CN104751157B - Detecting and tracking method based on FPGA - Google Patents
Detecting and tracking method based on FPGA Download PDFInfo
- Publication number
- CN104751157B CN104751157B CN201310753663.6A CN201310753663A CN104751157B CN 104751157 B CN104751157 B CN 104751157B CN 201310753663 A CN201310753663 A CN 201310753663A CN 104751157 B CN104751157 B CN 104751157B
- Authority
- CN
- China
- Prior art keywords
- target
- point
- fifo
- fpga
- information
- 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.)
- Active
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the target following technical fields under the complex background of ground, and in particular to a kind of detecting and tracking method based on FPGA.This method comprises the following steps:Step 1: being filtered denoising to image using medium filtering;Step 2: carrying out region growing to target point obtains target area;Step 3: carrying out feature extraction to target;Step 4: detection window adjusts;Step 5: carrying out object matching according to characteristic value;Step 6: target genuine-fake differentiates;Step 7: carrying out target information prediction according to movement locus;Step 8: target's center's position output.The present invention makes full use of target property, accurate positioning using a kind of track algorithm based on region growing;Using the design scheme of SOPC, algorithm realization method of making rational planning for gives full play to the convenience of the real-time and software processing of FPGA stream treatments, and execution speed is fast, realizes real-time tracking;It is tracked using adaptive threshold fuzziness and windowing, improves the accuracy of tracking and greatly reduce calculation amount.
Description
Technical field
The invention belongs to the target following technical fields under the complex background of ground, and in particular to a kind of detection based on FPGA
Tracking.
Background technology
Two important indicators of infrared object tracking system are stability and real-time respectively.Presently, there are many tracking
There is being set to inaccurate, particle filter algorithm calculation amount in algorithm, mean shift algorithms, the algorithm etc. based on projection
It is not easy to real-time implementation greatly very much.The processing platform of the tracking of present target is usually DSP+FPGA, the volume and power consumption of system
Larger, with the development of SOPC technologies, the SOPC design platforms based on monolithic FPGA become a present development trend.
Invention content
The detecting and tracking method based on FPGA that the purpose of the present invention is to provide a kind of, is easy out with solving monotrack
The problem of active, computationally intensive, real-time are difficult to meet the requirements, overcome the deficiencies in the prior art.
In order to solve the above-mentioned technical problem, the technical scheme is that:
A kind of detecting and tracking method based on FPGA, includes the following steps:
Step 1: being filtered denoising to image using medium filtering;
Step 2: carrying out region growing to target point obtains target area;
Step 3: carrying out feature extraction to target;
Step 4: detection window adjusts;
Step 5: carrying out object matching according to characteristic value;
Step 6: target genuine-fake differentiates;
Step 7: carrying out target information prediction according to movement locus;
Step 8: target's center's position output.
The step 1 first passes around medium filtering and does denoising to it, to current when carrying out target following
All data of pixel 3*3 neighborhoods are ranked up, and in-between value is taken to replace original pixel.
