame, a training set { (v) i ,c i ) I = 1,2,..n } and N samples, where v i Is the feature vector generated by the ith sample, c i E { +1, -1} is a sample tag, and the objective function for solving the support vector machine detector hyperplane h is:
wherein ,<h,v>represents the inner product of h and v; λ represents a regularization parameter;
updating hyperplane parameters using a passive algorithm:
wherein ,is the gradient of the loss function with respect to h, τ e (0, +_j) is a hyper-parameter that controls the h update rate.
Further, τ r Is 0.15 τ t Is 0.5 τ a 0.38.
The invention has the following beneficial effects:
according to the scale self-adaptive long-term correlation target tracking method provided by the invention, a scale self-adaptive strategy and an LCT target tracking frame are effectively fused, and firstly, a scale pool is introduced, so that an algorithm can self-adaptively select the optimal scale for finding the position of a tracking target. The multi-scale search can be combined with the position estimation filter more stably, and the situation that the scale estimation offset is too large is not easy to occur.
According to the scale self-adaptive long-term correlation target tracking method provided by the invention, compared with the tracking precision of an LCT algorithm, under classical target tracking scenes such as scale change, aspect ratio, low resolution, rapid motion, complete shielding, partial shielding, beyond-view, illumination change, view point transition, camera motion, similar objects and the like on Unmanned Aerial Vehicles (UAV 123) can be obviously improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of a scale-adaptive long-term correlation target tracking method in an embodiment of the invention;
FIG. 2 is a schematic diagram of a scale-adaptive long-term correlation target tracking method in an embodiment of the invention;
FIG. 3 is a schematic diagram of an adaptive scale model of a scale-adaptive long-term correlation target tracking method according to an embodiment of the present invention;
FIG. 4 is a graph of tracking accuracy of the present invention and other algorithms in a UAV123 dataset over 123 video sequences;
FIG. 5 is a graph of the success rate of the present invention and other algorithms in a UAV123 dataset over 123 video sequences;
FIG. 6 is a graph of tracking accuracy and success rate in the context of Scale Variation in a UAV123 dataset by the present invention and other algorithms;
FIG. 7 is a graph of tracking accuracy and success rate against UAV123 dataset Aspect Ratio Change (aspect ratio) background for the present invention and other algorithms;
FIG. 8 is a graph of tracking accuracy and success rate in the context of Low Resolution in the UAV123 dataset by the present invention and other algorithms;
FIG. 9 is a graph of tracking accuracy and success rate against Fast Motion (Fast Motion) background in a UAV123 dataset of the present invention and other algorithms;
FIG. 10 is a graph of tracking accuracy and success rate in the context of Full Occlusion in the UAV123 dataset by the present invention and other algorithms;
FIG. 11 is a graph of tracking accuracy and success rate against UAV123 dataset Partial Occlusion (partial occlusion) background for the present invention and other algorithms;
FIG. 12 is a graph of tracking accuracy and success rate in the context of the Out-of-View (Out-of-View) of the UAV123 dataset by the present invention and other algorithms;
FIG. 13 is a graph of tracking accuracy and success rate against UAV123 dataset Background Clutter (background clutter) background for the present invention and other algorithms;
FIG. 14 is a graph of tracking accuracy and success rate against UAV123 dataset illumination Variation (illumination variation) background for the present invention and other algorithms;
FIG. 15 is a graph of tracking accuracy and success rate in the context of a ViewPoint Change in UAV123 dataset by the present invention and other algorithms;
FIG. 16 is a graph of tracking accuracy and success rate against a Camera Motion background in a UAV123 dataset of the present invention and other algorithms;
FIG. 17 is a graph of tracking accuracy and success rate in the context of a Similar Object in the UAV123 dataset by the present invention and other algorithms.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Referring to fig. 1 and 2, two flowcharts of a scale-adaptive long-term correlation target tracking method according to an embodiment of the present invention are shown, respectively, and the method includes the following steps:
pretreatment: and reading the target position of the first frame to obtain a large background detection frame, a small background detection frame and Gaussian regression labels corresponding to each detection frame.
The processing procedure of the video of the first frame: extracting target direction gradient Histogram (HOG) characteristics according to the detection frame, and performing Fourier transform to obtain x f Adding a Gaussian kernel function to obtain a Gaussian response k in a frequency domain f Wherein the Gaussian kernel function isBoth regression models were mapped to kernel space, defined as k (x, x') =φ (x) ·φ(x'). Using x f and kf Calculating classifier parameters to obtain a time context regression model R c Regression model R of target appearance t And detector D rf, wherein Rc Responsible for translation estimation, R t Responsible for scale estimation, D rf Is responsible for re-detection.
The processing procedure of the current frame (t-th frame):
(1) According to the target position (x) of the t-1 th frame t-1 ，y t-1 ) Cutting a search window in a t frame, extracting HOG characteristics, and training a relevant filter template;
the training related filter template specifically comprises the following components:
wherein w represents the correlation filter, x m,n Representing an image block x having M x N pixels, y (M, N) representing x m,n The gaussian sample tab generated as a training sample,represents the mapping to kernel space, λ represents the regularization parameters.
The above can also be written as w= Σ m,n a(m,n)φ(x m,n )。
Wherein the coefficient a is defined by the following formula:f represents a discrete Fourier operator, x m,n The image block x is represented by m×n pixels, and x and y represent pixel coordinates.
(2) Translation estimation using context regression model R c Correlation filter score calculation response y t And estimates the position (x) t ，y t )；
The tracking task may calculate the correlation map by image blocks z of a new one of the image frames of size m x n. The response of step (2) may be determined according to the following equation:
where f represents the learned target appearance model, +.Finding the predicted position of the target.
(3) As shown in FIG. 3, a scale pool is built using a multi-scale search strategy, by correlation mapping y s And a target appearance regression model R t Adaptive estimation of optimal dimensionsObtaining an initial predicted target state of the t-th frame +.>
The scale search strategy is as follows: the template scale is fixed as S T ＝(s x ,s y ) The scale pool is set to s= { t 1 ,t 2 ,…t k For the current frame, at { t } i s t |t i The k dimensions are sampled in S to find the appropriate target scale, and bilinear interpolation is then employed so that the samples of each scale become equal to S T Samples were of uniform size.
The final target scale response value is calculated as follows: is the ith scale sample in the scale pool, with the size of t i S t . By bilinear interpolation, < > on->Will be adjusted to S T 。
(4) Setting a first threshold value tau r Second threshold τ t ，A correlation map representing the predictions; if->Using D rf Performing re-detection to find a candidate state set X, for each state X in X i 'calculate confidence score y' i If max (y' i )＞τ t Then->Obtaining the final predicted target state of the t-th frame +.>
(5) Updating the model R c ；
(6) Setting a third threshold value tau a If (3)Updating the model R t ；
In the steps (5) and (6), R is as follows c 、R t The model updates the coefficients f and A in the model frame by frame at the learning rate alpha as:
(7) Update detector D rf ；
In the embodiment of the invention, D rf Is a Support Vector Machine (SVM) detector. For SVM, a training set { (v) is given in each frame i ,c i ) I = 1,2,..n } and N samples, where v i Is the feature vector generated by the ith sample, c i E { +1, -1} is then the sample label, and the objective function for solving the SVM detector hyperplane h is:
wherein ,<h,v>representing the inner product of h and v.
Passive algorithms are used to efficiently update hyperplane parameters:
wherein ,is the gradient of the loss function with respect to h, τ e (0, +_j) is a hyper-parameter that controls the h update rate.
Obtaining final predicted target state of the t-th frame from steps (2) - (4)Obtaining R of the current frame from steps (5) - (7) c 、R t and Drf ；
(8) Repeating (1) - (7) for the t+1st frame until the video sequence ends.
According to the scale self-adaptive long-term correlation target tracking method provided by the embodiment of the invention, a scale self-adaptive strategy and an LCT target tracking frame are effectively fused, and firstly, a scale pool is introduced, so that an algorithm can self-adaptively select the optimal scale for finding the position of a tracking target. The multi-scale search can be combined with the position estimation filter more stably, and the situation that the scale estimation offset is too large is not easy to occur.
Based on the above embodiments, this embodiment provides a simulation experiment.
Simulation conditions: the simulation provided in this example was at Intel (R) Core (TM) i3-4170CPU@3.70GHz 3.70GHz, a hardware environment with 4.00GB memory, and a software environment with MATLAB R2016 a. The experimental parameters were set as follows: regularization parameter λ=10 -4 Gaussian kernel σ=0.1, learning rate α=0.01, threshold τ r ＝0.15，τ a ＝0.38，τ t =0.5, scale pool set to [1,0.99,1.01 ]]. The algorithm presented herein is then compared to LCTs and other existing classical target tracking algorithms.
The simulation content: the proposed method is evaluated on a large reference data set UAV-123 containing 123 videos, and the evaluation mode selects one pass success rate (OPE), i.e. the target position given by the first frame starts tracking, and the tracking is not reinitialized after failure.
Fig. 4 to 17 are graphs of experimental results of the present experiment, in which LCSA represents the scale-adaptive long-term correlation filter tracking method proposed by the present invention, LCT and kcf_ GaussHog, CSK, IVT, DFT represent other excellent target tracking algorithms, respectively. The LCSA algorithm provided by the invention has a score of 0.40 in the graph of fig. 5 versus the success rate curve and a score of 0.58 in the graph of fig. 4 versus the accuracy curve. With the long-term correlation tracking algorithm (LCT) as a reference algorithm, the LCSA can be obtained from experimental data on the UAV-123 to be improved by 2.56% in AUC success rate and 5.26% in accuracy compared with the LCT. Although the accuracy is slightly lower than LCT in the case of background clutter on UAV123, there is a performance improvement over classical target tracking algorithms such as LCT in classical target tracking scenarios such as scale change, aspect ratio, low resolution, fast motion, complete occlusion, partial occlusion, out-of-view, illumination change, field of view point transition, camera motion, similar objects, etc.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.

