CN110276836A - A kind of method and MR mixed reality intelligent glasses accelerating characteristic point detection - Google Patents
A kind of method and MR mixed reality intelligent glasses accelerating characteristic point detection Download PDFInfo
- Publication number
- CN110276836A CN110276836A CN201810204060.3A CN201810204060A CN110276836A CN 110276836 A CN110276836 A CN 110276836A CN 201810204060 A CN201810204060 A CN 201810204060A CN 110276836 A CN110276836 A CN 110276836A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- mixed reality
- chain
- module
- 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 42
- 239000011521 glass Substances 0.000 title claims abstract description 20
- 238000001514 detection method Methods 0.000 title abstract description 47
- 238000012545 processing Methods 0.000 claims abstract description 61
- 230000008569 process Effects 0.000 claims abstract description 18
- 238000005457 optimization Methods 0.000 claims description 60
- 238000007781 pre-processing Methods 0.000 claims description 32
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 claims description 11
- 239000010931 gold Substances 0.000 claims description 11
- 229910052737 gold Inorganic materials 0.000 claims description 11
- 230000008859 change Effects 0.000 claims description 4
- 235000013399 edible fruits Nutrition 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract description 7
- 238000010586 diagram Methods 0.000 description 13
- 230000008901 benefit Effects 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 239000004973 liquid crystal related substance Substances 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 229910052710 silicon Inorganic materials 0.000 description 2
- 239000010703 silicon Substances 0.000 description 2
- 241000196324 Embryophyta Species 0.000 description 1
- 241001136824 Pyrgotidae Species 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
- G06F3/013—Eye tracking input arrangements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/006—Mixed reality
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
Technical solution of the present invention provides a kind of method and MR mixed reality intelligent glasses that can accelerate characteristic point detection, the processing mode that mixed reality processor MPU is combined by parallel processing and pipeline processes, MR mixed reality pixel stream is handled, obtain characteristic point, primary processor exports MR application image after carrying out the operation of MR pose to characteristic point.Increase the handling capacity of data by pipeline processing mode, parallel processing manner improves data-handling efficiency.The detection characteristic point time that technical solution obtains through the invention, it is to calculate to complete in image pyramid compared with the total time of all image characteristic points with the detection characteristic point time in the prior art, it improves and extracts feature spot speed, to solve the problems, such as that image pose calculates, tracks Caton.
Description
Technical field
The present invention relates to characteristic point detection fields, more particularly to a kind of method that characteristic point can be accelerated to detect and MR mixing are now
Real intelligent glasses.
Background technique
With the fast development of Virtual Reality, augmented reality AR and mixed reality MR, to the processing speed of graph image
Requirement with accuracy rate is also in fast lifting.During to graph and image processing, particularly important (the feature detection of feature detection
Be link most time-consuming in MR graph and image processing, account for about the 50% of total operation time), characteristic point detection length velocity relation to pair
The height of the tracking efficiency of graph image.
Current feature point detecting method is: processor (CPU) carries out simple process to collected graphic image data,
Invalid information therein is removed, data volume is reduced with this.Image pyramid is generated later, after being generated to image pyramid,
Start to images all in image pyramid while carrying out data processing.Specifically, carrying out the seat of characteristic point to pyramid diagram picture
Mark is calculated, and calculates individual features point weight later, finally calculates the centroid angle of characteristic point, and image is obtained after the completion of all calculating
Pyramidal set of characteristic points, and unified output.Finally the detection characteristic point time is institute in calculating completion image pyramid in this way
There is the total time of image characteristic point.
During processing, due to calculate all pixels in image pyramid, due to the figure of VR, AR and MR
Shape image data amount is huge, when extracting characteristic point using the above method, will cause that processing speed is slow, and processor (CPU) is grown
Time occupy, result in it is subsequent according to characteristic point carry out the calculating of image pose, tracking when Caton the problem of.
Summary of the invention
The present invention provides a kind of method and MR mixed reality intelligent glasses that can accelerate characteristic point detection, passes through to be promoted and extract
Feature spot speed solves the problems, such as that image pose calculates, tracks Caton.
To achieve the goals above, the embodiment of the present invention adopts the following technical scheme that
In a first aspect, technical solution of the present invention provides a kind of method that characteristic point can be accelerated to detect, which comprises
Preprocessing Algorithm chain pre-processes MR mixed reality pixel stream, obtains pre-processed results.Core algorithm chain by locating parallel
The mode of reason detects the pre-processed results, obtains characteristic point.
In the first mode in the cards, implementation with reference to first aspect, the core algorithm chain is by simultaneously
The mode of row processing detects the pre-processed results, obtains characteristic point, comprising: core algorithm chain is raw according to representative value N
At core algorithm chain array.Core algorithm chain array replicates the pre-processed results, obtains pre-processed results array.Core
Each core algorithm chain in center algorithm chain array, respectively detects each pre-processed results in pre-processed results array,
Obtain characteristic point.
Wherein, core algorithm chain is according to representative value N, by self-replication N/2 times (N >=2 and be even number);Pre-processed results quilt
The number of duplication is identical as the quantity of core algorithm chain in core algorithm chain array.
In second of mode in the cards, implementation with reference to first aspect, Preprocessing Algorithm chain and core are calculated
Each optimization algorithm module in method chain has dynamic two-level cache, and for receiving, by previous optimization algorithm module, treated
MR pixel data, and form MR picture element caching data.The MR picture element caching data are carried out in optimization algorithm module parallel
Processing, obtains module arithmetic result.The dynamic two-level cache includes level cache and L2 cache.
In the third mode in the cards, second with reference to first aspect is possible, described in optimization algorithm module
In to the MR picture element caching data carry out parallel processing, obtain module arithmetic result, comprising: according to MR mixed reality pixel stream
Resolution ratio increases the computing circuit positive integer times in the optimization algorithm module, obtains computing circuit array.Computing circuit battle array
Each computing circuit in column carries out data processing to the MR picture element caching data by pipeline processing mode, obtains described
Module arithmetic result.
In the 4th kind of mode in the cards, implementation and first aspect with reference to first aspect first, second,
The third is possible, the method also includes: primal algorithm file in primal algorithm module is compiled, compiling result is obtained and makees
For gold reference model.According to the type of the primal algorithm, increase the dynamic two-level cache wherein.To the original calculation
French part is rewritten, and optimization algorithm file is obtained.Whether the optimization algorithm file is judged according to the gold reference model
Correctly.If correct, the optimization algorithm file download to programmable gate array is generated into hardware circuit, is obtained described excellent
Change algoritic module.
Second aspect, technical solution of the present invention provide a kind of MR mixed reality Brilliant Eyes that characteristic point can be accelerated to detect
Mirror, comprising: binocular depth camera module, for acquiring MR mixed reality pixel stream.Mixed reality processor is calculated for pre-processing
Method chain pre-processes the MR mixed reality pixel stream that the binocular depth camera module acquires, and obtains pretreatment knot
Fruit is also used to core algorithm chain and is detected by way of parallel processing to the pre-processed results, obtains characteristic point.Main place
Device is managed, the characteristic point for obtaining the mixed reality processor exports MR application image after carrying out the operation of MR pose.
In the first mode in the cards, in conjunction with the implementation of second aspect, the mixed reality processor packet
It includes: pretreatment circuit, for being acquired by the optimization algorithm module in Preprocessing Algorithm chain to the binocular depth camera module
The MR mixed reality pixel stream pre-processed, obtain the pre-processed results.Generator, for the core algorithm
Chain generates core algorithm chain array, is also used to replicate the pre-processed results according to representative value N, obtains pretreatment knot
Fruit array.According to representative value N, the core algorithm chain is replicated N/2 time (N >=2 and be even number), generation core algorithm chain battle array
Column, replicate the pre-processed results, generate pre-processed results array;It is also used to raw according to the core algorithm chain array
At parallel circuit, and it is cured in programmable gate chip.Programmadle logic door parallel circuit, for passing through the generator
Optimization algorithm module in the core algorithm chain array of generation in each core algorithm chain, in the pre-processed results array
Each pre-processed results are detected, and the characteristic point is obtained.Wherein, the number that pre-processed results are replicated, with core algorithm
The quantity of core algorithm chain is identical in chain array.
In second of mode in the cards, the first in conjunction with second aspect is possible, mixed reality processor, packet
It includes: dynamic two-level cache circuit, for receiving the optimization algorithm mould in the Preprocessing Algorithm chain and/or the core algorithm chain
Block treated MR pixel data, and form MR picture element caching data.Parallel sub-circuit, for slow by the dynamic two-stage
It deposits the MR picture element caching data obtained in circuit and carries out parallel processing, obtain module arithmetic result.Wherein, the dynamic two
Grade buffer circuit and the parallel sub-circuit are located at, and the optimization algorithm module and programmable gate in pretreatment circuit are electric parallel
In the optimization algorithm module on road.
In the third mode in the cards, second in conjunction with second aspect is possible, the mixed reality processor,
For: according to the resolution ratio of the MR mixed reality pixel stream of binocular depth camera module acquisition, by the Preprocessing Algorithm
Computing circuit positive integer times in chain and core algorithm chain increase, and obtain parallel sub-circuit.Pass through the processing mode pair of assembly line
The continuous received MR mixed reality pixel stream of the buffer circuit carries out data processing.
It, can in conjunction with the implementation and the first to three kind of second aspect of second aspect in four kinds of modes in the cards
Can, the generator is also used to: being compiled to the primal algorithm file in primal algorithm module, is obtained compiling result as yellow
Golden reference model.The dynamic two-level cache circuit is generated wherein according to the type of primal algorithm.To primal algorithm file into
Row is rewritten, and optimization algorithm file is obtained.Judge whether the optimization algorithm file is correct according to the gold reference model, if just
Really, the optimization algorithm file download to programmable gate array is generated into hardware circuit and obtains the optimization algorithm module.
Technical solution of the present invention provides a kind of method that characteristic point can be accelerated to detect, comprising: Preprocessing Algorithm chain is to MR
Mixed reality pixel stream is pre-processed, and pre-processed results are obtained.Core algorithm chain is by way of parallel processing to this result
It is detected, obtains characteristic point.Technical solution of the present invention also provides a kind of MR mixed reality intelligence that characteristic point can be accelerated to detect
Glasses, comprising: binocular depth camera module, for acquiring MR mixed reality pixel stream.Mixed reality processor, for pre-processing
Algorithm chain pre-processes the MR mixed reality pixel stream that binocular depth camera module acquires, and obtains pre-processed results, and
Pre-processed results are detected by way of parallel processing, obtain characteristic point.Primary processor, for handling mixed reality
The characteristic point that device obtains exports MR application image after carrying out the operation of MR pose.
It is an advantage of the invention that by increasing mixed reality processor (Mixed Reality Processing
Unit, MPU), it is handled MR data flow in a manner of parallel processing, can generate each layer in image pyramid simultaneously
Image, while characteristic point detection (improving detection speed), output test result are carried out to these images.Primary processor will mix
The data result that reality processor obtains carries out further data processing and exports.In Preprocessing Algorithm chain and core algorithm chain
Each algoritic module in data are handled by the way of pipeline processes, increase data throughout, reduce number
According to waiting time.
The detection characteristic point time that technical solution obtains through the invention is meter with the detection characteristic point time in the prior art
It calculates and completes all image characteristic points in image pyramid and compare total time, pass through parallel and assembly line two kinds of processing modes knot
Conjunction, which improves, extracts feature spot speed, to solve the problems, such as that image pose calculates, tracks Caton.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to make one simply to introduce, it should be apparent that, the accompanying drawings in the following description is the present invention
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is the flow diagram that the embodiment of the present invention one provides;
Fig. 2 is flow diagram provided by Embodiment 2 of the present invention;
Fig. 3 is the flow diagram of step 201 in Fig. 2;
Fig. 4 be in Fig. 3 in step 301 in color-space conversion module data handling procedure flow diagram;
Fig. 5 is the flow chart of core algorithm chain in step 204 in Fig. 2;
Fig. 5 a is with the pixel example figure for illustrating FAST algorithm;
Fig. 6 is the structural schematic diagram of core algorithm chain array in the embodiment of the present invention two;
Fig. 7 is the structural schematic diagram of the MR mixed reality intelligent glasses of acceleration characteristic point detection provided by the invention;
Fig. 8 is the structural schematic diagram one of mixed reality processor 72 in Fig. 7;
Fig. 9 is the structural schematic diagram two of mixed reality processor 72 in Fig. 7;
Figure 10 is that the structure for the MR mixed reality intelligent glasses that one kind provided in this embodiment can accelerate characteristic point to detect is shown
It is intended to two;
Figure 11 is that the structure for the MR mixed reality intelligent glasses that one kind provided in this embodiment can accelerate characteristic point to detect is shown
It is intended to three.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Fig. 1 is the flow diagram that the embodiment of the present invention one provides, as shown in Figure 1, present embodiments providing, comprising:
Step 101, Preprocessing Algorithm chain pre-process MR mixed reality pixel stream, obtain pre-processed results.
Step 102, core algorithm chain detect the pre-processed results by way of parallel processing, obtain feature
Point.
Core algorithm chain by self-replication N/2 times (N >=2 and be even number), generates core algorithm chain battle array according to representative value N
Column.The number that core algorithm chain array is replicated according to itself, replicates pre-processed results, obtains pre-processed results battle array
Column.
Core algorithm chain array includes multiple cores algorithm chain, during detecting characteristic point, these core algorithm chains
Characteristic point detection is carried out to each pre-processed results in pre-processed results array simultaneously, obtains characteristic point.
It include optimization algorithm module relevant to detection characteristic point algorithm in Preprocessing Algorithm chain and core algorithm chain, often
A optimization algorithm module all has dynamic two-level cache, for receiving by previous algoritic module treated MR pixel data, and
Form MR picture element caching data.Later, parallel processing is carried out to MR picture element caching data, obtains module arithmetic result.Specifically,
The computing circuit positive integer times in optimization algorithm module are increased according to MR mixed reality pixel stream resolution ratio, obtain computing circuit
Array.Each computing circuit in computing circuit array counts the MR picture element caching data by pipeline processing mode
According to processing, module arithmetic result is obtained.It is flowed out specifically, MR picture element caching data are continual from a upper optimization algorithm module
Afterwards, into the dynamic two-level cache of next algoritic module, in the dynamic two-level cache of next optimization algorithm module to its into
Row processing, directly carries out next stage operation after obtaining result, and the image being not to wait in all MR pixel data streams is all handled
After carry out next stage operation again.
Increase level cache and L2 cache in primal algorithm module, extracts the primal algorithm text in primal algorithm module
Part, and compiled using result as gold reference model.After being rewritten to primal algorithm file, according to gold reference model
Result correctness is rewritten in judgement.If result is correct, imput output circuit and respective drive are further increased, it finally will be above-mentioned
As a result it downloads or maps in programmable logic gate circuit, obtain hardware circuit, to obtain optimization algorithm module.
It is an advantage of the invention that by that the middle maximum that occupies of processor (CPU of serial processing mode) will transport in the prior art
The feature point detection algorithm part of calculation amount is transplanted in newly-increased mixed reality processor (MPU), thus with the side of parallel processing
Formula handles MR data flow, by generating the image of each layer in image pyramid simultaneously, while carrying out to these images special
Sign point detection (improving detection speed), output test result.Further, in Preprocessing Algorithm chain and core algorithm chain
Data are handled by the way of pipeline processes in each algoritic module, increase data throughout, reduce data
Waiting time.
The detection characteristic point time that technical solution obtains through the invention is meter with the detection characteristic point time in the prior art
It calculates and completes all image characteristic points in image pyramid and compare total time, pass through parallel and assembly line two kinds of processing modes knot
Conjunction, which improves, extracts feature spot speed, to solve the problems, such as that image pose calculates, tracks Caton.
Current embodiment two specifically describes technical solution of the present invention, as shown in Figure 2:
Step 201, Preprocessing Algorithm chain receive and pre-process MR mixed reality pixel stream.
Algoritic module in Preprocessing Algorithm chain handles MR mixed reality pixel stream, obtains pre-processed results.
Step 202, core algorithm chain replicate N/2 times, generate core algorithm chain array.
Specifically, representative value N is to be obtained according to characteristic point to the required precision of scale invariability.
Step 203, core algorithm chain array carry out characteristic point detection to pre-processed results array, obtain characteristic point result.
Step 204 replicates pre-processed results N/2 times, obtains pre-processed results array.
Specifically, having color-space conversion module and border extended algoritic module in Preprocessing Algorithm chain in step 201.
Pretreatment in existing specific explanations step 201, as shown in Figure 3:
Step 301, color-space conversion module receive MR mixed reality pixel stream.
Color image in MR mixed reality pixel stream is converted to gray level image by step 302.
In characteristic point detection, in order to reduce data volume, usually by image data color image information filtering or
The lesser gray level image information of data volume (effective image information can also be removed in the process) is converted to, in certain journey
Reduce data volume on degree.Such as have, CvtColor, it is color space conversion function, has RGB color being converted to grayscale image
The function of picture.
Step 303, border extended algoritic module are by the border extended of gray level image.
Specifically, image is become larger, is then filled out automatically in a manner of extrapolation after border extended algoritic module expands image border
Fill image boundary.Especially by first expanded images right boundary, new image is formed, is then expanded new image upper following
Boundary.Its purpose is to increase image-region to a certain extent, it is missed to avoid the characteristic point in original image edge.Tool
Body pre-processed results are the gray scale MR mixed reality pixel stream on expanded images boundary.
In the technical solution of the present invention, there is dynamic two-level cache: level cache and two in each optimization algorithm module
Grade caching, after obtaining MR picture element caching data, computing circuit carries out data processing to it, obtains module arithmetic as a result, specific
It is illustrated by taking the color-space conversion module in step 301 as an example, as shown in Figure 4:
Level cache be laterally caching, MR picture element caching data it is continual from a upper algoritic module outflow after uninterruptedly into
Enter level cache.
Specifically as shown, D00、D01、D02、D03、D04、D05……D0MFor level cache the first row data, D10、D11、D12、
D13、D14、D15……D1MFor the second row of level cache data, D20、D21、D22、D23、D24、D25……D2MFor level cache third
Row data.
L2 cache is longitudinal caching, for receiving the data flowed out from level cache.
Further, according to MR mixed reality pixel stream resolution ratio by the computing circuit positive integer times in optimization algorithm module
Increase, obtains computing circuit array.Such as resolution ratio is 720P, computing circuit increases by 720 times.
Each computing circuit in computing circuit array by pipeline processing mode to the MR picture element caching data into
Row data processing obtains module arithmetic result.Specific signal is as shown in computing circuit part in Fig. 4.
Such as figure, there are computing circuit 1 (abbreviation circuit 1), circuit 2 ... circuit N, after data begin to flow into L2 cache, electricity
Road 1 starts to handle it, follow-up data be directly entered without waiting it is processed in circuit 2, and so on until circuit N in
Also there are the data in processed, if circuit quantity is unable to meet demand at this time, data are stored in L2 cache.Circuit 1 is to current
After data processing, result is exported, continues to next pending data.Likewise, circuit 2 has handled current data
Bi Hou exports result, continues to next pending data.
Specifically, in first clock cycle T1Interior, it is slow that the data in level cache in first nine grids enter second level
It deposits, as shown in figure 4, specific data are as follows: D00、D01、D02、D10、D11、D12、D20、D21And D22.These data enter computing circuit 1
Output after processed.In second clock cycle T2When, the data in level cache in second nine grids enter L2 cache,
Specific data are as follows: D01、D02、D03、D11、D12、D13、D21、D22、D23.It is processed that these data enter computing circuit 2.When subsequent
Clock cycle T3、T4、……TN, and so on.
The data result obtained by circuit 1, into the dynamic two-level cache of next optimization algorithm module, next excellent
Change and it is handled in the dynamic two-level cache of algoritic module, directly carries out next stage operation after obtaining result.In module
The interior data processing such as above process constantly recycles.
Now step 204 is described in detail, is illustrated by taking a core algorithm chain as an example first, as shown in Figure 5:
Step 501, Zoom module zoom in and out pre-processed results, obtain zoomed image.
Specifically, Zoom module zooms in and out pre-processed results according to image pyramid.
Step 502, FAST algoritic module carry out characteristic point Preliminary detection to zoomed image, obtain Preliminary detection result.
Specifically, being to carry out coordinate calculating to the pixel in zoomed image, obtain corresponding weight value, it is full-time it is maximum be first
The characteristic point that step detection obtains.
Further, it is specifically described by taking image in Fig. 5 a as an example:
In piece image, non-angle point often occupies the majority, and non-Corner Detection is more much easier than Corner Detection, because
This weeds out non-angle point first will greatly improve Corner Detection speed.Since N (circumference pixel) is 16, so number is 1,5,
At least it should meet corner conditions there are three pixel in 9,13 this 4 circumference pixels, the center of circle is possible to be angle point.Cause
This first checks for 1 and 9 circumference pixels, if I1 and I9 between [Ip-t, Ip+t], then the center of circle certainly not angle point, otherwise
Reexamine 5 and 13 circumference pixels.If at least there are three circumference pixels to meet brightness higher than Ip in this 4 circumference pixels
+ t is lower than Ip-t, then further checks rest of pixels point on circumference.And so on, finally obtain feature as shown in Figure 5 a
Point.
Step 503, HarrisResponse algoritic module carry out secondary detection to Preliminary detection result.
Specifically, HarrisResponse is the variation by the gray scale in all directions of pixel in zoomed image.To examine
Surveying characteristic point part is positioned at smooth plane or on the angle point of male and fomale(M&F).
Step 504 assigns directionality to the characteristic point that secondary detection obtains.
Centroid angle calculating is carried out to secondary detection result specifically, assigning directionality and referring to, specific algorithm is
IcAngels。
Step 505, output characteristic point testing result.
Now the core algorithm chain array and pre-processed results array that refer in technical solution of the present invention are described in detail.
Specifically, the structural schematic diagram of core algorithm chain array is as shown in Figure 6.
Processor replicates above-mentioned core algorithm chain N/2 times according to representative value N, each independent in core algorithm chain array
Core algorithm chain parallel arranged.Pre-processed results have also been replicated N/2 times, during processing, every in core algorithm chain array
A core algorithm chain receives pre-processed results simultaneously, while carrying out data processing to these pre-processed results.
Received corresponding pre-processed results are handled by above-mentioned steps 501- step 505 in core algorithm chain, will be examined
The characteristic point measured exports successively.
It further illustrates, since Zoom module in step 501 can be according to each layer of image pyramid of image scaling ratio
Example zooms in and out pre-processed results.So in the core algorithm chain being each replicated actual treatment MR mixed reality data,
Image in respectively current MR mixed reality pixel stream: frame/1, frame/1.2, frame/1.22, frame/1.23... frame/1.2m.From
And the core algorithm chain in core algorithm chain array is respectively, frame/1 core algorithm chain, frame/1.2 core algorithm chains, frame/
1.22Core algorithm chain, frame/1.23Core algorithm chain ... frame/1.2mCore algorithm chain.
Specifically, frame/1.2mRelationship with 2/N is, it is assumed that N is to have 3 core algorithms in 6 core algorithm chain arrays
Chain, specially frame/1 core algorithm chain, frame/1.2 core algorithm chains, frame/1.22Core algorithm chain.
From the description to technical solution of the present invention it is found that having multiple optimizations in core algorithm chain array and Preprocessing Algorithm chain
Algoritic module, description of the internal structure and process flow of each optimization algorithm module according to Fig. 4.
The specific structure of level cache and L2 cache in algoritic module is specifically described for current FAST algoritic module, and
The specific structure of level cache and L2 cache in other optimization algorithm modules is not caused to limit, each optimization algorithm module
In level cache and the specific structure of L2 cache be to be determined according to its algorithm own characteristic:
The core of Fast algorithm is to calculate central point distance to be whether the difference in the region of m is greater than threshold value, and row caching is adopted
With n row (level cache), windows cache (L2 cache) uses m*m, while FAST is often required to carry out maximum value suppression to characteristic point
Operation processed, therefore increase by one group of caching, row caching (level cache) uses 2 rows, and windows cache is cached using m*m as maximum value
The input data of operation.Such as: central point distance is calculated as the difference in 7 region and whether is greater than threshold value, because this journey caching is adopted
With 15 rows, windows cache uses 7*7, while FAST is often required to carry out characteristic point maximum value inhibition operation, therefore increases by one group
Caching, row caching use 2 rows, input data of the windows cache using 3*3 as maximum value caching operation.
Now the acquisition of the optimization algorithm module referred in technical solution of the present invention is illustrated to existing.
Primal algorithm file is extracted from primal algorithm module, runs this algorithm file using VS (Visual Studio)
And pixel stream is inputted, if exporting correct processing result, using this result as gold reference model.According to HLS grammar correction
(modification process is roughly the same, but the type of each algorithm is different, so firsts and seconds in dynamic two-level cache for this algorithm file
The structure of caching is different, specifically describes such as above-mentioned FAST algoritic module), modified primal algorithm file is run, IP kernel is generated.
Later, increase binocular depth camera module driving circuit (input), IP kernel, HDMI display circuit (output), it is comprehensive to generate netlist
File download forms hardware circuit and obtains optimization algorithm module to FPGA.Reaching control operation to control terminal control routine should
The purpose of hardware circuit.
Specifically, above-mentioned modification process includes: that floating-point operation is revised as fixed-point calculation;Specified computing circuit handles image
Full-size;Modify all random access.
Specifically, being only illustrated by taking vivado platform as an example above, platform itself is not limited, be can be realized
Any platform of above-mentioned function all should belong to protection scope of the present invention.
Technical solution of the present invention provides the method for accelerating characteristic point detection, comprising: Preprocessing Algorithm chain mixes MR existing
Real image element stream is pre-processed, and pre-processed results are obtained.Core algorithm chain examines this result by way of parallel processing
It surveys, obtains characteristic point.
It is an advantage of the invention that it is in a manner of parallel processing to MR number by increasing mixed reality processor (MPU)
It is handled according to stream, the image of each layer in image pyramid can be generated simultaneously, while characteristic point detection is carried out to these images
(improving detection speed), output test result.The data result that primary processor obtains mixed reality processor is carried out into one
Step data is handled and is exported.Using pipeline processes in each algoritic module in Preprocessing Algorithm chain and core algorithm chain
Mode handles data, increases data throughout, reduces the waiting time of data.
The detection characteristic point time that technical solution obtains through the invention is meter with the detection characteristic point time in the prior art
It calculates and completes all image characteristic points in image pyramid and compare total time, pass through parallel and assembly line two kinds of processing modes knot
Conjunction, which improves, extracts feature spot speed, to solve the problems, such as that image pose calculates, tracks Caton.
The present invention also provides the MR mixed reality intelligent glasses for accelerating characteristic point detection, Fig. 7 is provided in this embodiment
Accelerate the structural schematic diagram of the MR mixed reality intelligent glasses of characteristic point detection, as shown in Figure 7:
Binocular depth camera module 71, for acquiring MR mixed reality pixel stream.
Mixed reality processor 72 (MPU) acquires the binocular depth camera module 71 for Preprocessing Algorithm chain
The MR mixed reality pixel stream is pre-processed, and pre-processed results are obtained, and is also used to core algorithm chain and is passed through parallel processing
Mode detects the pre-processed results, obtains characteristic point.
Primary processor 73, after the characteristic point for obtaining the mixed reality processor 72 carries out the operation of MR pose
Export MR application image.
As shown in figure 8, mixed reality processor 72 includes:
Circuit 81 is pre-processed, for imaging mould to the binocular depth by the optimization algorithm module in Preprocessing Algorithm chain
The MR mixed reality pixel stream of 71 acquisition of group is pre-processed, and the pre-processed results are obtained.
Generator 82 is also used to pair for according to representative value N, generating core algorithm chain array to the core algorithm chain
The pre-processed results that pretreatment circuit 81 obtains are replicated, and pre-processed results array is obtained.
Programmable gate parallel circuit 83, for passing through each core in the core algorithm chain array that generator 82 generates
Optimization algorithm module in algorithm chain detects each pre-processed results in the pre-processed results array, obtains institute
State characteristic point.
Wherein, the quantity phase of core algorithm chain in the number that the pre-processed results are replicated, with core algorithm chain array
Together.
The generator 81, is specifically used for: according to according to representative value N, the core chain, which is replicated N/2 times, (N >=2 and is
Even number), generate core chain array;The pre-processed results are replicated, pre-processed results array is generated, is also used to according to institute
It states core chain array and generates parallel circuit, and be cured in programmable gate chip.
As shown in figure 9, mixed reality processor 72, comprising:
Dynamic two-level cache circuit 91, it is excellent in the Preprocessing Algorithm chain and/or the core algorithm chain for receiving
Change algoritic module treated MR pixel data, and forms MR picture element caching data.
Parallel sub-circuit 92, for passing through the MR picture element caching data obtained in the dynamic two-level cache circuit
Parallel processing is carried out, module arithmetic result is obtained.Wherein, the dynamic two-level cache circuit and the parallel sub-circuit are located at,
In the optimization algorithm module for pre-processing the optimization algorithm module and programmable gate parallel circuit 83 in circuit 81.
Mixed reality processor 72, is used for: according to the MR mixed reality pixel stream of the acquisition of binocular depth camera module 71
Computing circuit positive integer times in the Preprocessing Algorithm chain and core algorithm chain are increased, obtain parallel sub-circuit by resolution ratio.
Pipeline data processing is carried out to the continuous received MR mixed reality pixel stream of the dynamic two-level cache circuit 91.
Specifically, after the continual outflow from a upper optimization algorithm module of MR mixed reality pixel stream, into next optimization
In the dynamic two-level cache of algoritic module, it is handled in the dynamic two-level cache of next optimization algorithm module, is obtained
As a result next stage operation is directly carried out after.Without the image in all MR pixel data streams after the current generation is all disposed
Next stage operation is carried out again.
Generator 81 is also used to: to generating institute in the primal algorithm module in Preprocessing Algorithm chain and the core algorithm chain
State dynamic two-level cache circuit;Primal algorithm file in the primal algorithm module is compiled, compiling result is obtained;Root
Judge whether the compiling result is correct according to gold reference model;If correct, the compiling result is downloaded to programmable logic
Gate array column-generation hardware circuit obtains the optimization algorithm module.
Figure 10 is the structural schematic diagram two of the MR mixed reality intelligent glasses of acceleration characteristic point detection provided in this embodiment,
As shown in Figure 10: binocular depth camera module acquires MR mixed reality pixel stream.Pre-process the optimization algorithm module pair in circuit
It is pre-processed, and pre-processed results are obtained, for reduce data volume and enlarged image edge is intended to reduce color data and
It avoids omitting the missing feature point in subsequent calculating.Generator obtains typical case to the required precision of scale invariability according to characteristic point
Value N, and then generate core algorithm chain array and pre-processed results array.And parallel circuit is generated according to core algorithm chain array,
And be cured in programmable gate chip, further obtain programmable gate parallel circuit.Programmable gate is electric parallel
The optimization algorithm module in each core algorithm chain in road, examines each pre-processed results in pre-processed results array
It surveys, obtains characteristic point.Primary processor exports MR application image after carrying out the operation of MR pose to obtained characteristic point.Wherein, pre- place
Each optimization algorithm module in reason circuit and Programmadle logic door parallel circuit has and can increase the dynamic of data throughout
State two-level cache circuit and the parallel sub-circuit that data processing speed can be increased.
Now further the MR mixed reality intelligent glasses that technical solution of the present invention provides are further illustrated, such as Figure 11 institute
Show:
Binocular depth camera module acquires MR and mixes existing pixel stream, and mixed reality processor (MPU) carries out at data it
Reason obtains that data are passed through bus transfer to primary processor (usually CPU) after characteristic point.It is acquired in binocular depth camera module
While MR mixes existing pixel stream, sensor module acquisition environmental information is simultaneously transmitted to primary processor, in sensor module at least
Include: light flies sensor, position and attitude sensor, Temperature Humidity Sensor, air detecting sensors.Control mould group is for receiving
External control instruction is simultaneously transferred to primary processor, can be touching instruction (touch panel) or sound instruction (microphone).DDR
(Double Data Rate Double Data Rate synchronous DRAM) and FLASH (flash memory) are for assisting primary processor work
Make.
After primary processor handles the information such as received characteristic point, environmental information, control instruction, output is aobvious to nearly eye
Show mould group and shows.(mobile industry processor interface, Mobile Industry can be specifically transmitted by MIPI interface
Processor Interface)。
Nearly eye display module (NED, Near by Display) is common by the nearly eye display end of left mesh and the nearly eye display end of right mesh
Composition.
Power supply is used to power to above each hardware mould group.
Wherein, there is digital light display medium in nearly eye display module, including but not limited to: LCOS (Liquid Crystal
On Silicon, liquid crystal on silicon), LCD (Liquid Crystal Display, liquid crystal display), post light waveguide raster
Eyeglass, prism group is semi-transparent/full impregnated shows optical module, and freeform optics prism is semi-transparent/full impregnated display component, waveguide optical half
Thoroughly/full impregnated shows Lens assembly, DMD (Digital Micromirror Device, digital micro-mirror device) etc..
Technical solution of the present invention provides a kind of MR mixed reality intelligent glasses that characteristic point can be accelerated to detect, comprising: binocular
Depth camera mould group, for acquiring MR mixed reality pixel stream.Mixed reality processor, for Preprocessing Algorithm chain to binocular depth
The MR mixed reality pixel stream of degree camera module acquisition is pre-processed, and obtains pre-processed results, and pass through parallel processing
Mode detects pre-processed results, obtains characteristic point.Primary processor, the characteristic point for obtaining mixed reality processor
MR application image is exported after carrying out the operation of MR pose.
It is an advantage of the invention that it is in a manner of parallel processing to MR data flow by increasing mixed reality processor
It is handled, the image of each layer in image pyramid can be generated simultaneously, while characteristic point detection is carried out to these images and (is promoted
Detection speed), output test result.The data result that primary processor obtains mixed reality processor carries out further data
It handles and exports.Further, using pipeline processes in the algoritic module in Preprocessing Algorithm chain and core algorithm chain
Mode handles data, increases data throughout, reduces the waiting time of data.
The detection characteristic point time that technical solution obtains through the invention is meter with the detection characteristic point time in the prior art
It calculates and completes all image characteristic points in image pyramid and compare total time, pass through parallel and assembly line two kinds of processing modes knot
Conjunction, which improves, extracts feature spot speed, to solve the problems, such as that image pose calculates, tracks Caton.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement;And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (10)
1. the method that one kind can accelerate characteristic point to detect, which is characterized in that the described method includes:
Preprocessing Algorithm chain pre-processes MR mixed reality pixel stream, obtains pre-processed results;
Core algorithm chain detects the pre-processed results by way of parallel processing, obtains characteristic point.
2. the method according to claim 1, wherein the core algorithm chain is by way of parallel processing to institute
It states pre-processed results to be detected, obtains characteristic point, comprising:
The core algorithm chain generates core algorithm chain array according to representative value N;
The core algorithm chain array replicates the pre-processed results, obtains pre-processed results array;
Each core algorithm chain in the core algorithm chain array, respectively to each pretreatment in the pre-processed results array
As a result it is detected, obtains the characteristic point;
Wherein, the core algorithm chain is according to representative value N, by self-replication N/2 times (N >=2 and be even number);The pretreatment knot
The number that fruit is replicated is identical as the quantity of core algorithm chain in core algorithm chain array.
3. the method according to claim 1, wherein the method, comprising:
Each optimization algorithm module in the Preprocessing Algorithm chain and the core algorithm chain has dynamic two-level cache, uses
In receiving by previous optimization algorithm module treated MR pixel data, and form MR picture element caching data;
Parallel processing is carried out to the MR picture element caching data in the optimization algorithm module, obtains module arithmetic result;
The dynamic two-level cache includes level cache and L2 cache.
4. according to the method described in claim 3, it is characterized in that, it is described in the optimization algorithm module to the MR pixel
Data cached carry out parallel processing, obtains module arithmetic result, comprising:
The computing circuit positive integer times in the optimization algorithm module are increased according to the MR mixed reality pixel stream resolution ratio,
Obtain computing circuit array;
Each computing circuit in the computing circuit array by pipeline processing mode to the MR picture element caching data into
Row data processing obtains the module arithmetic result.
5. method described in -4 according to claim 1, which is characterized in that the method also includes:
Primal algorithm file in primal algorithm module is compiled, obtains compiling result as gold reference model;
According to the type of the primal algorithm, increase the dynamic two-level cache wherein;
The primal algorithm file is rewritten, optimization algorithm file is obtained;
Judge whether the optimization algorithm file is correct according to the gold reference model;
If correct, the optimization algorithm file download to programmable gate array is generated into hardware circuit, is obtained described excellent
Change algoritic module.
6. the MR mixed reality intelligent glasses that one kind can accelerate characteristic point to detect, which is characterized in that the MR mixed reality intelligence
Glasses include:
Binocular depth camera module, for acquiring MR mixed reality pixel stream;
Mixed reality processor, the MR mixed reality that the binocular depth camera module is acquired for Preprocessing Algorithm chain
Pixel stream is pre-processed, and pre-processed results are obtained, and is also used to core algorithm chain by way of parallel processing to the pre- place
Reason result is detected, and characteristic point is obtained;
Primary processor, output MR is answered after the characteristic point for obtaining the mixed reality processor carries out the operation of MR pose
Use image.
7. MR mixed reality intelligent glasses according to claim 6, which is characterized in that the mixed reality processor packet
It includes:
Circuit is pre-processed, for acquiring by the optimization algorithm module in Preprocessing Algorithm chain to the binocular depth camera module
The MR mixed reality pixel stream pre-processed, obtain the pre-processed results;
Generator, for the core algorithm chain according to representative value N, to be generated core algorithm chain array, is also used to described pre-
The pre-processed results that processing circuit obtains are replicated, and pre-processed results array is obtained;According to representative value N, by the core
Center algorithm chain replicates N/2 time (N >=2 and be even number), and generation core algorithm chain array replicates the pre-processed results,
Generate pre-processed results array;It is also used to generate parallel circuit according to the core algorithm chain array, and is cured to programmable patrol
It collects in door chip;
Programmable gate parallel circuit, each core algorithm in the core algorithm chain array for being generated by the generator
Optimization algorithm module in chain detects each pre-processed results in the pre-processed results array, obtains the spy
Sign point;
Wherein, the number that the pre-processed results are replicated is identical as the quantity of core algorithm chain in core algorithm chain array.
8. MR mixed reality intelligent glasses according to claim 7, which is characterized in that mixed reality processor, comprising:
Dynamic two-level cache circuit, for receiving the optimization algorithm in the Preprocessing Algorithm chain and/or the core algorithm chain
MR pixel data after resume module, and form MR picture element caching data;
Parallel sub-circuit, it is parallel for being carried out to the MR picture element caching data obtained by the dynamic two-level cache circuit
Processing, obtains module arithmetic result;
Wherein, the dynamic two-level cache circuit and the parallel sub-circuit are located at, and pre-process the optimization algorithm module in circuit
In the optimization algorithm module of programmable gate parallel circuit.
9. MR mixed reality intelligent glasses according to claim 8, the mixed reality processor, are used for:
According to the resolution ratio of the MR mixed reality pixel stream of binocular depth camera module acquisition, by the Preprocessing Algorithm chain
Increase with the computing circuit positive integer times in core algorithm chain, obtains parallel sub-circuit;
The received MR mixed reality pixel stream continuous to the dynamic two-level cache circuit carries out pipeline data processing.
10. the generator is also used to: to primal algorithm mould according to MR mixed reality intelligent glasses described in claim 6-9
Primal algorithm file in block is compiled, and obtains compiling result as gold reference model;According to the class of the primal algorithm
Type generates the dynamic two-level cache circuit wherein;The primal algorithm file is rewritten, optimization algorithm file is obtained;
Judge whether the optimization algorithm file is correct according to the gold reference model;It, will be under the optimization algorithm file if correct
It is loaded onto programmable gate array generation hardware circuit and obtains the optimization algorithm module.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810204060.3A CN110276836A (en) | 2018-03-13 | 2018-03-13 | A kind of method and MR mixed reality intelligent glasses accelerating characteristic point detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810204060.3A CN110276836A (en) | 2018-03-13 | 2018-03-13 | A kind of method and MR mixed reality intelligent glasses accelerating characteristic point detection |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110276836A true CN110276836A (en) | 2019-09-24 |
Family
ID=67957529
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810204060.3A Pending CN110276836A (en) | 2018-03-13 | 2018-03-13 | A kind of method and MR mixed reality intelligent glasses accelerating characteristic point detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110276836A (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1979527A (en) * | 2005-12-09 | 2007-06-13 | 中国科学院沈阳自动化研究所 | Image corner rapid extraction method and implementation device |
CN103927745A (en) * | 2014-03-28 | 2014-07-16 | 北京中海新图科技有限公司 | Tracking and matching parallel computing method for wearable device |
US20170024632A1 (en) * | 2015-07-23 | 2017-01-26 | Mireplica Technology, Llc | Performance Enhancement For Two-Dimensional Array Processor |
CN106846496A (en) * | 2017-01-19 | 2017-06-13 | 杭州古珀医疗科技有限公司 | DICOM images based on mixed reality technology check system and operating method |
CN106960451A (en) * | 2017-03-13 | 2017-07-18 | 西安电子科技大学 | A kind of method for lifting the weak texture region characteristic point quantity of image |
CN106997617A (en) * | 2017-03-10 | 2017-08-01 | 深圳市云宙多媒体技术有限公司 | The virtual rendering method of mixed reality and device |
CN107680028A (en) * | 2016-08-01 | 2018-02-09 | 北京百度网讯科技有限公司 | Processor and method for zoomed image |
-
2018
- 2018-03-13 CN CN201810204060.3A patent/CN110276836A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1979527A (en) * | 2005-12-09 | 2007-06-13 | 中国科学院沈阳自动化研究所 | Image corner rapid extraction method and implementation device |
CN103927745A (en) * | 2014-03-28 | 2014-07-16 | 北京中海新图科技有限公司 | Tracking and matching parallel computing method for wearable device |
US20170024632A1 (en) * | 2015-07-23 | 2017-01-26 | Mireplica Technology, Llc | Performance Enhancement For Two-Dimensional Array Processor |
CN107680028A (en) * | 2016-08-01 | 2018-02-09 | 北京百度网讯科技有限公司 | Processor and method for zoomed image |
CN106846496A (en) * | 2017-01-19 | 2017-06-13 | 杭州古珀医疗科技有限公司 | DICOM images based on mixed reality technology check system and operating method |
CN106997617A (en) * | 2017-03-10 | 2017-08-01 | 深圳市云宙多媒体技术有限公司 | The virtual rendering method of mixed reality and device |
CN106960451A (en) * | 2017-03-13 | 2017-07-18 | 西安电子科技大学 | A kind of method for lifting the weak texture region characteristic point quantity of image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP3506161A1 (en) | Method and apparatus for recovering point cloud data | |
CN113963044B (en) | Cargo box intelligent loading method and system based on RGBD camera | |
CN110956126A (en) | Small target detection method combined with super-resolution reconstruction | |
CN110059793B (en) | Gradual modification of a generative antagonistic neural network | |
CN111368668B (en) | Three-dimensional hand recognition method and device, electronic equipment and storage medium | |
CN110689000A (en) | Vehicle license plate identification method based on vehicle license plate sample in complex environment | |
CN111046746A (en) | License plate detection method and device | |
CN111611925A (en) | Building detection and identification method and device | |
CN109712092A (en) | Archives scan image repair method, device and electronic equipment | |
CN113762461A (en) | Training neural networks with finite data using reversible enhancement operators | |
CN112464954A (en) | Lightweight target detection network applied to embedded equipment and training method | |
CN115049738A (en) | Method and system for estimating distance between person and camera | |
CN101694385B (en) | Small target detection instrument based on Fourier optics and detection method thereof | |
CN109708662B (en) | High-frame-frequency high-precision injection type star atlas simulation test platform based on target identification | |
CN108520532B (en) | Method and device for identifying motion direction of object in video | |
CN110276836A (en) | A kind of method and MR mixed reality intelligent glasses accelerating characteristic point detection | |
CN116310102A (en) | Three-dimensional reconstruction method, terminal and medium of transparent object image based on deep learning | |
CN106815861A (en) | A kind of optical flow computation method and apparatus of compact | |
CN111508024A (en) | Method for estimating pose of robot based on deep learning | |
CN116246062A (en) | Performing semantic segmentation training using image/text pairs | |
LU500193B1 (en) | Low-illumination image enhancement method and system based on multi-expression fusion | |
CN113256711B (en) | Pose estimation method and system of monocular camera | |
CN113011223A (en) | Image recognition method, system, equipment and storage medium | |
Liang et al. | Experimental Evaluation of Direct Monocular Visual Odometry Based on Nonlinear Optimization | |
CN111738920B (en) | FPGA chip oriented to panoramic stitching acceleration and panoramic image stitching method |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190924 |
|
WD01 | Invention patent application deemed withdrawn after publication |