The step 2 implementation method is as follows:
This step is divided into two parts:Copy detection area image finds seed point;Region life is carried out in detection zone
It is long;
Step 2.1 copy detection area image finds seed point
Target following is realized in window, using the track algorithm based on region growing, processing is marked to image, is delayed
The image of a frame detection zone is deposited, the infrared image of detection zone is stored in RAM in real time;It is maximum using mean value in eight neighborhood
Point is used as seed point, the realization of the part that can be realized with flowing water inside FPGA;Being carried out according to image for growing threshold is adaptive
Adjustment, adjustment formula are:
Step 2.2 region growing
Region growing is realized with two FIFO, first FIFO stores the information of seed point, therefore when FIFO non-emptys,
Show still there is seed point not grow, then needs that it is continued to grow;The coordinate of second FIFO storage target point
Information, i.e., relative to the abscissa value of detection zone and ordinate value;The coordinate value calculates the target point by multiply-add operation
Address, then obtain the address of its four neighborhood, then read the gray value of the pixel from RAM successively;First determine whether the point
Whether grew, if having grown, other processing need not be done;If do not grown, need to sentence
It break whether within region growing range, if within growth scope, by the point coordinate information is stored to second
In a FIFO;Until all growth completion indicates all targets to the seed point in first FIFO when i.e. FIFO is empty
It has grown and has completed;
Region growing is completed using asynchronous FIFO, wherein it is the fifth harmonic for reading clock to write clock, for region growing
FPGA optimization processings are as follows:
First clock provides the address of the upper neighborhood of target point, judges whether the right neighborhood of a upper seed point is raw
It grew, and continued to determine whether to be target point if not growing, and needed to set FIFO with effect if it is target point, and will be upper
The right neighborhood of one seed point is stored in FIFO as seed point, is otherwise not processed;
Second clock provides the address of the left neighborhood of target point, judges whether the lower neighborhood of a upper seed point is raw
It grew, and continued to determine whether to be target point if not growing, and then needed to set FIFO with effect if it is target point, and will
The lower neighborhood of a upper seed point is stored in FIFO as seed point, is otherwise not processed;
Third clock provides the address of the right neighborhood of target point;
4th clock, provides the address of the lower neighborhood of target point, judges whether the upper neighborhood of current target point is raw
It grew, and continued to determine whether to be target point if not growing, and needed to set FIFO with effect if if it is target point, and
It is stored in FIFO using the upper neighborhood of current seed point as seed point, is otherwise not processed;
5th clock, judges whether the left neighborhood of current target point had grown, and continues if not growing
Judge whether to be target point, if it is target point, then need to set FIFO with effect, and using the left neighborhood of current seed point as
Seed point is stored in FIFO, is otherwise not processed.
The step 3:Clarification of objective is extracted, the gray scale of target, area, boundary, type heart information;The gray scale of target is believed
Breath needs the average value of entire target gray value;The area of target refers to the number of pixels that target is included;Boundary refers to mesh
Target up and down boundary value, asked by bubbling method;The location information of target is the position where the object type heart;Due to target
Movement has relevance, therefore the information of the target of former frames should should have continuity with the information of current goal;
It is as follows to seek the area of target, gray scale, type heart calculation formula:
A indicates that the area of target, R indicate target area in formula,The abscissa of the object type heart and vertical seat are indicated respectively
Mark;Indicate average gray value.
The step 4:Tracking in the window of use eliminates interference by opening a window, and the size of window is needed according to specific mesh
Target size is adjusted;Using the ratio and object boundary and window edge distance of window area and target area as judgement
According to adjustment window size position in real time again so that window, which can include target, as more as possible can must exclude the interference signal.
The step 5:
Step 5.1 Gray-scale Matching
Whether the average gray difference value for the target that the average gray of the target of this frame image is predicted with upper frame meets one
Confidence level is added one by fixed condition if meeting, otherwise constant;
It is step 5.2, area matched
Whether the area for the target that the average gray of the target of this frame image is predicted with upper frame meets in certain condition,
Confidence level is added one if meeting, it is otherwise constant;
Step 5.3, location matches
The target location of this frame image is compared with the picture position after prediction, judges whether its difference meets
Confidence level is added one by certain condition if meeting, otherwise constant;
Step 5.4, Boundary Match
The object boundary of this frame image is compared with the image boundary after prediction, judges whether its difference meets
Confidence level is added one by certain condition if meeting, otherwise constant.
The step 6:By above judgement, confidence level changes, when confidence level is not less than a certain threshold value, then
Think that the target is true target, using the information of target write-in object chain as the foundation of prediction next frame target information, if
To be less than the threshold value, then it is assumed that the target is decoy, needs to abandon the target information, and whether judges the frame number lost
Otherwise big Mr. Yu's threshold value shows that target may be blocked if it is greater than then indicating that target has been lost, at this time with prediction
Target information replaces target information at this time.
The step 7:The prediction of target information mainly by the target informations of upper several frames to next frame target some
Information is judged, the position of next frame target, gray scale, area and boundary information are predicted, as judging that next frame target is true
Pseudo- foundation.
The step 8:The type heart of center, that is, target of target has been calculated by the region growing object type heart, because
If this present frame is true target, the type heart of the target need to only be exported, not lost if it is pseudo- target and target,
Export future position, if lose target, output center position, so far image trace flow terminate.
