CN103177455A - Method for realizing KLT (Karhunen Loeve Transform) moving target tracking algorithm based on multicore DSP (Digital Signal Processor) - Google Patents
Method for realizing KLT (Karhunen Loeve Transform) moving target tracking algorithm based on multicore DSP (Digital Signal Processor) Download PDFInfo
- Publication number
- CN103177455A CN103177455A CN2013100897012A CN201310089701A CN103177455A CN 103177455 A CN103177455 A CN 103177455A CN 2013100897012 A CN2013100897012 A CN 2013100897012A CN 201310089701 A CN201310089701 A CN 201310089701A CN 103177455 A CN103177455 A CN 103177455A
- Authority
- CN
- China
- Prior art keywords
- image
- core
- window
- dsp
- klt
- 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.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a method for realizing a KLT (Karhunen Loeve Transform) moving target tracking algorithm based on a multicore DSP (Digital Signal Processor). The method disclosed by the invention comprises the steps of: image smoothing, pyramid graph calculating, image gray value difference calculating, feature point selecting, feature point tracking and feature point consistency checking; and six modules for executing the steps are transplanted into the multicore DSP, wherein one core is allocated to each module, and the processing is carried out in a datastream manner. According to the method, a KLT tracking algorithm is adopted, and the advantage and processing speed of multicore parallel execution of the multicore DSP are used, so that the real-time and accurate tracking of moving targets can be realized.
Description
Technical field
The invention belongs to infrared target search tracking technique field, particularly a kind of moving object real-time tracking field of specifically belonging to.
Background technology
The detection of moving target, tracking are piths of digital image processing techniques.It has comprised a plurality of subject contents such as pattern-recognition, computer vision, vision signal processing, is also increasing to the application of this technology.Recent years, people continually develop new algorithm in the detection and tracking of moving target, but due on the calculated amount of algorithm and the constraint of the processing speed of hardware platform, make two Key Performance Indicators of motion target tracking: the balance of real-time and accuracy is difficult to grasp.In order to realize the requirement of real-time in Motion Object Tracking System, need to take some measures to improve the calculation process speed of algorithm, the one, be optimized on algorithm, the 2nd, improve on hardware system.
The algorithm of motion target tracking is a lot of at present, follow the tracks of or cry the Lucas optical flow method as particle filter (pf), meanshift tracking and KLT, each own advantage separately of these methods, for particle filter, it can be reasonable in global search to optimum solution, but that it finds the solution speed is relatively slow, because being based on color histogram, it plays calculating, so not too can distinguish the same color thing, meanshift follows the tracks of and is easy to be absorbed in local optimum, but speed is still very fast.It is also good that the KLT method shows aspect tracking, has very strong antijamming capability, especially on real-time computing velocity.1981, Kannde and Lucas have proposed track algorithm (the 1. Jianbo Shi based on unique point, Carlo Tomasi, Good Features to Track, IEEE Conference on Computer Vision and Pattern Recognition, 1994, 6.), Lucas and Tomasi expanded this algorithm afterwards, perfect (2.Tiziano Tommasini, Andrea Fusiello, Emanuele Trucco etc., Making Good Features Track Better, Proc. IEEE Int. Conf. on Computer Vision and Pattern Recognition, 1998 (6): 178 ~ 183.).KLT feature point tracking algorithm is followed the tracks of tracking target material texture-rich " corner point ", therefore has very strong anti-interference, and trackability is good, and practical application has also confirmed these characteristics.Although it is moderate that KLT feature point tracking algorithm calculates, if realize on existing hardware platform with software, also be difficult to reach requirement of real time.The KLT algorithm realizes that on GPU arithmetic speed increases, but be based on the consideration of cost, to use widely and also have certain difficulty (3.Sudipta N Sinba, Janmichael Frahm, Marc Pollefeys, et al.GPU-based video feature tracking and matching:TR 06-012 [R] .University of North Carolina, USA, 2006.).Sudipta N (4.Sudipta N. Sinha, Jan-Michael Frahm etc., GPU-based Video Fearure Tracking and Matching, Department of Computer Science, UNC Chapel Hill, Technical Report TR 06-012,2006,5.) etc. on P-IV 3GHz CPU computing machine, the picture of 1024*768 size is carried out KLT and follows the tracks of, calculate spended time as shown in the table:
Table one. the feature corresponding calculating spended time of counting
| Feature count (individual) | The time (ms) of average every frame cost |
| 100 | 500 |
| 400 | 550 |
| 750 | 600 |
| 1000 | 645 |
The modification of depending merely on as can be seen from the above table on algorithm does not reach requirement, therefore need to be broken through on hardware.