Claims (8)

1.一种尺度自适应的长期相关性跟踪方法，其特征在于，包括如下步骤：1. A scale-adaptive long-term correlation tracking method, characterized by including the following steps: (1)初始化目标检测框；在第一帧根据所述检测框提取目标的特征，初始化时间上下文回归模型Rc、目标外观回归模型Rt和检测器Drf；其中Rc负责平移估计，Rt负责尺度估计，Drf负责重检测；(1) Initialize the target detection frame; extract the characteristics of the target according to the detection frame in the first frame, and initialize the temporal context regression model R c , the target appearance regression model R t and the detector D rf ; where R c is responsible for translation estimation, R t is responsible for scale estimation, and D rf is responsible for re-detection; (2)针对第t帧，根据第t-1帧的目标位置(xt-1，yt-1)，在第t帧中裁剪搜索窗口，并提取HOG特征，训练相关滤波器模板；(2) For the t-th frame, according to the target position (x t-1 , y t-1 ) of the t-1th frame, crop the search window in the t-th frame, extract HOG features, and train the relevant filter template; (3)进行平移估计，使用Rc和基于所述相关滤波器模板得到的相关滤波分数计算响应并估计当前帧位置/> (3) Perform translation estimation and calculate the response using R c and the correlation filter score obtained based on the correlation filter template. And estimate the current frame position/> (4)使用多尺度搜索策略，构建尺度池，通过相关映射ys和Rt自适应的估计出最佳尺度得到第t帧的初始预测目标状态/> (4) Use a multi-scale search strategy to construct a scale pool, and adaptively estimate the best scale through correlation mapping y s and R t Get the initial predicted target state of frame t/> (5)如果使用Drf执行重检测，找到候选状态集合X，对X中的每个状态x′i计算置信分数y′i，如果max(y′i)＞τt，则/>i＝arg maxiy′i；其中，τr为第一阈值，τt为第二阈值，/>表示预测的相关映射；得到第t帧的最终预测目标状态/> (5)If Use D rf to perform re-detection, find the candidate state set X, and calculate the confidence score y′ i for each state x′ i in X. If max(y′ i ) >τ t , then/> i＝arg max i y′ i ; where, τ r is the first threshold, τ t is the second threshold,/> Represents the predicted correlation mapping; obtains the final predicted target state of the tth frame/> (6)更新Rc；(6) Update R c ; (7)如果更新Rt；其中，τa为第三阈值；(7)If Update R t ; where, τ a is the third threshold; (8)更新Drf；(8) Update D rf ; (9)针对第t+1帧，重复(2)～(8)，直到视频序列结束。(9) For the t+1th frame, repeat (2) to (8) until the end of the video sequence. 2.根据权利要求1所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，所述步骤(2)中训练的相关滤波器模板具体为：2. A scale-adaptive long-term correlation tracking method according to claim 1, characterized in that the correlation filter template trained in step (2) is specifically: 其中，w表示相关滤波器，xm,n表示图像块x有m*n个像素，y(m,n)表示以xm,n作为训练样本生成的高斯样本标签，表示到内核空间的映射，λ表示正则化参数。Among them, w represents the correlation filter, x m,n represents that the image block x has m*n pixels, y(m,n) represents the Gaussian sample label generated with x m,n as the training sample, represents the mapping to the kernel space, and λ represents the regularization parameter. 3.根据权利要求1所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，所述步骤(2)中训练的相关滤波器模板具体为： 3. A scale-adaptive long-term correlation tracking method according to claim 1, characterized in that the correlation filter template trained in step (2) is specifically: 其中系数a由下式定义：F表示离散傅立叶算子，xm,n表示图像块x有m*n个像素，x和y表示像素点坐标。The coefficient a is defined by the following formula: F represents the discrete Fourier operator, x m, n represents that the image block x has m*n pixels, and x and y represent the pixel point coordinates. 4.