The method can be realized as follows:Target tracker includes FPGA, SSRAM, FLASH;Infrared detector sends red
Outer image is to field programmable gate array (FPGA), then after image procossing, target position that FPGA will be calculated
Output is set, FLASH is used for the software program and hardware program of loading field programmable gate array (FPGA).
It is obtained by the present invention to have the beneficial effect that:
The present invention removes the interference noise in image by medium filtering first, and window is adaptively adjusted by NIOS II
Size, reduces the calculation amount of algorithm, while effectively removing the influence of interference signal, then calculates mesh by region growing module
Target feature, is then sent into NIOS II systems by PIO interface, and NIOS II are true to target by the target signature information of acquisition
Puppet is judged, while adjusting the size and location of tracking window, is finally predicted according to the dbjective state of present frame and preceding four frame
Go out the status information of next frame target.The SOPC implementations based on FPGA are used, realization method is carried out according to algorithm spy
Classifying rationally, cooperative work of software and hardware realize target following.Its advantage is that suitable for the target following under complex background, real-time
It is high.
(I) present invention makes full use of target property, positioning accurate using a kind of track algorithm based on region growing
Really.The algorithm can accurately judge the true and false of target under complicated earth background, realize tenacious tracking.
(II) present invention uses the design scheme of SOPC, algorithm realization method of making rational planning for give full play at FPGA flowing water
The convenience of real-time and the software processing of reason, execution speed is fast, realizes real-time tracking.
(III) present invention employs adaptive threshold fuzziness and windowing to track, and improves the accuracy of tracking and greatly reduces
Calculation amount.
Description of the drawings
Figure one is that the present invention is based on the target tracker implementation flow charts of FPGA;
Figure two is that region growing FPGA of the present invention realizes block diagram;
Figure three is the hardware realization platform composition frame chart based on FPGA;
Figure four is real time analysis proof diagram of the present invention.
Specific implementation mode
The present invention is described further below in conjunction with drawings and examples.
Detecting and tracking method of the present invention based on FPGA includes the following steps:
Step 1: being filtered denoising to image using medium filtering
Since infrared image has more noise, when carrying out target following, medium filtering pair is first passed around
It does denoising, is ranked up to all data of current pixel 3*3 neighborhoods, and in-between value is taken to replace original pixel.
Step 2: carrying out region growing to target point obtains target area
This step is divided into two parts:Copy detection area image finds seed point;Region life is carried out in detection zone
It is long.
Step 2.1 copy detection area image finds seed point
The present invention realizes target following in window, can remove some other interference in this way while shorten at image
The time of reason.Since this paper is using the track algorithm based on region growing, need that processing is marked to image, therefore need
The image of a frame detection zone is cached, the infrared image of detection zone is stored in RAM in real time.Region growing needs one kind
Son point and growing threshold, as growth foundation and condition.In order to exclude the influence of false-alarm noise, the present invention is using eight neighborhood
The middle maximum point of mean value is used as seed point, the realization of the part that can be realized with flowing water inside FPGA.Growing threshold according to figure
As adaptively being adjusted, adjustment formula is:
The method of adjustment can effectively prevent that growing threshold is excessive to cause the appearance of false target and growing threshold too small
The problem of caused true target is lost.
Step 2.2 region growing
Region growing is realized with two FIFO, first FIFO stores the information of seed point, therefore when FIFO non-emptys,
Show still there is seed point not grow, then needs that it is continued to grow.The coordinate of second FIFO storage target point
Information, i.e., relative to the abscissa value of detection zone and ordinate value.The coordinate value calculates the target point by multiply-add operation
Address, then obtain the address of its four neighborhood, then read the gray value of the pixel from RAM successively;First determine whether the point
Whether grew, if having grown, other processing need not be done.If do not grown, need to sentence
It break whether within region growing range, if within growth scope, by the point coordinate information is stored to second
In a FIFO;Until all growth completion indicates all targets to the seed point in first FIFO when i.e. FIFO is empty
It has grown and has completed.
Region growing is completed using asynchronous FIFO, wherein it is the fifth harmonic for reading clock to write clock, for region growing
FPGA optimization processings are as follows:
First clock provides the address of the upper neighborhood of target point, judges whether the right neighborhood of a upper seed point is raw
It grew, and continued to determine whether to be target point if not growing, and needed to set FIFO with effect if it is target point, and will be upper
The right neighborhood of one seed point is stored in FIFO as seed point, is otherwise not processed.