Following table has provided the KLT algorithm in each platform working time:
Table two KLT algorithm is at each platform working time (ms)
| Universal cpu | OpenCV | GPU | DSP | |
| Hardware platform | E7400 2.80GHz | E7400 2.80GHz | NVidia Geforece 7900GTX | TMS320C6678 |
| Algorithm execution time | 82.87 | 78.13 | 40 | 35.1 |
Summary of the invention
The object of the present invention is to provide a kind of employing KLT track algorithm, use advantage that the multi-core parallel concurrent of multi-core DSP carries out and its processing speed can realize moving target in real time, accurate implementation method based on the KLT Moving Target Tracking Algorithm of multi-core DSP of following the tracks of.
The technical solution that realizes the object of the invention is:
A kind of implementation method of the KLT Moving Target Tracking Algorithm based on multi-core DSP, step is as follows:
Step 1: utilize the video acquisition plate with network interface and SRIO interface that the video image that CCD obtains is transferred in multi-core DSP by the SRIO interface;
Step 2: the video image of input is carried out smooth operation, and namely input picture being carried out standard deviation is 0.7, and template window is
Gaussian filtering is realized the level and smooth of image, improves signal to noise ratio (S/N ratio) and the signal to noise ratio of target;
Step 3: will be the pyramid diagram picture through level and smooth image transitions, then the pyramid diagram of every one-level be looked like to ask unique point, and computed image gray difference value;
Step 4: the Characteristic of Image point is chosen, and namely gray level image is carried out the second order differentiate, obtains
Matrix, wherein
Matrix is
, wherein
The expression gray level image,
Be single order in certain neighborhood
Directional derivative,
Single order for correspondence
Directional derivative is then according to formula
Obtain this second-order matrix two eigenwerts (
,
) size judges whether can be used as unique point;
Concrete criterion is as follows:
(1) if
,
All hour, this zone is approximately the flat region;
(2) if
,
When differing larger, illustrate that this zone is fringe region;
(3) if
The time, illustrate that this point is the validity feature point, wherein,
Be the threshold value of oneself setting;
Step 5: utilize
The whole image of window gradient matrix traversal is chosen the window of texture-rich, namely chooses local irregularities in image and the window that has regular characteristic on macroscopic view, and the validity feature point the chosen size according to its eigenwert is sorted;
Step 6, the calculating of picture point side-play amount, any at image
The time chart picture frame
With
The time chart picture frame
In the position satisfy
Wherein
,
, be the coordinate of the pixel of image,
Be a bit of moment,
,
For the horizontal ordinate of pixel and the displacement of ordinate, namely exist
In each pixel, can by
The pixel translation of middle respective window
Obtain, purpose is obtained the pixel translational movement exactly
Step 7: carry out consistency check with Affine arithmetic for following the tracks of successful unique point;
The present invention compared with prior art, its remarkable advantage:
(1) in the feature point tracking step, in order to improve the speed of tracking, guarantee that simultaneously the unique point that traces into has of overall importance, smoothed image is converted to the pyramid diagram picture, ask the gradient of every one-level pyramid diagram picture, the unique point pointwise of then choosing is followed the tracks of step by step again.
(2) used Inline Function when optimized algorithm, Inline Function execution efficient is high, and code length is short, and easy to use, and the built-in function that has called DSP, makes the code operational efficiency high, and elapsed time is short.