根据权利要求3所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，跟踪任务是通过大小为m*n的图像帧中的新一帧的图像块z来计算相关映射，所述步骤(3)的响应根据下式确定：4. A scale-adaptive long-term correlation tracking method according to claim 3, characterized in that the tracking task is to calculate the correlation mapping through the image block z of a new frame in the image frame of size m*n. , the response to step (3) Determine according to the following formula: 其中f表示学习的目标外观模型，⊙表示Hadamard乘积，通过找到的最大值找到目标的预测位置。where f represents the learned target appearance model, ⊙ represents the Hadamard product, and is found by The maximum value of finds the predicted position of the target. 5.根据权利要求1所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，所述步骤(4)的尺度搜索策略为：5. A scale-adaptive long-term correlation tracking method according to claim 1, characterized in that the scale search strategy of step (4) is: 模板尺度固定为ST＝(sx,sy)，尺度池设置为S＝{t1,t2,…tk}，对于当前帧，在{tist|ti∈S}中对k个尺寸进行采样以找到合适的目标尺度，然后采用双线性插值，使得各个尺度的样本变成与ST样本一致的大小；The template scale is fixed to S T = (s x , s y ), and the scale pool is set to S = {t 1 , t 2 ,...t k }. For the current frame, in {t i s t |t i ∈S} Sampling k sizes to find the appropriate target scale, and then using bilinear interpolation to make the samples of each scale become the same size as the S T sample; 最终的目标尺度响应值计算如下： The final target scale response value is calculated as follows: 是尺度池中的第i个尺度样本，大小为tiSt。 is the i-th scale sample in the scale pool, with size t i S t . 6.根据权利要求3所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，更新Rc、更新Rt包括：6. A scale-adaptive long-term correlation tracking method according to claim 3, characterized in that updating R c and updating R t includes: 对于Rc和Rt，以学习速率α逐帧更新模型中的系数f和A为：For R c and R t , updating the coefficients f and A in the model frame by frame at the learning rate α is: 7.根据权利要求1所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，Drf为支持向量机检测器；7. A scale-adaptive long-term correlation tracking method according to claim 1, characterized in that D rf is a support vector machine detector; 更新Drf包括：Updated D rf includes: 在每一帧中，给定一个训练集{(vi,ci)|i＝1,2,...,N}和N个样本，其中vi是第i个样本生成的特征向量，ci∈{+1,-1}是样本标签，求解支持向量机检测器超平面h的目标函数是：In each frame, given a training set {( vi , ci )|i=1,2,...,N} and N samples, where vi is the feature vector generated by the i-th sample, c i ∈{+1,-1} is the sample label, and the objective function to solve the support vector machine detector hyperplane h is: 其中，l(h；(v,c))＝max{0,1-c<h,v>}，<h,v>表示h与v的内积；λ表示正则化参数；Among them, l(h; (v, c))=max{0,1-c<h,v>}, <h,v> represents the inner product of h and v; λ represents the regularization parameter; 使用被动算法更新超平面参数：Update hyperplane parameters using a passive algorithm: 其中，是损失函数关于h的梯度，τ∈(0，∞)是控制h更新速率的超参数。in, is the gradient of the loss function with respect to h, and τ∈(0,∞) is a hyperparameter that controls the update rate of h. 8.根据权利要求1所述的一种尺度自适应的长期相关性跟踪方法，其特征在于，τr为0.15，τt为0.5，τa为0.38。8. A scale-adaptive long-term correlation tracking method according to claim 1, characterized in that τ r is 0.15, τ t is 0.5, and τ a is 0.38.
CN202110265773.2A 2021-03-11 2021-03-11 A scale-adaptive long-term correlation target tracking method Expired - Fee Related CN113033356B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110265773.2A CN113033356B (en) 2021-03-11 2021-03-11 A scale-adaptive long-term correlation target tracking method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110265773.2A CN113033356B (en) 2021-03-11 2021-03-11 A scale-adaptive long-term correlation target tracking method