Second clock provides the address of the left neighborhood of target point, judges whether the lower neighborhood of a upper seed point is raw
It grew, and continued to determine whether to be target point if not growing, and then needed to set FIFO with effect if it is target point, and will
The lower neighborhood of a upper seed point is stored in FIFO as seed point, is otherwise not processed.
Third clock provides the address of the right neighborhood of target point.
4th clock, provides the address of the lower neighborhood of target point, judges whether the upper neighborhood of current target point is raw
It grew, and continued to determine whether to be target point if not growing, and needed to set FIFO with effect if if it is target point, and
It is stored in FIFO using the upper neighborhood of current seed point as seed point, is otherwise not processed.
5th clock, judges whether the left neighborhood of current target point had grown, and continues if not growing
Judge whether to be target point, if it is target point, then need to set FIFO with effect, and using the left neighborhood of current seed point as
Seed point is stored in FIFO, is otherwise not processed.
Step 3: carrying out feature extraction to target
Clarification of objective is extracted, the gray scale of target, area, boundary, type heart information.The half-tone information of target needs entire mesh
Mark the average value of gray value;The area of target refers to the number of pixels that target is included;Boundary refers to target up and down
Boundary value can be asked by bubbling method;The location information of target is the position where the object type heart.Since target movement has
Relevance, therefore the information of the target of former frames should should have continuity with the information of current goal.
It is as follows to seek the area of target, gray scale, type heart calculation formula:
A indicates that the area of target, R indicate target area in formula,The abscissa of the object type heart and vertical seat are indicated respectively
Mark;Indicate average gray value.
Step 4: detection window adjusts
Tracking in the window that the present invention uses is interfered by opening a window effectively to eliminate, but the size of window needs basis
The size of objectives is adjusted.If the too small target that will cannot completely include of window causes target information inaccurate, window
Noise can then be introduced greatly by making a slip of the tongue, present invention employs the ratio and object boundary of window area and target area and window edge away from
From adjusting window size position in real time as basis for estimation again so that window, which can include target, as more as possible can must exclude to do
Disturb signal.
Step 5: carrying out object matching according to characteristic value
Step 5.1 Gray-scale Matching
Whether the average gray difference value for the target that the average gray of the target of this frame image is predicted with upper frame meets one
Confidence level is added one by fixed condition if meeting, otherwise constant.
It is step 5.2, area matched
Whether the area for the target that the average gray of the target of this frame image is predicted with upper frame meets in certain condition,
Confidence level is added one if meeting, it is otherwise constant.
Step 5.3, location matches
The target location of this frame image is compared with the picture position after prediction, judges whether its difference meets
Confidence level is added one by certain condition if meeting, otherwise constant.
Step 5.4, Boundary Match
The object boundary of this frame image is compared with the image boundary after prediction, judges whether its difference meets
Confidence level is added one by certain condition if meeting, otherwise constant.
Step 6: target genuine-fake differentiates
By above judgement, confidence level changes, when confidence level is not less than a certain threshold value, then it is assumed that the target is
True target, using the information of target write-in object chain as the foundation for predicting next frame target information, if it is less than the threshold value,
Then think that the target is decoy, need to abandon the target information, and judge the whether big Mr. Yu's threshold value of frame number lost, such as
Fruit, which is more than, then indicates that target has been lost, and otherwise shows that target may be blocked, is replaced at this time with the target information of prediction
Target information at this time.
Step 7: carrying out target information prediction according to movement locus
The prediction of target information mainly sentences some information of next frame target by the target information of upper several frames
It is disconnected, the position of next frame target, gray scale, area and boundary information are predicted, as the foundation for judging next frame target genuine-fake.
Step 8: target's center's position output.
The type heart of center, that is, target of target has been calculated by the region growing object type heart, so if currently
Frame is true target, then only need to export the type heart of the target, is not lost if it is pseudo- target and target, and mesh is predicted in output
Cursor position, if losing target, output center position.
So far image trace flow terminates.
The present invention is based on the target following calculations based on FPGA that the infrared object tracking system of FPGA can be provided below
It is realized on subtraction unit.The target tracker that the present invention provides is as shown, it includes FPGA, SSRAM, FLASH.
