CN115272409A - Single-target long-time tracking method based on deep neural network - Google Patents

Single-target long-time tracking method based on deep neural network Download PDF

Info

Publication number
CN115272409A
CN115272409A CN202210831023.1A CN202210831023A CN115272409A CN 115272409 A CN115272409 A CN 115272409A CN 202210831023 A CN202210831023 A CN 202210831023A CN 115272409 A CN115272409 A CN 115272409A
Authority
CN
China
Prior art keywords
target
tracking
network
score
scale
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
Application number
CN202210831023.1A
Other languages
Chinese (zh)
Inventor
黄飞
彭蓉华
宁平
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.)
Changsha Chaochuang Electronic Technology Co ltd
Original Assignee
Changsha Chaochuang Electronic Technology Co ltd
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 Changsha Chaochuang Electronic Technology Co ltd filed Critical Changsha Chaochuang Electronic Technology Co ltd
Priority to CN202210831023.1A priority Critical patent/CN115272409A/en
Publication of CN115272409A publication Critical patent/CN115272409A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • 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
    • G06V10/54—Extraction of image or video features relating to texture
    • 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
    • G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30241—Trajectory

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to the technical field of target tracking, and discloses a single-target long-time tracking method based on a deep neural network, which comprises the following steps: acquiring a tracking target image and a tracking search area image, constructing a twin network-based feature extraction network, and extracting features; in the tracking process, selecting a target image through PNR (portable network noise ratio) scores, adding the target image into a template library, sending the target image into a model learning network for model learning and updating, convolving a model target feature map and a search feature map to obtain a response map, sending the response map into a scale regression network to obtain a scale position score map, and obtaining target scales; when the target is blocked or lost, target recapturing is carried out; and in the tracking process, recording the information of the target and the suspected target, and selecting a tracking position which is most matched with the historical track when an interfering object occurs. The invention reduces the model drift phenomenon; the scale estimation is more stable and accurate; and a re-searching mechanism and an anti-interference matching technology are adopted, so that the algorithm is suitable for long-time target tracking.

Description

Single-target long-time tracking method based on deep neural network
Technical Field
The invention belongs to the technical field of visual target tracking, and particularly relates to a single-target long-time tracking method based on a deep neural network.
Background
The single-target tracking refers to tracking a specified target in a video, wherein the target is specified by people at the beginning, and the tracking algorithm estimates the position and the scale of the specified target by utilizing the context information of the video or an image sequence. The single target tracking algorithm has wide application fields, including intelligent video monitoring, unmanned driving, the field of robots, the field of unmanned aerial vehicles and the like. The application scenes of the target tracking algorithm are diversified, for example, in shopping malls with a lot of people, roads, highways, sky and the like, so that the target tracking algorithm faces some difficult challenges, such as illumination change, target deformation, occlusion, complex background and the like.
At present, in a tracking algorithm based on a deep neural network, the tracking algorithm based on a twin network has good performance in speed and accuracy and is widely researched. Attention-based methods, RPN-network-scale-estimation-based methods, spatio-temporal modeling-based methods, etc. have emerged. However, in the real long-term tracking process, the appearance of the target usually changes dynamically, even changes rapidly, the target is shielded, and the like.
Disclosure of Invention
Aiming at the existing problems, the invention provides a template base updating and scale regression network scale estimation based on RMAD (namely PNR value) and an anti-interference long-time tracking algorithm based on Hungarian matching algorithm so as to improve the tracking accuracy and robustness of the tracking algorithm under the conditions of target deformation, target occlusion and the like.
The invention discloses a single-target long-time tracking method based on a deep neural network, which comprises the following steps:
acquiring a tracking target image and a tracking search area image, and constructing a twin network-based feature extraction network for feature extraction;
an online learning model: in the tracking process, selecting a target image through PNR score, adding the target image into a template library, and then sending the target image into a model learning network for model learning and updating;
and (3) target scale estimation: convolving the model target characteristic diagram and the search characteristic diagram to obtain a response diagram, sending the response diagram into a scale regression network to obtain a scale position score diagram, and obtaining a target scale;
target recapture: in the tracking process, the state of the target is judged, whether the target is shielded or lost is judged, and when the target is lost, the target searching range is expanded, and target recapture is carried out;
anti-interference matching and tracking: in the tracking process, the position, the scale, the confidence coefficient and the textural feature information of the target and the suspected target are recorded, when an interfering object occurs, the Hungarian matching algorithm is adopted to track and match the currently estimated tracking position and the historical tracking track, and the tracking position which is most matched with the historical track is selected, so that the influence of the interfering object is eliminated, and the long-term tracking of the target is guaranteed.
Further, the twin network employs the ResNet50 as a backbone network.
Further, the loss function of the model learning network is:
Figure BDA0003745567660000021
Figure BDA0003745567660000022
Figure BDA0003745567660000023
wherein v isjIs a feature map of the training image, cjRepresenting the coordinates of the centre of the object, the purpose being to find the model f, the object image being represented by AcTo indicate that a is a background pixel when a pixel in the image is a background pixel c0, when A is the targetc1,s =vf, representing target confidence score, ycScore, y, representing each coordinate positioncInitialization with a Gaussian distribution, yc,Ac,ηcThe parameters are learned by on-line training.
Further, the calculation formula of the PNR is as follows:
Figure BDA0003745567660000031
Figure BDA0003745567660000032
wherein S'maxRepresents the maximum value, S 'in the classification score response map S'minRepresents the minimum value, M, in the classification score response map SSDenotes the mean value of S' after removal of the maximum response value, TPNRIs the mean of all historical PNR values, TS′Is all historical S'maxIs measured.
Further, the scale regression network is trained according to the following algorithm:
the loss function is:
Figure BDA0003745567660000033
LCIOU＝-ln(CIOU)
Figure BDA0003745567660000034
wherein L isCIOUIs the CIOU loss in terms of (x)l,yl) The top left corner of the real target box is represented by (x)r,yr) The lower right corner representing the real target frame, (x)c,yc) Represents the coordinates of the center point, then
Figure BDA0003745567660000035
Figure BDA0003745567660000036
Figure BDA0003745567660000037
Figure BDA0003745567660000038
And obtaining the highest-score target position according to the classification score map, and taking the mean value of top-k target frames around the highest-score target position as the frame of the final target.
Further, the target state judgment is carried out according to the classification score of the target: when the score is less than 0.25, judging that the target is blocked or the target is not found currently, starting a counter, when the counter is more than 5, indicating that the target is lost, expanding the search area by 1 time every 2 frames according to the size of the target, searching the target by using a classification network, and when the score of the target is more than 0.8, considering that the target is found; when the search time exceeds a certain number of frames, the target is considered lost.
Further, when a plurality of suspected targets appear in the tracking process, information of the suspected targets is stored, wherein the information comprises the position of the center point, the scale, the classification confidence coefficient and the SILTP characteristics.
Further, for a given location (x)c,yc) The coding formula of the SILTP characteristic is as follows:
Figure BDA0003745567660000041
wherein g iscGray value of the central pixel, gkFor the other k pixels in a circle of radius R within the central pixel field, s is the piecewise function, e =0.3:
Figure BDA0003745567660000042
further, the cost in the Hungarian matching algorithm is calculated by cosine similarity as similarity between vectors,
Figure BDA0003745567660000043
wherein objiThe vector is composed of the position, the scale, the classification confidence coefficient and the SILTP characteristic value of the center point of the suspected target, trTThe vector is the center point position, the scale, the classification confidence coefficient and the SILTP characteristic value of the nearest point of the tracking track.
Compared with the prior art, the invention has the following beneficial effects:
the template base is updated by adding a reliable target template into the template base according to the PNR value of the classification score response graph, so that the model drift phenomenon is reduced;
the scale estimation adopts a scale regression network to ensure that the scale is more stable and accurate;
a long-term tracking strategy of the target is designed, and after the target is lost, the dynamic expansion of the search range and the re-identification are started;
and an anti-interference matching algorithm is adopted, so that similar interference is reduced, and the tracking algorithm can track the target for a long time.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a classification score plot for the occurrence of an analog;
FIG. 3 is a block diagram of the deep neural network of the present invention;
FIG. 4 is a graph of the tracking results of the present invention.
Detailed Description
The present invention is further described with reference to the drawings, but the present invention is not limited thereto in any way, and any modifications or alterations based on the teaching of the present invention shall fall within the scope of the present invention.
In order to achieve the purpose, the invention adopts the following technical scheme:
s1, acquiring a tracking target image and a tracking search area image, and constructing a twin network-based feature extraction network for feature extraction.
S2, an online learning model: in the tracking process, a target image is selected through the PNR score and added into a template library, and then the target image is sent into a model learning network for model learning and updating.
S3, target scale estimation: and (4) convolving the model target characteristic diagram and the search characteristic diagram to obtain a response diagram, and sending the response diagram into a scale regression network to obtain a scale position score diagram and obtain a target scale.
S4, target recapture: and in the tracking process, judging the state of the target, judging whether the target is shielded or lost, and enlarging the target searching range to perform target recapture when the target is lost.
S5, anti-interference matching tracking: in the tracking process, the positions, the dimensions, the confidence degrees and the texture feature information of the target and the suspected target are recorded, when an interfering object occurs, the Hungarian matching algorithm is adopted to track and match the currently estimated tracking position and the historical tracking track, and the tracking position which is most matched with the historical track is selected, so that the influence of the interfering object is eliminated, and the long-term tracking of the target is guaranteed.
The invention is further described below with reference to the accompanying drawings.
Example 1
As shown in FIG. 1, the invention constructs a single-target long-time tracking algorithm based on an RMAD deep neural network, and the basic idea is to extract the depth characteristics of a target and a search area by using a twin network, establish a template library for the target characteristics, obtain the target template characteristics by adopting an online learning model method, convolve the target template characteristics and the search area characteristics to obtain a classification score map, and obtain the position information of the target. Then, determining the target size by using a scale regression network; and by a tracking state judgment mechanism, the search range is expanded to search again when the target is lost. When the interferents appear, tracking and matching are carried out by adopting a Hungarian matching algorithm, and the influence of the interferents is eliminated.
The method specifically comprises the following steps:
s1, constructing a feature extraction network based on a twin network: the resnet50 is used as a backbone network, the input of the network is composed of a tracking target image and a tracking search area image, and a characteristic diagram of the tracking target and a characteristic diagram of a search area are obtained after the input of the network.
S2, after the result of the first step is obtained, sending the feature map into a classification network to obtain a classification score map: the classification scoring network consists of an online learning model network and a related convolution network. The model network input of online learning is a template set with 50 frames of target feature templates, the template set comprises 15 frames of initial target feature templates and target feature templates in the subsequent tracking process, and online learning is performed every 20 frames. The model network for online learning is specifically trained according to the following algorithm:
1) It employs a loss function of
Figure BDA0003745567660000061
Figure BDA0003745567660000062
Figure BDA0003745567660000071
Wherein v isjIs a feature map of the training image, cjRepresenting the target center coordinates. The objective is to find the model f, A for the target imagecTo indicate that a is a background pixel when a pixel in the image is a background pixel c0, when A is the targetc1.s = v f, representing the target confidence score, ycScore, y, representing each coordinate positioncThe initialization is performed using a gaussian distribution. Wherein y isc,Ac,ηcThe parameters are learned by on-line training.
And then judging whether the target characteristic image is added into the template library or not according to the PNR peak signal-to-noise ratio score of the classification score response graph S'. When the PNR value is larger than the threshold value TPNRAnd the classification score is also greater than the threshold TS′And adding the current target feature graph into the template library. When the template library is full, the oldest template except the initial template is replaced, and the calculation formula of PNR is as follows:
Figure BDA0003745567660000072
Figure BDA0003745567660000073
wherein S'maxRepresents the maximum value, S ', in the classification score response map S'minRepresents the minimum value, M, in the classification score response map SSDenotes the mean value of S' after removal of the maximum response value, TPNRIs the mean of all historical PNR values, TS′Is all historical S'maxIs measured.
S3, calculating a rectangular frame of the tracking target through a scale regression network: obtaining the model target characteristic diagram in the step (2), carrying out convolution on the model target characteristic diagram and the search characteristic diagram to obtain a response diagram, and sending the response diagram into a scale regression network to obtain a four-dimensional position scoring diagram Aw×h×4Where each point is composed of a four-dimensional vector, and t (i, j) = (l, t, r, b) represents the distance from the target center point (i, j) to the four sides of the target rectangular frame. The scale regression network is specifically trained according to the following algorithm:
having a loss function of
Figure BDA0003745567660000074
LCIOU＝-ln(CIOU)
Figure BDA0003745567660000081
Wherein L isCIOUIs the CIOU loss in terms of (x)l,yl) The top left corner of the real target box is represented by (x)r,yr) The lower right corner representing the real target frame, (x)c,yc) Represents the coordinates of the center point, then
Figure BDA0003745567660000082
And obtaining the position of the target with the highest score according to the classification score map, and taking the mean value of top-k target frames around the position of the target as the frame of the final target.
S4, judging the target state according to the classification score of the target: and when the score is less than 0.25, judging that the target is blocked or the target is not found currently. And starting a counter, when the counter is more than 5, indicating that the target is lost, expanding the search area by 1 time every 2 frames according to the size of the target, searching the target by using a classification network, and when the target score is more than 0.8, considering that the target is found. When the search time exceeds a certain number of frames, the target is considered lost.
S5, anti-interference matching: when a plurality of peak points higher than a certain threshold appear on the classification score map, as shown in fig. 2, it indicates that a plurality of suspected targets appear in the tracking process, and at this time, the information obj of the suspected targets is usediAll the information is stored, the information comprises the position of a central point, the scale, the classification confidence coefficient and the SILTP characteristic, and the Hungarian matching algorithm is used for tracking the historical track tr of the targetT(p1,p2,…pT) And matching to obtain a final tracking result. The SILTP (Scale Invariant Local texture Pattern) operator is an improved version of the LBP operator, has stronger robustness to noise and illumination change of a Local image, and has stronger texture recognition characteristic. For a given position (x)c,yc) The coding formula of SILTP is：
Figure BDA0003745567660000083
Wherein g iscGray value of the central pixel, gkThe other N pixels of R (circle of radius) within the central pixel domain, s is the piecewise function, e =0.3:
Figure BDA0003745567660000091
the method for extracting the target texture features comprises the following steps: and (3) taking a target image, carrying out gray level conversion, calculating an SILTP value of each pixel point in the target image, and adopting 4 neighborhoods for anticlockwise calculation. The cost in Hungarian matching is calculated by cosine similarity, and the formula is as follows:
Figure BDA0003745567660000092
wherein objiThe vector is the information of the suspected target (center point position, scale, classification confidence, SILTP eigenvalue), trTThe vector is the information of the nearest point of the tracking trajectory (center point position, scale, classification confidence, SILTP feature value).
Compared with the prior art, the invention has the following beneficial effects:
the template base is updated by adding a reliable target template into the template base according to the PNR value of the classification score response graph, so that the model drift phenomenon is reduced;
the scale estimation adopts a scale regression network to ensure that the scale is more stable and accurate;
a long-term tracking strategy of the target is designed, and after the target is lost, the dynamic expansion of the search range and the re-identification are started;
and an anti-interference matching algorithm is adopted, so that similar interference is reduced, and a tracking algorithm can track the target for a long time.
The word "preferred" is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, use of the word "preferred" is intended to present concepts in a concrete fashion. The term "or" as used in this application is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless specified otherwise or clear from context, "X employs A or B" is intended to include either of the permutations as a matter of course. That is, if X employs A; b is used as X; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.
Also, although the disclosure has been shown and described with respect to one or an implementation, equivalent alterations and modifications will occur to others skilled in the art based upon a reading and understanding of this specification and the annexed drawings. The present disclosure includes all such modifications and alterations, and is limited only by the scope of the appended claims. In particular regard to the various functions performed by the above described components (e.g., elements, etc.), the terms used to describe such components are intended to correspond, unless otherwise indicated, to any component which performs the specified function of the described component (e.g., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary implementations of the disclosure. In addition, while a particular feature of the disclosure may have been disclosed with respect to only one of several implementations, such feature may be combined with one or other features of the other implementations as may be desired and advantageous for a given or particular application. Furthermore, to the extent that the terms "includes," has, "" contains, "or variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term" comprising.
Each functional unit in the embodiments of the present invention may be integrated into one processing module, or each unit may exist alone physically, or a plurality of or more than one unit are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium. The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. Each apparatus or system described above may execute the storage method in the corresponding method embodiment.
In summary, the above-mentioned embodiment is an implementation manner of the present invention, but the implementation manner of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be regarded as equivalent replacements which are included in the protection scope of the present invention.

Claims (9)

1. A single-target long-time tracking method based on a deep neural network is characterized by comprising the following steps:
acquiring a tracking target image and a tracking search area image, constructing a twin network-based feature extraction network, and extracting features;
in the tracking process, selecting a target image through PNR score, adding the target image into a template library, and sending the target image into a model learning network for model learning and updating to obtain a model target characteristic diagram and a search characteristic diagram;
convolving the model target characteristic diagram and the search characteristic diagram to obtain a response diagram, sending the response diagram into a scale regression network to obtain a scale position score diagram, and obtaining a target scale;
in the tracking process, the target state is judged, whether the target is shielded or lost is judged, and when the target is shielded or lost, the target searching range is expanded, and target recapture is carried out;
in the tracking process, the position, the scale, the confidence coefficient and the textural feature information of the target and the suspected target are recorded, when an interfering object occurs, the Hungarian matching algorithm is adopted to track and match the currently estimated tracking position and the historical tracking track, and the tracking position which is most matched with the historical track is selected, so that the influence of the interfering object is eliminated, and the long-term tracking of the target is guaranteed.
2. The deep neural network-based single-target long-time tracking method according to claim 1, wherein the twin network adopts ResNet50 as a backbone network.
3. The deep neural network-based single-target long-time tracking method according to claim 1, wherein the loss function of the model learning network is as follows:
Figure FDA0003745567650000011
Figure FDA0003745567650000012
Figure FDA0003745567650000013
wherein v isjIs a feature map of the training image, cjRepresenting the coordinates of the centre of the object, the purpose being to find the model f, the object image being represented by AcTo indicate that a is a background pixel when a pixel in the image is a background pixelcAbout 0, when A is the targetc1,s =vf, representing target confidence score, ycScore, y, representing each coordinate positioncInitialization with a Gaussian distribution, yc,Ac,ηcThe parameters are learned by on-line training.
4. The deep neural network-based single-target long-time tracking method according to claim 1, wherein the calculation formula of the PNR is as follows:
Figure FDA0003745567650000021
Figure FDA0003745567650000022
wherein S'maxRepresents the maximum value, S 'in the classification score response map S'minRepresents the minimum value, M, in the classification score response map SSDenotes the mean value of S' after removal of the maximum response value, TPNRIs the mean of all historical PNR values, TS′Is all historical S'maxIs measured.
5. The deep neural network-based single-target long-time tracking method according to claim 1, wherein the scale regression network is trained according to the following algorithm:
the loss function is:
Figure FDA0003745567650000023
LCIOU＝-ln(CIOU)
Figure FDA0003745567650000024
wherein L isCIOUIs the CIOU loss, expressed as (x)l,yl) The top left corner of the real target box is represented by (x)r,yr) The lower right corner representing the real target frame, (x)c,yc) Represents the coordinates of the center point, then
Figure FDA0003745567650000025
Figure FDA0003745567650000026
Figure FDA0003745567650000031
Figure FDA0003745567650000032
And obtaining the highest-score target position according to the classification score map, and taking the mean value of top-k target frames around the highest-score target position as the frame of the final target.
6. The deep neural network-based single-target long-time tracking method according to claim 1, wherein the target state judgment is performed according to the classification score of the target: when the score is less than 0.25, judging that the target is blocked or the target is not found currently, starting a counter, when the counter is more than 5, indicating that the target is lost, expanding the search area by 1 time every 2 frames according to the size of the target, searching the target by using a classification network, and when the score of the target is more than 0.8, considering that the target is found; when the search time exceeds a certain number of frames, the target is considered lost.
7. The single-target long-time tracking method based on the deep neural network as claimed in claim 1, wherein when a plurality of suspected targets appear in the tracking process, information of the suspected targets is saved, wherein the information includes a central point position, a scale, a classification confidence and an SILTP feature.
8. The deep neural network-based single-target long-time tracking method of claim 7, wherein for a given position (x)c,yc) The coding formula of the SILTP characteristic is as follows:
Figure FDA0003745567650000033
wherein g iscGray value of the central pixel, gkFor the other k pixels in a circle of radius R within the central pixel field, s is the piecewise function, e =0.3:
Figure FDA0003745567650000034
9. the single-target long-term tracking method based on the deep neural network as claimed in claim 8, wherein the cost in the Hungarian matching algorithm is calculated by cosine similarity as similarity between vectors,
Figure FDA0003745567650000041
wherein objiThe vector is composed of the position, the scale, the classification confidence coefficient and the SILTP characteristic value of the center point of the suspected target, trTThe vector is the center point position, the scale, the classification confidence coefficient and the SILTP characteristic value of the nearest point of the tracking track.
CN202210831023.1A 2022-07-14 2022-07-14 Single-target long-time tracking method based on deep neural network Pending CN115272409A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210831023.1A CN115272409A (en) 2022-07-14 2022-07-14 Single-target long-time tracking method based on deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210831023.1A CN115272409A (en) 2022-07-14 2022-07-14 Single-target long-time tracking method based on deep neural network

Publications (1)

Publication Number Publication Date
CN115272409A true CN115272409A (en) 2022-11-01

Family

ID=83764930

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210831023.1A Pending CN115272409A (en) 2022-07-14 2022-07-14 Single-target long-time tracking method based on deep neural network

Country Status (1)

Country Link
CN (1) CN115272409A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117830883A (en) * 2024-03-04 2024-04-05 成都庆龙航空科技有限公司 Aiming method of unmanned aerial vehicle
CN117830883B (en) * 2024-03-04 2024-05-31 成都庆龙航空科技有限公司 Aiming method of unmanned aerial vehicle

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117830883A (en) * 2024-03-04 2024-04-05 成都庆龙航空科技有限公司 Aiming method of unmanned aerial vehicle
CN117830883B (en) * 2024-03-04 2024-05-31 成都庆龙航空科技有限公司 Aiming method of unmanned aerial vehicle

Similar Documents

Publication Publication Date Title
CN108665481B (en) Self-adaptive anti-blocking infrared target tracking method based on multi-layer depth feature fusion
CN111354017A (en) Target tracking method based on twin neural network and parallel attention module
CN109785366B (en) Related filtering target tracking method for shielding
CN109214403B (en) Image recognition method, device and equipment and readable medium
CN108446634B (en) Aircraft continuous tracking method based on combination of video analysis and positioning information
CN111476817A (en) Multi-target pedestrian detection tracking method based on yolov3
CN111311647B (en) Global-local and Kalman filtering-based target tracking method and device
CN110766723B (en) Unmanned aerial vehicle target tracking method and system based on color histogram similarity
CN112036455B (en) Image identification method, intelligent terminal and storage medium
CN111340842B (en) Correlation filtering target tracking method based on joint model
CN114119659A (en) Multi-sensor fusion target tracking method
CN114897932A (en) Infrared target tracking implementation method based on feature and gray level fusion
CN110516528A (en) A kind of moving-target detection and tracking method based under movement background
CN109448024B (en) Visual tracking method and system for constructing constraint correlation filter by using depth data
CN115311449A (en) Weak supervision image target positioning analysis system based on class reactivation mapping chart
CN114627156A (en) Consumption-level unmanned aerial vehicle video moving target accurate tracking method
CN111091583B (en) Long-term target tracking method
CN110147768B (en) Target tracking method and device
CN112580438A (en) Point cloud identification method in three-dimensional scene
CN116777956A (en) Moving target screening method based on multi-scale track management
CN116665097A (en) Self-adaptive target tracking method combining context awareness
Zhang et al. Target tracking for mobile robot platforms via object matching and background anti-matching
Chun-Zhao et al. Drivable road boundary detection for intelligent vehicles based on stereovision with plane-induced homography
CN115272409A (en) Single-target long-time tracking method based on deep neural network
CN113837243B (en) RGB-D camera dynamic visual odometer method based on edge 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