(3) utilized eight nuclear superiority of multi-core DSP-TMS320C6678, whole algorithm is divided into several parts, executed in parallel in multinuclear, improved whole efficient respectively, shortened operation time, thereby realize real-time tracking.
(4) the tasks carrying Operation Mode Selection of DSP multinuclear be stream socket, namely task is moved according to the transmission of data, a task promotes another task run.This mode relatively is fit to the higher application of requirement of real-time.
Below in conjunction with accompanying drawing, the present invention is described in further detail.
Description of drawings
Fig. 1 is the schematic flow sheet of implementation method that the present invention is based on the KLT Moving Target Tracking Algorithm of multi-core DSP.
Fig. 2 is system hardware one-piece construction figure of the present invention.
Fig. 3 is the consistency check process flow diagram that the present invention follows the tracks of successful unique point.
Fig. 4 is each core maintenance data stream mode method of operation of the present invention.
Fig. 5 is the cut-away view of the multi-core DSP that uses of the present invention.
Embodiment
In conjunction with Fig. 1, the present invention is based on the implementation method of the KLT motion mark track algorithm of multi-core DSP, step is as follows:
The first step, owing to need to every frame of video being processed, therefore the data volume of processing is larger, algorithm complex is high, traditional processor generally can not satisfy the requirement of speed, that native system adopts is the high high speed digital signal processor eight core TMS320C6678 of a cost performance of the up-to-date release of TI company, introduced its inner structure as Fig. 5, can find out that it has powerful multiply-add operation device and parallel processing structure, this processor is eight core floating type DSP, and its each core maximum operation frequency reaches 1.25GHz.The single instruction cycle can be carried out 32 fixed-point data computings, perhaps carries out 16 floating data computings.Whole chip provides 320GMAC fixed point calculation or 160GFLOP Floating-point Computation ability.
As shown in Figure 2, whole hardware system is divided for four modules such as image capture module, image pretreatment module, Digital Image Processing module, display modules, at first utilize the video acquisition plate that the video image that CCD obtains is transferred in DSP-TMS320C6678 by SRIO, compatibility between two plates and error rate are all considerable, namely transmit image data packet loss seldom almost in the process of transmission, can reach requirement.
Second step carries out smooth operation with the video image of input in DSP inside, namely input picture is carried out standard deviation and is
Template window is
Gaussian filtering is realized the level and smooth of image, can not lose the energy of target, can weaken to a great extent again the energy of background, thereby improve signal to noise ratio (S/N ratio) and the signal to noise ratio of target, reaches the effect of Background suppression.
The 3rd step will be the pyramid diagram picture through level and smooth image transitions, then the pyramid diagram of every one-level be looked like to ask unique point, and computed image gray difference value;
In the 4th step, the Characteristic of Image point is chosen, and namely gray level image is carried out the second order differentiate, obtains
Then matrix judges whether can be used as unique point according to two eigenwert sizes of this matrix.The below has provided
The specific definition of matrix.
Suppose
The expression gray level image
Be single order in certain neighborhood
Directional derivative,
Single order for correspondence
Directional derivative,
Matrix expression is as follows
If
,
For
Two eigenwerts of matrix specifically judge as follows:
(1) if
,
All hour, this zone is approximately the flat region.
(2) if
,
When differing larger, illustrate that this zone is fringe region.
(3) if
(wherein
The threshold value of setting is fixed by oneself) time, illustrate that this point is the validity feature point.
In the 5th step, utilize
The whole image of window gradient matrix traversal is chosen the window of texture-rich, and the size of the unique point that chooses according to eigenwert sorted, and forms the list of unique point, is used for when feature point tracking the point of place of lost.
The 6th step, the calculating of picture point side-play amount,
The time chart picture frame
With
The time chart picture frame
In the position satisfy
Namely exist
In each pixel, can by
The pixel translation of middle respective window
Obtain, purpose is obtained exactly
Specifically being calculated as follows of side-play amount:
Suppose
Characteristic window constantly is
, wherein
Be window coordinates,
Characteristic window constantly is
Due to reasons such as noises, have
, wherein
For in the time
The interior noise that produces that changes due to illumination condition.
Will
Square and in whole window upper integral, just obtained the gray scale difference quadratic sum (SSD) of video in window
(1)
Wherein,
,
,
Usually can be taken as 1.If emphasize the effect of core texture adopt gauss of distribution function, this patent
Adopt gauss of distribution function.
When
For with
When comparing insignificant a small amount of, will
Taylor expansion is removed high-order term, obtains
With formula (2) substitution formula (1), and right simultaneously to the both sides of formula (1)
Get 0 after differentiate, can obtain
At this moment
The minimal value of getting.Variable being changed to of formula (3)
If order
Formula (4) can be expressed as
（7）
For every two width images, solve an equation (7) can obtain the displacement of characteristic window
In the 7th step, as Fig. 3, need to carry out consistency check with Affine arithmetic for following the tracks of successful unique point.Because the tracking of unique point realizes by multiple image, the information of image tends to distorted, therefore needs consistency check.For the signature tracking from the frame to the frame, carry out consistency check with simple translation transformation not enough often, therefore select affine maps to realize conforming inspection.
the 8th step, requirement due to real-time and accuracy, requirement to hardware platform is very high, especially require high to the processing speed of chip, therefore above-mentioned algorithm is resolved into several parts, respectively the level and smooth of image, calculate pyramid diagram, computed image gray difference value, the unique point consistency check, unique point is selected, six modules of feature point tracking, consider the calculated amount of each part and coupling and the associativity between every part, this algorithm has been resolved into six modules to be transplanted in DSP-TMS320C6678, core of each module assignment wherein, the maintenance data stream method of operation such as Fig. 4, the data that first core is handled well pass to second core, the data of successively previous core being handled well pass to next core.the communication of intermodule realizes by information transmission mechanism, call the DSP Integrated Development Environment SYS/BIOS and IPC instrument are provided, SYS/BIOS mainly completes internuclear task scheduling, IPC realize internuclear synchronously with communicate by letter, receive the sequence image that transmits from the video acquisition plate by UDP, and to the above-mentioned algorithm process of these data applications, utilize the optimization of the powerful fixed and floating hybrid operation ability of C66x and multinuclear streamline, realized the real-time follow-up of moving target, then the result that will process is transferred to upper realization of PC by network interface and shows in real time, with circle the aiming circle that is traced to, use little square and indicate unique point in circle.
Claims (3)
1. implementation method based on the KLT Moving Target Tracking Algorithm of multi-core DSP is characterized in that comprising the following steps:
Step 1: utilize the video acquisition plate with network interface and SRIO interface that the video image that CCD obtains is transferred in multi-core DSP by the SRIO interface;
Step 2: the video image of input is carried out smooth operation, and namely input picture being carried out standard deviation is 0.7, and template window is
Gaussian filtering is realized the level and smooth of image, improves signal to noise ratio (S/N ratio) and the signal to noise ratio of target;
Step 3: will be the pyramid diagram picture through level and smooth image transitions, then the pyramid diagram of every one-level be looked like to ask unique point, and computed image gray difference value;
Step 4: the Characteristic of Image point is chosen, and namely gray level image is carried out the second order differentiate, obtains
Matrix, wherein
Matrix is
, wherein
The expression gray level image,
Be single order in certain neighborhood
Directional derivative,
Single order for correspondence
Directional derivative is then according to formula
Obtain this second-order matrix two eigenwerts (
,
) size judges whether can be used as unique point;
Concrete criterion is as follows:
(1) if
,
All hour, this zone is approximately the flat region;
(2) if
,
When differing larger, illustrate that this zone is fringe region;
(3) if
The time, illustrate that this point is the validity feature point, wherein,
Be the threshold value of oneself setting;
Step 5: utilize
The whole image of window gradient matrix traversal is chosen the window of texture-rich, namely chooses local irregularities in image and the window that has regular characteristic on macroscopic view, and the validity feature point the chosen size according to its eigenwert is sorted;
Step 6, the calculating of picture point side-play amount, any at image
The time chart picture frame
With
The time chart picture frame
In the position satisfy
Wherein
,
, be the coordinate of the pixel of image,
Be a bit of moment,
,
For the horizontal ordinate of pixel and the displacement of ordinate, namely exist
In each pixel, can by
The pixel translation of middle respective window
Obtain, purpose is obtained the pixel translational movement exactly
Step 7: carry out consistency check with Affine arithmetic for following the tracks of successful unique point;
step 8, above-mentioned algorithm is resolved into several parts, respectively the level and smooth of image, calculate pyramid diagram, computed image gray difference value, unique point is selected, feature point tracking, six modules of unique point consistency check, these six modules are transplanted in multi-core DSP, core of each module assignment wherein, maintenance data stream mode is processed, being task moves according to the transmission of data, a task promotes the another one task run, specifically the data handled well of first core pass to second core, the data of successively previous core being handled well pass to next core, the communication of intermodule realizes by message passing mechanism.
2. the implementation method of KLT Moving Target Tracking Algorithm based on multi-core DSP described according to right 1, be characterised in that: step 1 is described, utilization has the video acquisition plate of network interface and SRIO interface, by SRIO with transmission of video images to DSP, video acquisition plate and dsp board carry out data transmission by udp protocol.
3. the implementation method of KLT Moving Target Tracking Algorithm based on multi-core DSP described according to right 1 is characterised in that: the side-play amount of computed image unique point, and with the tracking of realization character point, specifically being calculated as follows of side-play amount:
Suppose
Characteristic window constantly is
, wherein
Be window coordinates,
Characteristic window constantly is
, due to reasons such as noises, have
, wherein
For in the time
The interior noise that produces that changes due to illumination condition,
Will
Square and in whole window upper integral, just obtained the gray scale difference quadratic sum of video in window
(1)
Wherein,
,
,
Usually can be taken as 1, if emphasize the effect of core texture adopt gauss of distribution function, this patent
Adopt gauss of distribution function;
When
For with
When comparing insignificant a small amount of, will
Taylor expansion is removed high-order term, obtains
With formula (2) substitution formula (1), and right simultaneously to the both sides of formula (1)
Get 0 after differentiate, can obtain
At this moment
The minimal value of getting, variable being changed to of formula (3)
If order
Formula (4) can be expressed as
For every two width images, solve an equation (7) can obtain the displacement of characteristic window
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310089701.2A CN103177455B (en) | 2013-03-20 | 2013-03-20 | Based on the implementation method of the KLT Moving Target Tracking Algorithm of multi-core DSP |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310089701.2A CN103177455B (en) | 2013-03-20 | 2013-03-20 | Based on the implementation method of the KLT Moving Target Tracking Algorithm of multi-core DSP |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103177455A true CN103177455A (en) | 2013-06-26 |
| CN103177455B CN103177455B (en) | 2016-04-20 |
Family
ID=48637285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310089701.2A Expired - Fee Related CN103177455B (en) | 2013-03-20 | 2013-03-20 | Based on the implementation method of the KLT Moving Target Tracking Algorithm of multi-core DSP |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103177455B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103369207A (en) * | 2013-06-30 | 2013-10-23 | 浙江元亨通信技术股份有限公司 | Multi-nuclear DSP (Digital Signal Processor) video enhancer |
| CN103473532A (en) * | 2013-09-06 | 2013-12-25 | 上海大学 | Pedestrian fast detection system and method of vehicle-mounted platform |
| CN104978728A (en) * | 2014-04-08 | 2015-10-14 | 南京理工大学 | Image matching system of optical flow method |
| CN105096285A (en) * | 2014-05-23 | 2015-11-25 | 南京理工大学 | Image fusion and target tracking system based on multi-core DSP |
| CN106157327A (en) * | 2015-04-16 | 2016-11-23 | 南京理工大学 | A kind of motion target tracking method based on matrix low-rank representation theory |
| CN107111878A (en) * | 2016-06-07 | 2017-08-29 | 深圳市大疆创新科技有限公司 | Data processing method, apparatus and system |
| CN108476321A (en) * | 2017-04-26 | 2018-08-31 | 深圳市大疆创新科技有限公司 | Image processing method and equipment |
| CN109978801A (en) * | 2019-03-25 | 2019-07-05 | 联想(北京)有限公司 | A kind of image processing method and image processing apparatus |
| WO2020216249A1 (en) * | 2019-04-23 | 2020-10-29 | 深圳市道通智能航空技术有限公司 | Target tracking and processing method based on multiple processors |
| CN113763295A (en) * | 2020-06-01 | 2021-12-07 | 杭州海康威视数字技术股份有限公司 | Image fusion method, method and device for determining image offset |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101561270A (en) * | 2009-05-27 | 2009-10-21 | 天津理工大学 | Embedded omnidirectional ball vision object detection and mobile monitoring system and embedded omnidirectional ball vision object detection and mobile monitoring method |
| CN101923719A (en) * | 2009-06-12 | 2010-12-22 | 新奥特（北京）视频技术有限公司 | Particle filter and light stream vector-based video target tracking method |
| CN102148934A (en) * | 2011-04-02 | 2011-08-10 | 北京理工大学 | Multi-mode real-time electronic image stabilizing system |
| CN101923718B (en) * | 2009-06-12 | 2012-12-19 | 新奥特（北京）视频技术有限公司 | Optimization method of visual target tracking method based on particle filtering and optical flow vector |
-
2013
- 2013-03-20 CN CN201310089701.2A patent/CN103177455B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101561270A (en) * | 2009-05-27 | 2009-10-21 | 天津理工大学 | Embedded omnidirectional ball vision object detection and mobile monitoring system and embedded omnidirectional ball vision object detection and mobile monitoring method |
| CN101923719A (en) * | 2009-06-12 | 2010-12-22 | 新奥特（北京）视频技术有限公司 | Particle filter and light stream vector-based video target tracking method |
| CN101923718B (en) * | 2009-06-12 | 2012-12-19 | 新奥特（北京）视频技术有限公司 | Optimization method of visual target tracking method based on particle filtering and optical flow vector |
| CN102148934A (en) * | 2011-04-02 | 2011-08-10 | 北京理工大学 | Multi-mode real-time electronic image stabilizing system |
Non-Patent Citations (3)
| Title |
|---|
| JIANBO SHI等: "《Good Features to Track》", 《IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》, 30 June 1994 (1994-06-30), pages 593 - 600 * |
| TIZIANO TOMMASINI等: "《Making Good Features Track Better》", 《IEEE INT. CONF. ON COMPUTER VISION AND PATTERN RECOGNITION》, 30 June 1998 (1998-06-30), pages 178 - 183 * |
| 段鸿等: "《基于Kanade-Lucas-Tomasi算法的人脸特征点跟踪方法》", 《计算机辅助设计与图形学学报》, vol. 16, no. 3, 31 March 2004 (2004-03-31), pages 279 - 283 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103369207A (en) * | 2013-06-30 | 2013-10-23 | 浙江元亨通信技术股份有限公司 | Multi-nuclear DSP (Digital Signal Processor) video enhancer |
| CN103473532A (en) * | 2013-09-06 | 2013-12-25 | 上海大学 | Pedestrian fast detection system and method of vehicle-mounted platform |
| CN103473532B (en) * | 2013-09-06 | 2016-11-23 | 上海大学 | The quick pedestrian detecting system of airborne platform and method |
| CN104978728A (en) * | 2014-04-08 | 2015-10-14 | 南京理工大学 | Image matching system of optical flow method |
| CN104978728B (en) * | 2014-04-08 | 2017-11-14 | 南京理工大学 | A kind of image matching system of optical flow method |
| CN105096285A (en) * | 2014-05-23 | 2015-11-25 | 南京理工大学 | Image fusion and target tracking system based on multi-core DSP |
| CN106157327A (en) * | 2015-04-16 | 2016-11-23 | 南京理工大学 | A kind of motion target tracking method based on matrix low-rank representation theory |
| CN106157327B (en) * | 2015-04-16 | 2018-12-14 | 南京理工大学 | A kind of motion target tracking method based on matrix low-rank representation theory |
| US10909690B2 (en) | 2016-06-07 | 2021-02-02 | SZ DJI Technology Co., Ltd. | Data processing method, apparatus, and system |
| CN107111878A (en) * | 2016-06-07 | 2017-08-29 | 深圳市大疆创新科技有限公司 | Data processing method, apparatus and system |
| CN107111878B (en) * | 2016-06-07 | 2019-05-24 | 深圳市大疆创新科技有限公司 | Data processing method, apparatus and system |
| CN108476321A (en) * | 2017-04-26 | 2018-08-31 | 深圳市大疆创新科技有限公司 | Image processing method and equipment |
| CN109978801A (en) * | 2019-03-25 | 2019-07-05 | 联想(北京)有限公司 | A kind of image processing method and image processing apparatus |
| CN109978801B (en) * | 2019-03-25 | 2021-11-16 | 联想(北京)有限公司 | Image processing method and image processing device |
| WO2020216249A1 (en) * | 2019-04-23 | 2020-10-29 | 深圳市道通智能航空技术有限公司 | Target tracking and processing method based on multiple processors |
| CN113763295A (en) * | 2020-06-01 | 2021-12-07 | 杭州海康威视数字技术股份有限公司 | Image fusion method, method and device for determining image offset |
| CN113763295B (en) * | 2020-06-01 | 2023-08-25 | 杭州海康威视数字技术股份有限公司 | Image fusion method, method and device for determining image offset |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103177455B (en) | 2016-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103177455B (en) | Based on the implementation method of the KLT Moving Target Tracking Algorithm of multi-core DSP | |
| CN105631798B (en) | Low Power Consumption Portable realtime graphic object detecting and tracking system and method | |
| CN109100730B (en) | Multi-vehicle cooperative rapid map building method | |
| CN101692257B (en) | Method for registering complex curved surface | |
| CN102148934B (en) | Multi-mode real-time electronic image stabilizing system | |
| CN104978728B (en) | A kind of image matching system of optical flow method | |
| CN102438118B (en) | High-speed vision capture apparatus of moving object characteristic | |
| CN104778690A (en) | Multi-target positioning method based on camera network | |
| CN110222769A (en) | A kind of Further aim detection method based on YOLOV3-tiny | |
| CN104484868B (en) | The moving target of a kind of combination template matches and image outline is taken photo by plane tracking | |
| CN106686278A (en) | Heterogeneous processor parallel coordination processing method facing real-time object identification | |
| CN101742545A (en) | Positioning method and system in WiFi environment | |
| CN110222307B (en) | Parallel implementation method for eigenvalue decomposition of real symmetric matrix based on FPGA | |
| CN103065330B (en) | Based on particle filter method for tracking target and the device of pipeline and parallel design technology | |
| CN103150559A (en) | Kinect three-dimensional depth image-based head identification and tracking method | |
| CN205486304U (en) | Portable realtime graphic object detection of low -power consumption and tracking means | |
| CN102494675A (en) | High-speed visual capturing method of moving target features | |
| CN109410330A (en) | One kind being based on BIM technology unmanned plane modeling method | |
| CN103700117A (en) | Robust optical flow field estimating method based on TV-L1 variation model | |
| CN102201122B (en) | Motion capture system, data noise reduction method and system of motion capture | |
| CN102194248A (en) | Method for detecting and responding false-true collision based on augmented reality | |
| CN103617624B (en) | The real-time global search method based on cooperative target of measuring for SPEED VISION | |
| CN104978749A (en) | FPGA (Field Programmable Gate Array)-based SIFT (Scale Invariant Feature Transform) image feature extraction system | |
| CN103345765B (en) | Based on moving object detection devices and methods therefor under the mobile platform of DSP+FPGA | |
| CN108282597A (en) | A kind of real-time target tracing system and method based on FPGA |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160420 Termination date: 20180320 |