Infrared detector sends infrared image to field programmable gate array (FPGA), then passes through image procossing
Afterwards, FPGA exports the target location being calculated, and FLASH is used for the software program of loading field programmable gate array (FPGA)
And hardware program.
The present invention is verified with reference to experimental result:
By test, the accuracy and real-time of this system are verified.
The time that this system mainly needs is broadly divided into two parts to consider, a part is FPGA region growings part
Time, another part is NIOS II run times.By being tested in FPGA, it can be determined that going out the time meets the requirements,
Middle acquisition clock is 1M, needs a clock more than 3154 to complete altogether, i.e., 3.154ms meets real-time demand.
Claims (8)
1. a kind of detecting and tracking method based on FPGA, it is characterised in that:This method comprises the following steps:
Step 1: being filtered denoising to image using medium filtering;
Step 2: carrying out region growing to target point obtains target area;
Step 3: carrying out feature extraction to target;
Step 4: detection window adjusts;
Step 5: carrying out object matching according to characteristic value;
Step 6: target genuine-fake differentiates;
Step 7: carrying out target information prediction according to movement locus;
Step 8: target's center's position output;
The step 1 first passes around medium filtering and does denoising to it, to current pixel when carrying out target following
All data of 3*3 neighborhoods are ranked up, and in-between value is taken to replace original pixel;
The step 2 implementation method is as follows:
This step is divided into two parts:Copy detection area image finds seed point;Region growing is carried out in detection zone;
Step 2.1 copy detection area image finds seed point
Target following is realized in window, and using the track algorithm based on region growing, processing, caching one are marked to image
The infrared image of detection zone is stored in RAM by the image of frame detection zone in real time;Made using the maximum point of mean value in eight neighborhood
Realization for seed point, the part can be realized inside FPGA with flowing water;Growing threshold is adaptively adjusted according to image, is adjusted
Whole formula is:
Step 2.2 region growing
Region growing, the information of first FIFO storage seed point are realized with two FIFO, therefore when FIFO non-emptys, is shown
Still there is seed point not grow, then need that it is continued to grow;The coordinate information of second FIFO storage target point,
I.e. relative to the abscissa value of detection zone and ordinate value;The coordinate value calculates the ground of the target point by multiply-add operation
Then location obtains the address of its four neighborhood, then read the gray value of the pixel from RAM successively;First determine whether the seed point
Whether grew, if having grown, other processing need not be done;If do not grown, need to sentence
It break whether within region growing range, if within growth scope, the coordinate information of the seed point be stored to
In second FIFO;Until all growth completion indicates all to the seed point in first FIFO when i.e. FIFO is empty
Target point, which has been grown, to be completed;
Region growing is completed using asynchronous FIFO, wherein it is the fifth harmonic for reading clock to write clock, it is excellent for the FPGA of region growing
It is as follows to change processing:
First clock provides the address of the upper neighborhood of target point, judges whether the right neighborhood of a upper seed point had grown,
Continue to determine whether it is target point if not growing, need to set FIFO with effect if it is target point, and by upper one kind
The right neighborhood of son point is stored in FIFO as seed point, is otherwise not processed;
Second clock provides the address of the left neighborhood of target point, judges whether the lower neighborhood of a upper seed point had grown,
Continue to determine whether it is target point if not growing, then need to set FIFO with effect if it is target point, and by upper one
The lower neighborhood of seed point is stored in FIFO as seed point, is otherwise not processed;
Third clock provides the address of the right neighborhood of target point;
4th clock, provides the address of the lower neighborhood of target point, judges whether the upper neighborhood of current target point had grown,
Continue to determine whether it is target point if not growing, need to set FIFO with effect if if it is target point, and will work as
The upper neighborhood of preceding seed point is stored in FIFO as seed point, is otherwise not processed;
5th clock, judges whether the left neighborhood of current target point had grown, and continues to judge if not growing
Whether it is target point, if it is target point, then needs to set FIFO with effect, and using the left neighborhood of current seed point as seed
Point deposit FIFO, is otherwise not processed.