Publications (2)

Publication Number Publication Date
CN113033356A CN113033356A (en) 2021-06-25
CN113033356B true CN113033356B (en) 2023-10-17

Family

ID=76469899

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110265773.2A Expired - Fee Related CN113033356B (en) 2021-03-11 2021-03-11 A scale-adaptive long-term correlation target tracking method

Country Status (1)

Country Link
CN (1) CN113033356B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113902773A (en) * 2021-09-24 2022-01-07 南京信息工程大学 Long-term target tracking method using double detectors
CN114972443A (en) * 2022-06-28 2022-08-30 深圳一清创新科技有限公司 Target tracking method and device and unmanned vehicle
CN117292306A (en) * 2023-11-27 2023-12-26 四川迪晟新达类脑智能技术有限公司 Edge equipment-oriented vehicle target detection optimization method and device

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110569723A (en) * 2019-08-02 2019-12-13 西安工业大学 A Target Tracking Method Combining Feature Fusion and Model Update
KR102175491B1 (en) * 2019-05-08 2020-11-06 군산대학교산학협력단 Method and apparatus for tracking object based on correlation filter

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102175491B1 (en) * 2019-05-08 2020-11-06 군산대학교산학협력단 Method and apparatus for tracking object based on correlation filter
CN110569723A (en) * 2019-08-02 2019-12-13 西安工业大学 A Target Tracking Method Combining Feature Fusion and Model Update

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于多特征和尺度估计的核相关滤波跟踪算法;张伟;温显斌;;天津理工大学学报(第03期);全文 *

Also Published As

Publication number Publication date
CN113033356A (en) 2021-06-25

Similar Documents

Publication Publication Date Title
CN108986140B (en) Target scale self-adaptive tracking method based on correlation filtering and color detection
CN107358623B (en) A Correlation Filter Tracking Method Based on Saliency Detection and Robust Scale Estimation
CN106981071B (en) A Target Tracking Method Based on Unmanned Boat Application
CN113033356B (en) A scale-adaptive long-term correlation target tracking method
CN111914664A (en) Vehicle multi-target detection and trajectory tracking method based on re-identification
CN109146912B (en) Visual target tracking method based on target analysis
CN103971386A (en) Method for foreground detection in dynamic background scenario
CN109977971A (en) Dimension self-adaption Target Tracking System based on mean shift Yu core correlation filtering
CN103729854A (en) Tensor-model-based infrared dim target detecting method
US11821986B1 (en) Target tracking method, system, device and storage medium
CN109166139B (en) A Scale-Adaptive Object Tracking Method Combined with Fast Background Suppression
CN110706253B (en) Target tracking method, system and device based on apparent feature and depth feature
CN107368802B (en) A moving target tracking method based on KCF and human brain memory mechanism
CN112785622A (en) Long-time tracking method and device for unmanned ship on water surface and storage medium
Xing et al. Feature adaptation-based multipeak-redetection spatial-aware correlation filter for object tracking
CN110751671A (en) A Target Tracking Method Based on Kernel Correlation Filtering and Motion Estimation
CN110135435B (en) Saliency detection method and device based on breadth learning system
CN109448024B (en) Visual tracking method and system for constructing constraint correlation filter by using depth data
CN110827327B (en) Fusion-based long-term target tracking method
Chen et al. Single‐Object Tracking Algorithm Based on Two‐Step Spatiotemporal Deep Feature Fusion in a Complex Surveillance Scenario
CN110472607A (en) A kind of ship tracking method and system
CN110827319A (en) An Improved Staple Target Tracking Method Based on Local Sensitive Histogram
Huang et al. Robust Visual Tracking Models Designs Through Kernelized Correlation Filters.
CN111681266A (en) Ship tracking method, system, device and storage medium
CN111951298A (en) A target tracking method fused with time series information

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
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20231017

CF01 Termination of patent right due to non-payment of annual fee