CN109978910A - Target tracking system and method based on DSP - Google Patents
Target tracking system and method based on DSP Download PDFInfo
- Publication number
- CN109978910A CN109978910A CN201910130679.9A CN201910130679A CN109978910A CN 109978910 A CN109978910 A CN 109978910A CN 201910130679 A CN201910130679 A CN 201910130679A CN 109978910 A CN109978910 A CN 109978910A
- Authority
- CN
- China
- Prior art keywords
- target
- dsp
- video
- algorithm
- circuit
- 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
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000004891 communication Methods 0.000 claims abstract description 7
- 238000006243 chemical reaction Methods 0.000 claims abstract description 5
- 230000005540 biological transmission Effects 0.000 claims abstract description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 82
- 230000006870 function Effects 0.000 claims description 41
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 8
- 230000002093 peripheral effect Effects 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 230000003068 static effect Effects 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 108091001599 RGB-10 Proteins 0.000 claims description 3
- 230000006978 adaptation Effects 0.000 claims description 3
- 230000008054 signal transmission Effects 0.000 claims description 2
- 125000004122 cyclic group Chemical group 0.000 claims 1
- 230000003139 buffering effect Effects 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 7
- 238000010801 machine learning Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000006641 stabilisation Effects 0.000 description 2
- 238000011105 stabilization Methods 0.000 description 2
- 230000000087 stabilizing effect Effects 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5011—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals
- G06F9/5016—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals the resource being the memory
-
- 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
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a target tracking system based on DSP in the technical field of machine vision and a method thereof, comprising a video acquisition circuit for acquiring a target video; a video decoding circuit for analog/digital conversion of a video signal; the video display circuit is used for transmitting the video signal to an upper computer for video display; the data buffer circuit is used for buffering the video signal to be processed; the video processing circuit consists of an FPGA and a DSP and is used for reading and calculating the video signals in the cache in real time, outputting the real-time position of a target and realizing the function of tracking the target; the communication circuit is used for data communication and transmission among the modules; and the power supply circuit is used for supplying power to each module and the auxiliary circuit in the system. The invention improves the precision and the real-time performance of target tracking, and ensures that the system runs stably and is convenient to maintain.
Description
Technical field
The present invention relates to technical field of machine vision, and in particular, to a kind of Target Tracking System and its side based on DSP
Method.
Background technique
Target following is a very important research direction in machine vision, and target tracking algorism is in embedded system
On realization be always the field in need to put forth effort the difficult point captured.In recent years, with the fast development of machine vision algorithm, mesh
The relevant technologies such as mark tracking have been widely used in the fields such as defense military, public safety, industrial production, have with target following
The algorithm of pass also emerges one after another.However, the target tracking algorism proposed at present is although many kinds of, significant effect, but these
Algorithm is to carry out algorithm realization based on general purpose computer, server or work station mostly, and the exploitation of algorithm generally needs to borrow
Help this kind of third party software library such as OpenCV, MATLAB, tensorflow or software, and cannot directly embedded end into
Row deployment, this defect allow for this kind of target tracking algorism and are difficult to apply in actual engineering project, greatly limit calculation
Method uses field.Though secondly, at present have related technical personnel propose the Target Tracking System based on embedded platform, this
Algorithm used in class system is the image processing algorithm of this quasi-tradition such as target difference, template matching, and Scheme uses
Be the tupe artificially limited, either for complexity or adaptability, actual engineering can hardly be adapted to
Environment.
DSP has become as a kind of embedded system with powerful calculation processing power based on machine learning field
The Key Platform of interior target tracking algorism technology landing.Target tracking algorism in machine learning field is compared to traditional mesh
Track algorithm is marked with obviously advantage, the computation model of such algorithm are by re-starting training to signal, data
And what autonomous learning obtained, this is capable of the value of depth excavation training data, so that eventually for the algorithm mould of signal processing
Type is optimal, thus the accuracy and adaptability of boosting algorithm.But since machine learning algorithm usually requires the big of progress
The mathematical operation of amount, it is more demanding for the calculated performance of platform, so applying this kind of target tracking algorism embedded flat
It is always to put forth effort the difficult point captured in the field on platform, therefore, the research and realization of the method for tracking target based on DSP are very
It is necessary and with high realistic meaning.
Through the retrieval to the prior art, Chinese invention patent CN20160596748.1, entitled one kind is based on feature
Matched Ground Target Tracking device, including programmable gate array FPGA and digital signal processor DSP;FPGA is used for outside
Camera feeds image sequence and extracts characteristics of image, and then completes adjacent interframe characteristic matching, by success interframe characteristic matching result
Send DSP to;Cross-correlation is carried out according to the interframe geometric transform relation of DSP feedback accurately to match;Digital signal processor DSP is used
Adjacent image Inter-frame Transformation relationship is calculated in the characteristic matching result exported according to the FPGA.
The core processing algorithm that the invention uses is still the target tracking algorism in traditional field, i.e., extracts target first
Key point, then carry out the key point comparison calculation between frame and frame, to obtain matching target, realize target following.This
Although kind of an algorithm reduces operand to a certain extent, the operational efficiency of algorithm is improved, this extraction target critical
The mode of point is easily lost target signature, is easy to obscure tracking target in background information situation similar with foreground information,
Reduce the accuracy of target following.In addition to this, this algorithm is when deformation and change of scale occurs in target, tracking effect
It will big heavy discount.In contrast, the KCF algorithm used in the present invention then has preferable effect in above-mentioned performance test
Fruit, and the bottom hardware of present invention combination DSP, improve algorithm, so that there is short time disappearance, deformation in target
And change of scale Shi Junneng shows excellent performance, algorithm has certain robustness.
Summary of the invention
In view of the drawbacks of the prior art, the object of the present invention is to provide a kind of Target Tracking System based on DSP and its sides
Method.The present invention then proposes the side that a kind of target tracking algorism (KCF algorithm) of classics is realized and optimized on embedded platform
Method, and a kind of hardware platform that this method is realized is provided simultaneously, while system platform cooperation software algorithm operational effect stabilization,
It is easy to maintain, the possibility of realization is provided for all kinds of algorithms landing in field of machine vision.
The present invention relates to a kind of Target Tracking Systems based on DSP, comprising:
Video capture circuit, the acquisition for target video;
Video decoding circuit, the analog/digital conversion for vision signal;
Video display circuit is shown for video signal transmission to host computer to be carried out video;
Data caching circuit, the caching for vision signal to be processed;
Video processing circuits is made of FPGA and dsp processor, for being read in real time to the vision signal in caching
With calculating, target real time position is exported, realizes target following function;
Telecommunication circuit, for the data communication and transmission between each module;
Power circuit, for powering to module each in system and auxiliary circuit.
Preferably, the data caching circuit is made of FPGA and two panels SDRAM chip, and the SDRAM chip is for scheming
As the caching of data, the FPGA is used for the control of data access.
Preferably, the video display circuit is made of ADV7123 chip and its peripheral circuit, is used for RGB10 digit
Word component signal is converted to the output of RGB analog component signal, wherein the mixed synchronization of line synchronising signal and field sync signal composition
Signal does signal condition and load driving by being connected internally to the outlet of IOG vision signal, and using AD8041.
Preferably, the dsp processor is for processing target tracking core algorithm cartridge tracking result output.
Preferably, specific step is as follows for the target following core algorithm for realizing in the dsp processor:
Step 1: calling tracker_init function first initializes tracker, inputs in the picture of first frame
Hold and frame selects tracking target;
Step 2: it calls get_features function to carry out feature extraction to target area and obtains template tmpl;
Step 3: create_gaussian_peak function is called to generate Gaussian matrix according to obtained template characteristic figure
prob;
Step 4: call train function that training pattern is mapped to Fourier, according to template tmpl and Gaussian matrix
Prob training obtains object detector parameter alpha;
Step 5: the detector model obtained according to former frame picture training calls detect function check target current
Position in frame, and frame selects tracking target;
Step 6: re-calling get_features function and carry out feature extraction to the target detected in present frame, from
And obtain new target template tmpl;Train function is called to calculate new object detector parameter, Jin Ergeng in Fourier
New α;
Step 7: step 2 is repeated to step 6 to subsequent each frame picture.
Preferably, shifting of the target following core algorithm in the dsp processor with a kind of KCF algorithm on a hardware platform
It plants and realizes and improve.
Preferably, the KCF algorithm acquires positive negative sample using the circular matrix of target peripheral region, with spread training
Collection, and use ridge regression training objective detector.
Preferably, the KCF algorithm utilizes circular matrix in the property of Fourier diagonalizable, by entire calculating process
It is mapped to Fourier to be calculated, converts vector in Fourier for matrix inversion operation complicated in script time domain
Hadamad accumulates operation.
Preferably, in the KCF algorithm, original OpenCV function is reconstructed, to image interpolation, matrixing, HOG feature
Extraction, matrix operation, the appropriate cutting in basis of two-dimensional Fourier transform algoritic module reservation function and optimization, can be adapted to
The hardware bottom of DSP;It is hidden during algorithm constantly apply with releasing memory to eliminate DSP for the strategy distributed using static memory
Suffer from;High operation flow function in algorithm is realized by C64X DSP library function, so that calculating process is tight with bottom hardware
Lattice adaptation;The strategy for turning fixed-point calculation using floating-point operation improves the efficiency of algorithm.
The invention further relates to a kind of method for tracking target based on DSP, comprising:
Step 1: enter self-test after system electrification, working condition is entered after the completion of self-test;
Step 2: video capture circuit starting shooting, and picture frame is spread and carries out analog/digital turn into video decoding circuit
It changes, then exports transformation result to FPGA;
Step 3: digital video signal is sent to PC control platform and shown by video display circuit, and waits behaviour
Make personnel's frame and selects target area;
Step 4: after the completion of the choosing of target area frame, system enters target following state, and by the figure containing the target area
First frame picture of the piece as tracking video, is sent into cache chip;
Step 5: DSP reads in frame data from buffer zone, and invocation target track algorithm updates detector parameters, output
The target area being calculated, while exporting control amount makes the camera in video capture circuit that target area be followed to be moved
Tracking;
Step 6: arriving step 5 to subsequent frame Data duplication above-mentioned steps four, tracks target if replacing, repeatedly step 1
To step 5.
Compared with prior art, the present invention have it is following the utility model has the advantages that
1, the target tracking algorism in machine learning field is reconstructed, is allowed to be detached from the thirds such as OpenCV, MATLAB
Square software or software library, and enable independently to be deployed in DSP platform and run.
2, the Memory Allocation mechanism of optimization algorithm, the strategy distributed using static memory are eliminated DSP and constantly applied in algorithm
With the hidden danger during releasing memory.
3, for floating-point operation, matrixing, matrix operation a large amount of in algorithm etc., floating-point operation is taken to turn fixed point fortune
The efficiency for calculating, opening software flow, improving algorithm using the methods of library function optimization, the speed of service of boosting algorithm.
4, for function calculatings modules such as image interpolation, HOG feature extractions in algorithm, on the basis of reservation function into
Row is suitably cut out and is optimized, and the hardware bottom layer of DSP can be adapted to.
5, the classical target tracking algorism (KCF algorithm) in machine learning field has been deployed in DSP platform, greatly
The application field of the algorithm is expanded.
6, compared to other embedded target tracking systems, present invention uses the target following calculations in machine learning field
Method greatly improves the accuracy and tracking accuracy of target following while guaranteeing real-time performance of tracking.
7, the present invention combines DSP hardware bottom to propose series of optimum of KCF during realizing on embedded platform
Method, and a kind of hardware platform that this method is realized is provided simultaneously, hardware platform cooperation software algorithm operational effect stabilization,
It is easy to maintain, the possibility of realization is provided for all kinds of algorithms landing in field of machine vision.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon:
Fig. 1 is the system block diagram of hardware platform in the present invention.
Fig. 2 is the algorithm block diagram of target tracking algorism in the present invention.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, various modifications and improvements can be made.These belong to the present invention
Protection scope.
Embodiment
In view of the above demand, the present invention relates to a kind of Target Tracking System and its method based on DSP realizes a kind of base
In DSP and it is aided with FPGA and the Target Tracking System of other peripheral circuits, the system is in coring correlation filtering (KCF)
On the basis of, defect in the prior art is optimized and is improved, so that system improves core while guaranteeing tracking accuracy
The real-time of center algorithm.
The hardware platform of this system has video capture circuit, video decoding circuit, video display circuit, data buffer storage electricity
Road, video processing circuits and telecommunication circuit etc..Wherein video acquisition module includes simulation camera and optical lens, is mainly responsible for
The acquisition of target video;Video decoding circuit is made of TVP5150 chip and its peripheral circuit, is mainly responsible for and is acquired camera
PAL analog signal be converted into meeting the digital signal of 8 4:2:2YCBCR formats of BT.656 standard;Video display circuit by
ADV7123 chip and its peripheral circuit composition, are mainly responsible for RGB10 bit digital component signal being converted to RGB analogue component letter
Number output, wherein line synchronising signal and field sync signal composition mixed synchronization signal gone out by being connected internally to IOG vision signal
Mouthful, and signal condition and load driving are done using AD8041;Data caching circuit is made of FPGA and two panels SDRAM chip, is examined
Consider DSP and read and handle the faster speed of data needs, therefore uses SDRAM chip as the caching of image data, FPGA
It is the control chip for data access;Video processing circuits is made of FPGA and dsp chip, and wherein FPGA is mainly used for video
The pretreatment of signal, DSP are responsible for core algorithm and tracking result output of processing target tracking etc.;Telecommunication circuit is mainly responsible for mould
Intend the communication protocol between each module and between host computer, including CAN bus agreement, USB communication protocol, serial communication association
View etc..It is also equipped with power circuit in addition to foregoing circuit module, in system to be powered for functions circuit and chip, and right
Power supply is filtered.In the partial circuit using TPS54320 chip by 12V DC power input pressure stabilizing be 3.3V direct current
Power supply output and the output of 1.2V DC power supply;Using TPS7A8101 chip by 3.3V DC supply input pressure stabilizing be 1.8V direct current
Power supply output and the output of 2.5V DC power supply.
In this embodiment, the carrying out practically process of system is as follows.
Step 1: enter self-test after system electrification, working condition is entered after the completion of self-test;
Step 2: as shown in Figure 1, video capture circuit starting shooting, and by picture frame spread into video decoding circuit into
Row analog/digital conversion then exports transformation result to FPGA;
Step 3: digital video signal is sent to PC control platform and shown by video display circuit, and waits behaviour
Make personnel's frame and selects target area;
Step 4: after the completion of the choosing of target area frame, system enters target following state, and by the figure containing the target area
First frame picture of the piece as tracking video, is sent into cache chip;
Step 5: DSP reads in frame data from buffer zone, and invocation target track algorithm updates detector parameters, output
The target area being calculated, while exporting control amount makes the camera in video capture circuit that target area be followed to be moved
Tracking;
Step 6: arriving step 5 to subsequent frame Data duplication above-mentioned steps four, tracks target if replacing, repeatedly step 1
To step 5.
In above-mentioned implementation steps, the target tracking algorism in DSP is the core procedure that realization system carries out target following,
The specific implementation steps are as follows for the algorithm:
Step 1: as shown in Fig. 2, call tracker_init function to initialize tracker first, input first
The image content and frame of frame select tracking target;
Step 2: it calls get_features function to carry out feature extraction to target area and obtains template tmpl;
Step 3: create_gaussian_peak function is called to generate Gaussian matrix according to obtained template characteristic figure
prob;
Step 4: call train function that training pattern is mapped to Fourier, according to template tmpl and Gaussian matrix
Prob training obtains object detector parameter alpha.
Step 5: the detector model obtained according to former frame picture training calls detect function check target current
Position in frame, and frame selects tracking target;
Step 6: re-calling get_features function and carry out feature extraction to the target detected in present frame, from
And obtain new target template tmpl;Train function is called to calculate new object detector parameter, Jin Ergeng in Fourier
New α;
Step 7: step 2 is repeated to step 6 to subsequent each frame picture.
On the core algorithm of target following, the present invention provides a kind of shifting of KCF algorithm on this system hardware platform
Plant realization and improved method.KCF algorithm is a kind of typical discriminate target tracking algorism, and the core concept of the algorithm is root
According to the tracking target in current frame image, training obtains an optimal objective on the basis of calculation template feature and Gaussian matrix
Then detector model calculates the response of picture in next frame picture using the detector, then value frame choosing according to response
Target area out, and template characteristic and Gaussian matrix are updated as new template, to update optimal objective detector model
Subsequent frame iteration is carried out, which achievees the purpose that target following in a manner of this target area of differentiation frame by frame.Wherein in mesh
It marks in the training process of detector model and introduces kernel function, to enhance the generalization ability of model, improve the accuracy rate of target identification.
Circular matrix of algorithm during realization dexterously using target peripheral region acquires positive negative sample, with extension
Training set, and use ridge regression training objective detector;In addition to this, this processing method can use circular matrix in Fu
The property of leaf domain diagonalizable, is mapped to Fourier for entire calculating process and calculates, by square complicated in script time domain
Battle array inversion operation is converted into the Hadamad product operation of vector in Fourier, greatly reduces the time complexity of algorithm operation,
Arithmetic speed is improved, this is but also transplanting of the algorithm on the embedded platforms such as DSP is possibly realized.
However, traditional KCF algorithm is generally basede on OpenCV or MATLAB exploitation, can not be transported on embedded platform
Row;Meanwhile KCF is related to a large amount of floating-point operation, this meeting is so that the processing speed of embedded platform greatly reduces;In addition, algorithm
In the operations amount such as two-dimensional Fourier transform, image interpolation, matrixing it is big, the treatment effeciency of embedded platform can be made
Further decline.
In view of problem above and case study, this TMS320C6414 series DSP based on TI company provided by the invention
KCF realize and optimization method, key technology include the following:
Original OpenCV function is reconstructed, including image interpolation, matrixing, HOG feature extraction, matrix fortune
The classic algorithms module such as calculation, two-dimensional Fourier transform carries out appropriate cutting and optimization on the basis of reservation function, can
It is adapted to the hardware bottom layer of DSP.
The Memory Allocation mechanism of optimization algorithm, using static memory distribute strategy, eliminate DSP algorithm constantly apply with
Hidden danger during releasing memory.
For the operational efficiency for improving algorithm, the high operation flow function in algorithm is carried out in fact by C64X DSP library function
It is existing, so that calculating process and bottom hardware rigid adaptation.
For floating-point operation a large amount of in algorithm, turn the strategy of fixed-point calculation using floating-point operation, improves the efficiency of algorithm.
Software pipeline operation is opened, successive ignition is converted by multilayer circulation and executes parallel.
A large amount of data access in algorithm and common mathematical operation are optimized using instrinsic function, promoted
The speed of service of algorithm.
In conclusion the positive effect of the present invention is that: traditional target tracking algorism is although many kinds of, realizes
Significant effect, but these algorithms are to carry out algorithm realization, algorithm based on general purpose computer, server or work station mostly
Exploitation generally need by this kind of third party software library such as OpenCV, MATLAB, tensorflow or software, and cannot
It is directly disposed in embedded end, therefore, algorithm is limited significantly using field.The present invention then proposes a kind of target of classics
The method that track algorithm (KCF algorithm) is realized and optimized on embedded platform, and provide what a kind of this method was realized simultaneously
Hardware platform.The hardware platform cooperates software algorithm operational effect stable, easy to maintain, is all kinds of calculations in field of machine vision
Method landing provides the possibility realized.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (10)
1. a kind of Target Tracking System based on DSP characterized by comprising
Video capture circuit, the acquisition for target video;
Video decoding circuit, the analog/digital conversion for vision signal;
Video display circuit is shown for video signal transmission to host computer to be carried out video;
Data caching circuit, the caching for vision signal to be processed;
Video processing circuits is made of FPGA and dsp processor, by read with based in real time to the vision signal in caching
It calculates, exports target real time position, realize target following function;
Telecommunication circuit, for the data communication and transmission between each module;
Power circuit, for powering to module each in system and auxiliary circuit.
2. the Target Tracking System according to claim 1 based on DSP, characterized in that the data caching circuit by
FPGA and SDRAM chip composition, the SDRAM chip are used for the caching of image data, and the FPGA is used for the control of data access
System.
3. the Target Tracking System according to claim 1 based on DSP, characterized in that the video display circuit by
ADV7123 chip and its peripheral circuit composition, it is defeated for RGB10 bit digital component signal to be converted to RGB analog component signal
Out, wherein the mixed synchronization signal of line synchronising signal and field sync signal composition is exported by being connected internally to IOG vision signal, and
Signal condition and load driving are done using AD8041.
4. the Target Tracking System according to claim 1 based on DSP, characterized in that the dsp processor is for handling
Target following core algorithm and tracking result output.
5. the Target Tracking System according to claim 4 based on DSP, characterized in that realize in the dsp processor
Specific step is as follows for target following core algorithm:
Step 1: calling tracker_init function first initializes tracker, inputs the image content of first frame simultaneously
Frame selects tracking target;
Step 2: it calls get_features function to carry out feature extraction to target area and obtains template tmpl;
Step 3: create_gaussian_peak function is called to generate Gaussian matrix prob according to obtained template characteristic figure;
Step 4: calling train function that training pattern is mapped to Fourier, is instructed according to template tmpl and Gaussian matrix prob
Get object detector parameter alpha;
Step 5: detect function check target is called in the current frame according to the detector model that former frame picture training obtains
Position, and frame selects tracking target;
Step 6: re-calling get_features function and carry out feature extraction to the target detected in present frame, thus
To new target template tmpl;It calls train function to calculate new object detector parameter in Fourier, and then updates α;
Step 7: step 2 is repeated to step 6 to subsequent each frame picture.
6. the Target Tracking System according to claim 5 based on DSP, characterized in that the target in the dsp processor
Core algorithm is tracked to be realized and improved with a kind of transplanting of KCF algorithm on a hardware platform.
7. the Target Tracking System according to claim 6 based on DSP, characterized in that the KCF algorithm uses target week
The circular matrix for enclosing region acquires positive negative sample, with spread training collection, and uses ridge regression training objective detector.
8. the Target Tracking System according to claim 6 based on DSP, characterized in that the KCF algorithm utilizes Cyclic Moment
Entire calculating process is mapped to Fourier and calculated by battle array in the property of Fourier diagonalizable, will be in script time domain
Complicated matrix inversion operation is converted into the Hadamad product operation of vector in Fourier.
9. the Target Tracking System according to claim 6 based on DSP, characterized in that in the KCF algorithm, reconstruct is former
Some OpenCV functions, to image interpolation, matrixing, HOG feature extraction, matrix operation, two-dimensional Fourier transform algorithm mould
The basis of block reservation function is appropriate to be cut and optimizes, and the hardware bottom layer of DSP can be adapted to;The plan distributed using static memory
Slightly, DSP is eliminated constantly to apply and hidden danger during releasing memory in algorithm;High operation flow function in algorithm is passed through into C64X
DSP library function is realized, so that calculating process and bottom hardware rigid adaptation;Turn the plan of fixed-point calculation using floating-point operation
Slightly, the efficiency of algorithm is improved.
10. a kind of method for tracking target based on DSP characterized by comprising
Step 1: enter self-test after system electrification, working condition is entered after the completion of self-test;
Step 2: video capture circuit starting shooting, and picture frame is spread and carries out analog/digital conversion into video decoding circuit, with
Transformation result is exported to FPGA afterwards;
Step 3: digital video signal is sent to PC control platform and shown by video display circuit, and waits operator
Member's frame selects target area;
Step 4: after the completion of the choosing of target area frame, system enters target following state, and the picture containing the target area is made
For the first frame picture for tracking video, it is sent into cache chip;
Step 5: DSP reads in frame data from buffer zone, and invocation target track algorithm updates detector parameters, and output calculates
Obtained target area, at the same export control amount make the camera in video capture circuit follow target area moved with
Track;
Step 6: arriving step 5 to subsequent frame Data duplication above-mentioned steps four, tracks target if replacing, repeatedly step 1 to step
Rapid five.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910130679.9A CN109978910A (en) | 2019-02-21 | 2019-02-21 | Target tracking system and method based on DSP |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910130679.9A CN109978910A (en) | 2019-02-21 | 2019-02-21 | Target tracking system and method based on DSP |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109978910A true CN109978910A (en) | 2019-07-05 |
Family
ID=67077169
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910130679.9A Pending CN109978910A (en) | 2019-02-21 | 2019-02-21 | Target tracking system and method based on DSP |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109978910A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111221770A (en) * | 2019-12-31 | 2020-06-02 | 中国船舶重工集团公司第七一七研究所 | Kernel correlation filtering target tracking method and system |
CN112465781A (en) * | 2020-11-26 | 2021-03-09 | 华能通辽风力发电有限公司 | Method for identifying defects of main parts of wind turbine generator based on video |
CN113572941A (en) * | 2021-08-16 | 2021-10-29 | 中国科学院长春光学精密机械与物理研究所 | Multifunctional image acquisition device applied to CPCI computer |
CN113709399A (en) * | 2021-08-31 | 2021-11-26 | 中国电子科技集团公司第五十八研究所 | Visual target tracking system based on DSP + FPGA |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106599836A (en) * | 2016-12-13 | 2017-04-26 | 北京智慧眼科技股份有限公司 | Multi-face tracking method and tracking system |
CN108154523A (en) * | 2017-12-25 | 2018-06-12 | 北京航空航天大学 | A kind of real-time modeling method system and method in airborne photoelectric platform |
-
2019
- 2019-02-21 CN CN201910130679.9A patent/CN109978910A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106599836A (en) * | 2016-12-13 | 2017-04-26 | 北京智慧眼科技股份有限公司 | Multi-face tracking method and tracking system |
CN108154523A (en) * | 2017-12-25 | 2018-06-12 | 北京航空航天大学 | A kind of real-time modeling method system and method in airborne photoelectric platform |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111221770A (en) * | 2019-12-31 | 2020-06-02 | 中国船舶重工集团公司第七一七研究所 | Kernel correlation filtering target tracking method and system |
CN112465781A (en) * | 2020-11-26 | 2021-03-09 | 华能通辽风力发电有限公司 | Method for identifying defects of main parts of wind turbine generator based on video |
CN113572941A (en) * | 2021-08-16 | 2021-10-29 | 中国科学院长春光学精密机械与物理研究所 | Multifunctional image acquisition device applied to CPCI computer |
CN113709399A (en) * | 2021-08-31 | 2021-11-26 | 中国电子科技集团公司第五十八研究所 | Visual target tracking system based on DSP + FPGA |
CN113709399B (en) * | 2021-08-31 | 2024-03-08 | 中国电子科技集团公司第五十八研究所 | Visual target tracking system based on DSP+FPGA |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109978910A (en) | Target tracking system and method based on DSP | |
Yeh et al. | Lightweight deep neural network for joint learning of underwater object detection and color conversion | |
Li et al. | Lightweight detection networks for tea bud on complex agricultural environment via improved YOLO v4 | |
Xu et al. | Recurrent convolutional neural network for video classification | |
CN113011403B (en) | Gesture recognition method, system, medium and device | |
WO2017209660A1 (en) | Learnable visual markers and method of their production | |
CN111833360B (en) | Image processing method, device, equipment and computer readable storage medium | |
CN115619685A (en) | Transformer method for tracking structure for image restoration | |
Barbole et al. | A review on fruit detection and segmentation techniques in agricultural field | |
CN108648213A (en) | A kind of implementation method of KCF track algorithms on TMS320C6657 | |
CN105954292A (en) | Underwater structure surface crack detection device and method based on compound-eye bionic vision | |
Guo et al. | Efficient shadow detection and removal using synthetic data with domain adaptation | |
Jin et al. | Auxiliary detection head for one-stage object detection | |
CN104915641A (en) | Method for obtaining face image light source orientation based on android platform | |
US11361189B2 (en) | Image generation method and computing device | |
CN112070227A (en) | Neural network training method, device and equipment | |
CN116740643A (en) | Bird recognition system and method based on visual image | |
Wang et al. | Identification of growing points of cotton main stem based on convolutional neural network | |
CN109271974A (en) | A kind of lightweight face joint-detection and recognition methods and its system | |
CN109377501A (en) | Remote sensing images naval vessel dividing method and system based on transfer learning | |
Wang et al. | Multi-patch and feature fusion network for single image Dehazing | |
Psaltis et al. | Deep 3d flow features for human action recognition | |
Stommel et al. | A fast, robust and low bit-rate representation for SIFT and SURF features | |
Wang et al. | Efficient and robust homography estimation using compressed convolutional neural network | |
CN114640785A (en) | Site model updating method and system |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190705 |
|
RJ01 | Rejection of invention patent application after publication |