2. the detecting and tracking method described in accordance with the claim 1 based on FPGA, it is characterised in that:The step 3:Target
Feature extraction, the gray scale of target, area, boundary, type heart information;The half-tone information of target needs being averaged for entire target gray value
Value;The area of target refers to the number of pixels that target is included;Boundary refers to the boundary value up and down of target, passes through bubbling
Method acquires;The location information of target is the position where the object type heart;Since target movement has relevance, former frames
The information of target should should have continuity with the information of current goal;
It is as follows to seek the area of target, gray scale, type heart calculation formula:
A indicates that the area of target, R indicate target area in formula,The abscissa and ordinate of the object type heart are indicated respectively;Indicate average gray value.
3. the detecting and tracking method based on FPGA according to claim 2, it is characterised in that:The step 4:It uses
Tracking in window eliminates interference by opening a window, and the size needs of window are adjusted according to the size of objectives;Using window
The ratio and object boundary of area and target area adjust window size position in real time with window edge distance as basis for estimation,
It as more as possible can must exclude the interference signal again so that window can include target.
4. the detecting and tracking method described in accordance with the claim 3 based on FPGA, it is characterised in that:The step 5:
Step 5.1 Gray-scale Matching
Whether the average gray difference value for the target that the average gray of the target of this frame image is predicted with upper frame meets in certain item
Confidence level is added one by part if meeting, otherwise constant;
It is step 5.2, area matched
Whether the area for the target that the area of the target of this frame image is predicted with upper frame meets in certain condition, if met
Confidence level is then added one, it is otherwise constant;
Step 5.3, location matches
The target location of this frame image is compared with the picture position after prediction, judges whether its difference meets certain
Confidence level is added one by condition if meeting, otherwise constant;
Step 5.4, Boundary Match
The object boundary of this frame image is compared with the image boundary after prediction, judges whether its difference meets certain
Confidence level is added one by condition if meeting, otherwise constant.
5. the detecting and tracking method based on FPGA according to claim 4, it is characterised in that:The step 6:By with
On judgement, confidence level changes, when confidence level is not less than a certain threshold value, then it is assumed that the target is true target, by the mesh
Foundation of the object chain as prediction next frame target information is written in target information, if it is less than the threshold value, then it is assumed that the target
It for decoy, needs to abandon the target information, and judges the whether big Mr. Yu's threshold value of frame number lost, if it is greater than then indicating
Target has been lost, and otherwise shows that target may be blocked, and replaces target at this time to believe with the target information of prediction at this time
Breath.
6. the detecting and tracking method based on FPGA according to claim 5, it is characterised in that:The step 7:Target is believed
The prediction of breath is judged some information of next frame target by the target information of upper several frames, and next frame target is predicted
Position, gray scale, area and boundary information, as the foundation for judging next frame target genuine-fake.
7. the detecting and tracking method based on FPGA according to claim 6, it is characterised in that:The step 8:Target
The type heart of center, that is, target has been calculated by the region growing object type heart, so if present frame is true target, then only
The type heart of the target need to be exported, not lost if it is pseudo- target and target, export future position, if lost
Target, then output entirely detect the center position in the visual field, and so far image trace flow terminates.
8. the detecting and tracking method based on FPGA according to claim 7, it is characterised in that:The method can be real as follows
It is existing:Target tracker includes FPGA, SSRAM, FLASH;Infrared detector sends infrared image to FPGA, then passes through image
After processing, FPGA exports the target location being calculated, and FLASH is used to load the software program and hardware program of FPGA.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310753663.6A CN104751157B (en) | 2013-12-31 | 2013-12-31 | Detecting and tracking method based on FPGA |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310753663.6A CN104751157B (en) | 2013-12-31 | 2013-12-31 | Detecting and tracking method based on FPGA |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104751157A CN104751157A (en) | 2015-07-01 |
CN104751157B true CN104751157B (en) | 2018-07-27 |
Family
ID=53590813
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310753663.6A Active CN104751157B (en) | 2013-12-31 | 2013-12-31 | Detecting and tracking method based on FPGA |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104751157B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108269269A (en) * | 2016-12-30 | 2018-07-10 | 纳恩博(北京)科技有限公司 | Method for tracking target and device |
CN107341760A (en) * | 2017-06-27 | 2017-11-10 | 北京计算机技术及应用研究所 | A kind of low-altitude target tracking system based on FPGA |
CN109254941B (en) * | 2017-07-13 | 2020-04-24 | 凌云光技术集团有限责任公司 | FPGA-based serial signal clock synchronization method, serial-to-parallel conversion method and device |
CN107945214A (en) * | 2017-11-21 | 2018-04-20 | 杭州电子科技大学 | Mean shift video target tracking method based on SOPC |
CN109165600B (en) * | 2018-08-27 | 2021-11-26 | 浙江大丰实业股份有限公司 | Intelligent search platform for stage performance personnel |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1546993A (en) * | 2003-12-04 | 2004-11-17 | 上海交通大学 | Infrared target detecting, tracking and identifying system |
CN103325124A (en) * | 2012-03-21 | 2013-09-25 | 东北大学 | Target detecting and tracking system and method using background differencing method based on FPGA |
CN103400153A (en) * | 2013-07-15 | 2013-11-20 | 中国航天科工集团第三研究院第八三五八研究所 | Serial filtering matching method and system for real-time image identification |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7796780B2 (en) * | 2005-06-24 | 2010-09-14 | Objectvideo, Inc. | Target detection and tracking from overhead video streams |
-
2013
- 2013-12-31 CN CN201310753663.6A patent/CN104751157B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1546993A (en) * | 2003-12-04 | 2004-11-17 | 上海交通大学 | Infrared target detecting, tracking and identifying system |
CN103325124A (en) * | 2012-03-21 | 2013-09-25 | 东北大学 | Target detecting and tracking system and method using background differencing method based on FPGA |
CN103400153A (en) * | 2013-07-15 | 2013-11-20 | 中国航天科工集团第三研究院第八三五八研究所 | Serial filtering matching method and system for real-time image identification |
Also Published As
Publication number | Publication date |
---|---|
CN104751157A (en) | 2015-07-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110517288B (en) | Real-time target detection tracking method based on panoramic multi-path 4k video images | |
CN104751157B (en) | Detecting and tracking method based on FPGA | |
CN101739686B (en) | Moving object tracking method and system thereof | |
CN110766058B (en) | Battlefield target detection method based on optimized RPN (resilient packet network) | |
CN106526585B (en) | Tracking before target detection based on the filtering of Gaussian particle gesture probability hypothesis density | |
CN107169998A (en) | A kind of real-time tracking and quantitative analysis method based on hepatic ultrasound contrast enhancement image | |
CN105719283A (en) | Road surface crack image detection method based on Hessian matrix multi-scale filtering | |
CN108399430B (en) | A kind of SAR image Ship Target Detection method based on super-pixel and random forest | |
CN110503092B (en) | Improved SSD monitoring video target detection method based on field adaptation | |
CN116665095B (en) | Method and system for detecting motion ship, storage medium and electronic equipment | |
CN110706208A (en) | Infrared dim target detection method based on tensor mean square minimum error | |
CN109636810A (en) | Lung neoplasm dividing method, system and the region growing method of CT images, segmentation terminate judgment method, cutting adhesion region method | |
CN112329764A (en) | Infrared dim target detection method based on TV-L1 model | |
CN114998628A (en) | Template matching-based twin network long-term target tracking method | |
Huang et al. | Research on evaluation method of underwater image quality and performance of underwater structure defect detection model | |
CN116934808A (en) | River surface flow velocity measurement method based on water surface floater target tracking | |
CN111127506A (en) | Sequence image-based marine moving target comprehensive detection method | |
CN111241965A (en) | Target tracking method for occlusion detection based on multi-example learning | |
CN114155411A (en) | Intelligent detection and identification method for small and weak targets | |
Kong et al. | Adaptive image edge detection model using improved Canny algorithm | |
Liu et al. | Jellyfish recognition and density calculation based on image processing and deep learning | |
Niu et al. | Application of CEM algorithm in the field of tunnel crack identification | |
CN113869150A (en) | Infrared small target detection method for weighting local contrast of three-layer window | |
CN113834447A (en) | High-dynamic laser light bar self-adaptive imaging processing method under outdoor complex environment | |
Qian et al. | Specular reflections removal of gastrointestinal polyps based on endoscopic image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |