CN105630906A - Person searching method, apparatus and system - Google Patents
Person searching method, apparatus and system Download PDFInfo
- Publication number
- CN105630906A CN105630906A CN201510964095.3A CN201510964095A CN105630906A CN 105630906 A CN105630906 A CN 105630906A CN 201510964095 A CN201510964095 A CN 201510964095A CN 105630906 A CN105630906 A CN 105630906A
- Authority
- CN
- China
- Prior art keywords
- personnel
- image
- body region
- feature
- region
- 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 75
- 238000012545 processing Methods 0.000 claims abstract description 41
- 230000004927 fusion Effects 0.000 claims abstract description 26
- 230000011218 segmentation Effects 0.000 claims abstract description 21
- 210000000746 body region Anatomy 0.000 claims description 137
- 238000000605 extraction Methods 0.000 claims description 37
- 238000011524 similarity measure Methods 0.000 claims description 31
- 239000000284 extract Substances 0.000 claims description 23
- 230000008569 process Effects 0.000 claims description 19
- 238000012163 sequencing technique Methods 0.000 claims description 9
- 239000004744 fabric Substances 0.000 claims description 2
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000012544 monitoring process Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000005286 illumination Methods 0.000 description 4
- 235000008733 Citrus aurantifolia Nutrition 0.000 description 3
- 235000011941 Tilia x europaea Nutrition 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000004571 lime Substances 0.000 description 3
- 238000000638 solvent extraction Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- AXDJCCTWPBKUKL-UHFFFAOYSA-N 4-[(4-aminophenyl)-(4-imino-3-methylcyclohexa-2,5-dien-1-ylidene)methyl]aniline;hydron;chloride Chemical compound Cl.C1=CC(=N)C(C)=CC1=C(C=1C=CC(N)=CC=1)C1=CC=C(N)C=C1 AXDJCCTWPBKUKL-UHFFFAOYSA-N 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000009415 formwork Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000010981 turquoise Substances 0.000 description 1
- 210000003462 vein Anatomy 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
- G06Q50/265—Personal security, identity or safety
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention discloses a person searching method, apparatus and system. The method comprises: receiving an image that contains a to-be-searched person; performing segmentation processing on the image, so as to obtain a person whole-body region, a person upper-body region and/or a person lower-body region; separately extracting various types of feature information of images in the person whole-body region., the person upper-body region and/or the person lower-body region; separately calculating a similarity degree between each type of feature information of each region and corresponding feature information in a database; performing a weighted operation according to a preset weight value of the similarity degree, so as to obtain a fusion similarity degree of images in the person whole-body region, the person upper-body region and/or the lower-body region that correspond to each person image in the database; and separately obtaining searching results of person images, which is sorted according to the fusion similarity degree, in the database. The method, apparatus and system disclosed by the present invention have the advantages of an accurate retrieval result and high retrieval efficiency.
Description
Technical field
The present invention relates to picture searching and field of video monitoring, be specifically related to a kind of people search's method, Apparatus and system.
Background technology
Along with extensively building of safe city faces the universal of monitoring with various places, video monitoring data quantitative change obtains increasing, this brings huge challenge to criminal investigation and case detection, and how to extract target suspect from these high-volume databases rapidly and accurately becomes the key solved a case.
The video investigation pattern that Traditional Man browses needs to expend substantial amounts of manpower and time, it is easy to affect the opportunity of solving a case adversely. People search's technology is easy to video investigation person and is found suspected target moving frame and track quickly and accurately, and public security department is improved case-solving rate, safeguards that life property safety of people is significant. Collection to be measured is ranked up by the distance that current pedestrian retrieval method is all according to inquiry pedestrian's object and all pedestrian's its appearance features to be measured. But under actual video monitoring environment, the factors such as the different visual angle of photographic head, illumination, aberration are different, cause that same a group traveling together macroscopic features under multi-cam often exists significant difference, a kind of method effectively is not had can relatively accurately to extract the feature of image, when carrying out multiple features fusion cannot accurate weights assigned coefficient so that retrieval result is inaccurate.
Summary of the invention
Therefore, the technical problem to be solved in the present invention is in that to overcome the inaccurate defect of retrieval result of people search's method of the prior art.
For this, a kind of people search's method of the present invention, comprise the following steps:
Receive the image comprising personnel to be searched;
Described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively;
Calculating the similarity between individual features information in each type of characteristic information in each region and data base respectively, described data base is for storing polytype characteristic information of image in the polytype characteristic information and personnel's lower part of the body region including image in polytype characteristic information of image in the personnel whole body region that all personnel's image is corresponding, personnel region above the waist;
Weighted value shared by default described similarity computes weighted, and obtains the fusion similarity of image in described personnel whole body region corresponding to each personnel's image in described data base, personnel region above the waist and/or personnel's lower part of the body region respectively;
Obtain the Search Results of personnel's image in the described data base according to described fusion sequencing of similarity respectively.
Preferably, described described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist include:
Judge whether whether described personnel to be searched proportion in the picture incomplete in the picture less than threshold value and described personnel to be searched respectively;
When described personnel to be searched proportion in the picture is not full-time in the picture less than threshold value or described personnel to be searched, adopt GRABCUT algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
When described personnel to be searched proportion in the picture is full-time in the picture more than or equal to threshold value or described personnel to be searched, adopt DDN algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist.
Preferably, described polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature.
Preferably, described extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively and include:
When extracting multidimensional Dominant Color Features, respectively the color of image in each region is carried out rough segmentation and essence point, obtain rough segmentation color histogram and essence point color histogram respectively;
When extracting multichannel multiple color spaces assemblage characteristic, respectively image in each region is carried out piecemeal, add up the color histogram of every piece of each color space of subgraph respectively, each color histogram is carried out comprehensively acquisition Color rectangular histogram;
When extracting HOG feature, respectively image in each region is carried out piecemeal, add up gradient orientation histogram and the color histogram of every piece of subgraph respectively, undertaken described gradient orientation histogram and color histogram comprehensively obtaining colored gradient orientation histogram.
Preferably, in the described each type of characteristic information calculating each region respectively and data base, similarity between individual features information includes:
Tanimoto distance metric method is adopted to carry out Similarity Measure for CEDD feature, CSD feature, multichannel multiple color spaces assemblage characteristic, FCTH feature;
Euclidean distance measure is adopted to carry out Similarity Measure for CLD feature, CSURF feature;
1 normal form distance metric method is adopted to carry out Similarity Measure for SCD feature;
The Point matching method of minimizing in neighborhood is adopted to carry out Similarity Measure for HOG feature;
COS distance similarity calculating method is adopted to carry out Similarity Measure for multidimensional Dominant Color Features.
Preferably, described, described image is carried out dividing processing, it is thus achieved that before the step in personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist, also include:
Described image is carried out white balance process, it is thus achieved that the image after white balance process.
A kind of people search's device of the present invention, including:
Receive unit, for receiving the image comprising personnel to be searched;
Cutting unit, for carrying out dividing processing to described image, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Feature extraction unit, for extracting polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively;
First computing unit, for calculating in each type of characteristic information in each region and data base the similarity between individual features information respectively, described data base is for storing polytype characteristic information of image in the polytype characteristic information and personnel's lower part of the body region including image in polytype characteristic information of image in the personnel whole body region that all personnel's image is corresponding, personnel region above the waist;
Second computing unit, compute weighted for weighted value shared by the described similarity preset, obtain the fusion similarity of image in described personnel whole body region corresponding to each personnel's image in described data base, personnel region above the waist and/or personnel's lower part of the body region respectively;
Search Results obtains unit, for obtaining the Search Results of personnel's image in the described data base according to described fusion sequencing of similarity respectively.
Preferably, described cutting unit includes:
Judging unit, for judging whether whether described personnel to be searched proportion in the picture incomplete in the picture less than threshold value and described personnel to be searched respectively;
First segmentation result obtains unit, for when described personnel to be searched proportion in the picture not full-time in the picture less than threshold value or described personnel to be searched, adopt GRABCUT algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Second segmentation result obtains unit, for when described personnel to be searched proportion in the picture full-time in the picture more than or equal to threshold value or described personnel to be searched, adopt DDN algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist.
Preferably, described polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature.
Preferably, described feature extraction unit includes:
Multidimensional Dominant Color Features extraction unit, for when extracting multidimensional Dominant Color Features, the color of image in each region carrying out rough segmentation and essence point respectively, obtains rough segmentation color histogram and essence point color histogram respectively;
Multichannel multiple color spaces assemblage characteristic extraction unit, for when extracting multichannel multiple color spaces assemblage characteristic, respectively image in each region is carried out piecemeal, add up the color histogram of every piece of each color space of subgraph respectively, each color histogram is carried out comprehensively acquisition Color rectangular histogram;
HOG feature extraction unit, for when extracting HOG feature, respectively image in each region is carried out piecemeal, add up gradient orientation histogram and the color histogram of every piece of subgraph respectively, undertaken described gradient orientation histogram and color histogram comprehensively obtaining colored gradient orientation histogram.
Preferably, described first computing unit includes:
First computation subunit, for adopting Tanimoto distance metric method to carry out Similarity Measure for CEDD feature, CSD feature, multichannel multiple color spaces assemblage characteristic, FCTH feature;
Second computation subunit, for adopting Euclidean distance measure to carry out Similarity Measure for CLD feature, CSURF feature;
3rd computation subunit, for adopting 1 normal form distance metric method to carry out Similarity Measure for SCD feature;
4th computation subunit, for adopting the Point matching method of minimizing in neighborhood to carry out Similarity Measure for HOG feature;
5th computation subunit, for adopting COS distance similarity calculating method to carry out Similarity Measure for multidimensional Dominant Color Features.
Preferably, before described cutting unit, also include:
White balance processing unit, for carrying out white balance process to described image, it is thus achieved that the image after white balance process.
A kind of people search's system of the present invention, is placed on camera and the processor of multiple some position including cloth;
Described camera, comprises the image of personnel for Real-time Collection; By Real-time Collection to image be sent to described processor;
Described processor, including real-time processing device and people search's device;
Described real-time processing device, for receive described camera send Real-time Collection to image; Described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and personnel's lower part of the body region above the waist; Extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and personnel's lower part of the body region respectively; Polytype characteristic information of image in described personnel whole body region, personnel region above the waist and personnel's lower part of the body region is stored in data base;
Described people search's device, for receiving the image comprising personnel to be searched; Described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist; Extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively; Calculate the similarity between individual features information in each type of characteristic information in each region and data base respectively; Weighted value shared by default described similarity computes weighted, and obtains the fusion similarity of image in described personnel whole body region corresponding to each personnel's image in described data base, personnel region above the waist and/or personnel's lower part of the body region respectively; Obtain the Search Results of personnel's image in the described data base according to described fusion sequencing of similarity respectively.
Technical solution of the present invention, has the advantage that
1. the embodiment of the present invention provide people search's method, Apparatus and system, by image is carried out dividing processing, effectively it is partitioned into the whole body of personnel targets, upper and lower half body region, such that it is able to carry out feature extraction according to image in region, making to extract targeted region more accurately, also reducing regional extent, thus reducing the noise impact on feature extraction, improve the accuracy and speed of feature extraction, and then the precision of retrieval result can be improved and improve recall precision. By extracting polytype characteristic information of image, and the sequence of result is scanned for according to fusion similarity, more ensure that the result searched all matches with personnel to be searched in polytype feature, further increase the accuracy of retrieval result.
2. people search's method that the embodiment of the present invention provides, Apparatus and system, by judging whether whether personnel to be searched proportion in the picture incomplete in the picture less than threshold value and personnel to be searched respectively, prejudge the testing result of image, different partitioning algorithms is selected according to testing result, namely when personnel to be searched proportion in the picture is not full-time in the picture less than threshold value or personnel to be searched, adopt GRABCUT algorithm, other situations then adopt DDN algorithm, combine GRABCUT algorithm and the feature of DDN algorithm both algorithms, thus for different image anticipation results, more targetedly, better, obtain personnel whole body more accurately, on, lower part of the body region, more accurate basis is provided for follow-up feature extraction, and then the precision of feature extraction and the accuracy of retrieval result can be improved further.
3. the embodiment of the present invention provide people search's method, Apparatus and system, by arranging multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and the polytype characteristic information of FCTH feature, it is adapted to different scene, there is different sign abilities, the robustness of boosting algorithm.
4. the embodiment of the present invention provide people search's method, Apparatus and system, by the image received first being carried out AWB process, can eliminate due to image difference when multi-point is monitored, remove the impact of illumination etc., equilibrium figures image brightness, impact when reducing interference to feature extraction, improves the precision of feature extraction and the accuracy of retrieval result.
Accompanying drawing explanation
In order to be illustrated more clearly that the technical scheme in the specific embodiment of the invention, below the accompanying drawing used required during detailed description of the invention is described is briefly described, apparently, accompanying drawing in the following describes is some embodiments of the present invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is the flow chart of a concrete example of people search's method in the embodiment of the present invention 1;
Fig. 2 is the theory diagram of a concrete example of people search's device in the embodiment of the present invention 2;
Fig. 3 is the theory diagram of a concrete example of people search's system in the embodiment of the present invention 3.
Detailed description of the invention
Below in conjunction with accompanying drawing, technical scheme is clearly and completely described, it is clear that described embodiment is a part of embodiment of the present invention, rather than whole embodiments. Based on the embodiment in the present invention, the every other embodiment that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of protection of the invention.
In describing the invention, it is necessary to explanation, term " first ", " second ", " the 3rd " etc. only for descriptive purposes, and it is not intended that instruction or hint relative importance.
As long as just can be combined with each other additionally, technical characteristic involved in invention described below difference embodiment does not constitute conflict each other.
Embodiment 1
The present embodiment provides a kind of people search's method, as it is shown in figure 1, comprise the steps:
S1, receive and comprise the image of personnel to be searched; This image can derive from network, mobile phone, historical query image etc.
S2, image is carried out dividing processing, it is thus achieved that personnel whole body region, it is also possible to obtain personnel region or personnel's lower part of the body region above the waist, or personnel whole body region, personnel region and personnel's lower part of the body region above the waist all obtain.
Polytype characteristic information of image in S3, respectively extraction personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region. Preferably, if the one or more regions obtained in step s 2 in personnel whole body region, personnel region above the waist and personnel's lower part of the body region, then correspondence extracts polytype characteristic information of image in this region. The region that there is no in step s 2 is not then extracted, and also cannot extract, and hereinafter relevant with region step is explained with reference to this.
Similarity between individual features information in S4, each type of characteristic information calculating each region respectively and data base, data base is for storing polytype characteristic information of image in the polytype characteristic information and personnel's lower part of the body region including image in polytype characteristic information of image in the personnel whole body region that all personnel's image is corresponding, personnel region above the waist. Such as, a type of characteristic information in personnel whole body region calculates similarity with the characteristic information of this kind of type in personnel whole body region in data base. Preferably, data base can be divided into different word banks according to the time period, for instance, can select when doing Similarity Measure is the individual features information in word bank so that hunting zone is substantially reduced, and more targetedly, improves search efficiency and the accuracy of retrieval result.
S5, weighted value shared by the similarity preset compute weighted, and obtain the fusion similarity of image in personnel whole body region, personnel region above the waist and/or the personnel's lower part of the body region that in data base, each personnel's image is corresponding respectively. Preferably, using the inverse of the standard deviation of all similarities of each type of characteristic information as weighted value shared by default similarity. In personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region that each personnel's image is corresponding, the fusion similarity of image is that after every kind of Similarity-Weighted, accumulation calculating obtains again, it is achieved thereby that the dynamic adjustment of Feature Fusion.
S6, obtain according to merging the Search Results of personnel's image in the data base of sequencing of similarity respectively. Preferably, this Search Results can be displayed by web client, is shown by the picture sorted, it is possible to allows user find the target occurred under multi-point in very short time, greatly improves the efficiency of search. And user can also select the number of display result, for instance 50,100 etc.
Above-noted persons' searching method, by image is carried out dividing processing, effectively it is partitioned into the whole body of personnel targets, upper and lower half body region, such that it is able to carry out feature extraction according to image in region, making to extract targeted region more accurately, also reducing regional extent, thus reducing the noise impact on feature extraction, improve the accuracy and speed of feature extraction, and then the precision of retrieval result can be improved and improve recall precision. By extracting polytype characteristic information of image, and the sequence of result is scanned for according to fusion similarity, more ensure that the result searched all matches with personnel to be searched in polytype feature, further increase the accuracy of retrieval result.
Preferably, above-mentioned steps S2 includes:
S2-1, judge whether whether personnel to be searched proportion in the picture incomplete in the picture less than threshold value and personnel to be searched respectively; These two judge all to be possible after whom before no matter who, it is preferable that first determine whether that whether personnel to be searched proportion in the picture is less than threshold value, then judge whether personnel to be searched are incomplete in the picture. Or can also judge that whether scoring that full human model provides is lower than threshold value etc.
S2-2, when personnel to be searched proportion in the picture is complete in the picture less than threshold value (namely personnel only occupy very small part in the picture) or personnel to be searched or scoring lower than threshold value time, adopt GRABCUT algorithm that image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist; Such as, GRABCUT algorithm has good effect for the situation of serious shielding, particularly as follows: before first automatically choosing, background seed points, utilize kmeans algorithm that front, background seed are clustered (such as 5 class) respectively, then utilize GMM Gaussian Background model that the seed points after cluster is set up model, before finally utilizing minimal cut algorithm to carry out according to the distance between pixel and affiliated model, the judgement of background provide the whole body of personnel targets, upper and lower half body region.
S2-3, when other situations, namely when personnel to be searched proportion in the picture is full-time in the picture more than or equal to threshold value or personnel to be searched or scoring is higher than threshold value, adopt DDN algorithm that image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist. Such as, DDN algorithm can accurately be partitioned into pedestrian's foreground area of standard state and provide accurate pedestrian's whole body, upper and lower half body region, particularly as follows: first pass through off-line learning, full people's sample is trained, obtains deep decomposition network parameter W, B. Then zoom to network input picture size by unified for input picture and extract HOG feature, and HOG feature and network parameter W, B computing are obtained label image, before being derived from image, background area and the whole body of personnel targets, upper and lower half body region.
Above-noted persons' searching method, by judging whether whether personnel to be searched proportion in the picture incomplete in the picture less than threshold value and personnel to be searched respectively, prejudge the testing result of image, different partitioning algorithms is selected according to testing result, namely when personnel to be searched proportion in the picture is not full-time in the picture less than threshold value or personnel to be searched, adopt GRABCUT algorithm, other situations then adopt DDN algorithm, combine GRABCUT algorithm and the feature of DDN algorithm both algorithms, thus for different image anticipation results, more targetedly, better, obtain personnel whole body more accurately, on, lower part of the body region, more accurate basis is provided for follow-up feature extraction, and then the precision of feature extraction and the accuracy of retrieval result can be improved further.
Preferably, polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature etc. Multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD character reveal color characteristic, and HOG feature, CSURF feature instantiation go out textural characteristics, and CEDD feature, FCTH character reveal color and vein in conjunction with feature.
Preferably, above-mentioned steps S3 includes:
S3-1, when extracting multidimensional Dominant Color Features, respectively the color of image in each region is carried out rough segmentation and essence point, obtains rough segmentation color histogram and essence point color histogram respectively; Particularly as follows: the color rough segmentation to image, it is thus achieved that the step of 11 kinds of mass-tones: calculating the rgb value of current pixel point, RGB quantifies respectively to 32 bin, and obtains the index value of each pixel, table look-up and obtain the mass-tone value (0��10) of correspondence. Image carrying out the division of grid, and adds up the step of the average of each grid, variance: RGB image is converted into YcrCb color space, Y, Cr, Cb color component value simultaneously added up in each piece adds up, and the pixel number in statistics block; Calculate each color component in the average of each piece of the inside of 8x8, mean square deviation. Essence point color histogram obtains step: RGB color is transformed into HSL color model, in H component non-uniform quantizing to 11 bin of each pixel value, will can obtain 11 kinds of color value, and corresponding color index value is following table:
| Bin indexes | Color value |
| 0 | At prime |
| 1 | Orange |
| 2 | True yellow |
| 3 | Yellowish green |
| 4 | Just green |
| 5 | Turquoise |
| 6 | Just light blue |
| 7 | Shallow dark blue |
| 8 | Just dark blue |
| 9 | Royal purple |
| 10 | Fuchsin |
SL value is fitted (fit parameter values is empirical value), achromatic color value in 5 kinds of value of colors (saturated coloured silk color, dark, grey black color, white lime coloured silk color, bright) and 4 (black, grey black, white lime, in vain) can be obtained; Being combined by the colour of the color of H Yu SL matching, can obtain 11x5 kind value of color, add 4 kinds of achromatic color values of SL, in total 55+4=59, color is color; 59 kinds of colours of refinement are carried out global statistics, to combination 55 in value of color, the pixel number of 11 bin of statistics H, achromatic pixel number in statistics SL, it is divided into 3 classes (black, grey (grey black, white lime), white), can obtain, after comprehensive, the color feature value that 59+11+3=73 kind is fine; The fine color eigenvalue asked for is normalized, it is thus achieved that essence point color histogram.
S3-2, extract multichannel multiple color spaces assemblage characteristic time, respectively image in each region is carried out piecemeal, add up the color histogram of every piece of each color space of subgraph (such as 4 color spaces) respectively, each color histogram is carried out comprehensively acquisition Color rectangular histogram;
S3-3, extract CLD feature time, first divide the image into 64 pieces with the grid of 8*8, each piece extracts dominant color. Then dominant color is done the dct transform of 8*8, can obtain a series of DCT coefficient, then these coefficients are done Z-shaped scanning, find out low frequency coefficient therein and quantify, using the low frequency coefficient a kind of color characteristic as this target.
S3-4, extract CSD feature time, image is transformed into HMMD color space and quantifies, with the window of fixed size, quantized image is carried out slider-operated, and the color structure histogram information of image in statistical window, finally the rectangular histogram that statistics obtains is carried out nonlinear quantization, obtain final color structure histogram.
S3-5, extract SCD feature time, image is transformed into hsv color space, then becomes 16 deciles, S and V component to be quantized into 4 deciles H element quantization. Finally carry out Haar transform to often tieing up rectangular histogram coefficient, obtain 256 and maintain number, and 256 maintain numerical symbol.
S3-6, extract HOG feature time, respectively image in each region is carried out piecemeal, adds up gradient orientation histogram and the color histogram of every piece of subgraph respectively, undertaken gradient orientation histogram and color histogram comprehensively obtaining colored gradient orientation histogram. By increasing colouring information in gradient orientation histogram, it is possible to improve the robustness of algorithm.
S3-7, extract CSURF feature time, calculate RGB average (�� R, �� G, �� B) and the variance (�� R, �� G, �� B) of image respectively, carry out color space conversion according to the following formula.
Image after above-mentioned conversion is carried out piecemeal, adds up the haar response in 16 subdomains of every piece of subgraph, then through the CSURF feature being combined into subgraph after Gauss weighting, finally every piece of subgraph tandem compound is become final textural characteristics.
S3-8, extract CEDD feature time, this feature can be divided into color and texture two parts, color part is that image processes the 24 dimension mass-tones obtained through fuzzy theory, and texture is the texture information obtained after 6 texture formworks and image convolution, eventually passes Joint Distribution and obtains CEDD feature.
S3-9, extract FCTH feature time, this feature can be divided into color and texture two parts, color part image processes the 24 dimension mass-tones obtained through fuzzy theory, texture part has used the wavelet transformation of high frequency, split into 8 intervals according to different frequency band coefficients by fuzzy theory, in this, as texture information, eventually pass Joint Distribution and obtain FCTH feature.
Above-noted persons' searching method, by arranging multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and the polytype characteristic information of FCTH feature, it is adapted to different scene, there is different sign abilities, the robustness of boosting algorithm.
Preferably, above-mentioned steps S4 includes:
S4-1, for CEDD feature, CSD feature, multichannel multiple color spaces assemblage characteristic, FCTH feature adopt Tanimoto distance metric method carry out Similarity Measure; Tanimoto distance metric method is: assume to be calculated two characteristic vector respectively X, Y, wherein X=[x1,x2,��,xn], Y=[y1,y2,��,yn], then
Count1+=(X/SumX) * (Y/SumY);
Count2+=(Y/SumY) * (Y/SumY);
Count3+=(X/SumX) * (X/SumX);
The Similarity value of two characteristic vectors is:
S=(1-(Count1/ (Count2+Count3-Count1))).
S4-2, for CLD feature, CSURF feature adopt Euclidean distance measure carry out Similarity Measure.
S4-3,1 normal form distance metric method is adopted to carry out Similarity Measure for SCD feature.
S4-4, the Point matching method of minimizing in neighborhood is adopted to carry out Similarity Measure for HOG feature; In neighborhood, the Point matching method of minimizing is: for two images compared, first calculate the HOG feature of each sub-image and the Euclidean distance of interior other sub-images HOG feature of adjacent domain, by calculated minimum as the similarity of this block subgraph in two images, finally the similarity of cumulative all subgraphs is as the similarity of two images.
S4-5, for multidimensional Dominant Color Features adopt COS distance similarity calculating method carry out Similarity Measure. COS distance similarity calculating method is: calculates the COS distance of essence point color histogram and rough segmentation color histogram respectively, eventually passes the Similarity value of two figure that Weighted Fusion provides as this kind of feature.
Preferably, before step S2, above-noted persons' searching method also includes:
S7, image is carried out white balance process, it is thus achieved that white balance process after image. White balance processes: input picture is transformed into YCbCr color space, then input picture is carried out piecemeal, calculate the average Mb of Cb, the Cr in each region, Mr, calculate aggregate-value Db, Dr of absolute value according to the following formula,
Db=��i,j(| Cb (i, j)-Mb |)/N, Dr=��i,j(|Cr(i,j)-Mr|)/N
Finding the white point in image according to Db, Dr, its computing formula is as follows:
| Cb (i, j)-(Mb+Db �� sign (Mb)) | < 1.5 �� Db,
| Cr (i, j)-(1.5 �� Mr+Dr �� sign (Mr)) | < 1.5 �� Dr,
Finally choose front the 10% of white point as final white reference point, calculate the gain of each passage, then the color value of each passage is corrected, obtain the process image after white balance.
Above-noted persons' searching method, by the image received first being carried out AWB process, can eliminate due to image difference when multi-point is monitored, remove the impact of illumination etc., equilibrium figures image brightness, impact when reducing interference to feature extraction, improves the precision of feature extraction and the accuracy of retrieval result.
Preferably, after step s 3, before step S4, above-noted persons' searching method also includes:
S8, the high dimensional feature in the characteristic information extracted is carried out dimension-reduction treatment, reduce characteristic length; In order to tackle huge database search, reduce characteristic length and provide separating capacity and be characterized by solving the requisite means of matching efficiency and effect. Adopt PCA LDA algorithm, by the method for off-line training, it is possible to reducing while characteristic length, primitive character is mapped to the space more having separating capacity, greatly improves effect and the efficiency of search.
Embodiment 2
Corresponding to embodiment 1, the present embodiment provides a kind of people search's device, as in figure 2 it is shown, include:
Receive unit 1, for receiving the image comprising personnel to be searched;
Cutting unit 2, for carrying out dividing processing to image, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Feature extraction unit 3, for extracting polytype characteristic information of image in personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively;
First computing unit 4, for calculating in each type of characteristic information in each region and data base the similarity between individual features information respectively, data base is for storing polytype characteristic information of image in the polytype characteristic information and personnel's lower part of the body region including image in polytype characteristic information of image in the personnel whole body region that all personnel's image is corresponding, personnel region above the waist;
Second computing unit 5, computes weighted for weighted value shared by the similarity preset, and obtains the fusion similarity of image in personnel whole body region, personnel region above the waist and/or the personnel's lower part of the body region that in data base, each personnel's image is corresponding respectively;
Search Results obtains unit 6, for obtaining the Search Results of personnel's image in the data base according to fusion sequencing of similarity respectively.
Above-noted persons' searcher, by image is carried out dividing processing, effectively it is partitioned into the whole body of personnel targets, upper and lower half body region, such that it is able to carry out feature extraction according to image in region, making to extract targeted region more accurately, also reducing regional extent, thus reducing the noise impact on feature extraction, improve the accuracy and speed of feature extraction, and then the precision of retrieval result can be improved and improve recall precision. By extracting polytype characteristic information of image, and the sequence of result is scanned for according to fusion similarity, more ensure that the result searched all matches with personnel to be searched in polytype feature, further increase the accuracy of retrieval result.
Preferably, cutting unit 2 includes:
Judging unit, for judging whether whether personnel to be searched proportion in the picture incomplete in the picture less than threshold value and personnel to be searched respectively;
First segmentation result obtains unit, for when personnel to be searched proportion in the picture not full-time in the picture less than threshold value or personnel to be searched, adopt DDN algorithm that image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Second segmentation result obtains unit, for when personnel to be searched proportion in the picture full-time in the picture more than or equal to threshold value or personnel to be searched, adopt GRABCUT algorithm that image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist.
Above-noted persons' searcher, by judging whether whether personnel to be searched proportion in the picture incomplete in the picture less than threshold value and personnel to be searched respectively, prejudge the testing result of image, different partitioning algorithms is selected according to testing result, namely when personnel to be searched proportion in the picture is not full-time in the picture less than threshold value or personnel to be searched, adopt GRABCUT algorithm, other situations then adopt DDN algorithm, combine GRABCUT algorithm and the feature of DDN algorithm both algorithms, thus for different image anticipation results, more targetedly, better, obtain personnel whole body more accurately, on, lower part of the body region, more accurate basis is provided for follow-up feature extraction, and then the precision of feature extraction and the accuracy of retrieval result can be improved further.
Preferably, polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature.
Preferably, feature extraction unit 3 includes:
Multidimensional Dominant Color Features extraction unit, for when extracting multidimensional Dominant Color Features, the color of image in each region carrying out rough segmentation and essence point respectively, obtains rough segmentation color histogram and essence point color histogram respectively;
Multichannel multiple color spaces assemblage characteristic extraction unit, for when extracting multichannel multiple color spaces assemblage characteristic, respectively image in each region is carried out piecemeal, add up the color histogram of every piece of each color space of subgraph respectively, each color histogram is carried out comprehensively acquisition Color rectangular histogram;
HOG feature extraction unit, for when extracting HOG feature, respectively image in each region is carried out piecemeal, add up gradient orientation histogram and the color histogram of every piece of subgraph respectively, undertaken gradient orientation histogram and color histogram comprehensively obtaining colored gradient orientation histogram. Etc..
Above-noted persons' searcher, by arranging multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and the polytype characteristic information of FCTH feature, it is adapted to different scene, there is different sign abilities, the robustness of boosting algorithm.
Preferably, the first computing unit 4 includes:
First computation subunit, for adopting Tanimoto distance metric method to carry out Similarity Measure for CEDD feature, CSD feature, multichannel multiple color spaces assemblage characteristic, FCTH feature;
Second computation subunit, for adopting Euclidean distance measure to carry out Similarity Measure for CLD feature, CSURF feature;
3rd computation subunit, for adopting 1 normal form distance metric method to carry out Similarity Measure for SCD feature;
4th computation subunit, for adopting the Point matching method of minimizing in neighborhood to carry out Similarity Measure for HOG feature;
5th computation subunit, for adopting COS distance similarity calculating method to carry out Similarity Measure for multidimensional Dominant Color Features.
Preferably, before cutting unit 2, also include:
White balance processing unit, for carrying out white balance process to image, it is thus achieved that the image after white balance process.
Above-noted persons' searcher, by the image received first being carried out AWB process, can eliminate due to image difference when multi-point is monitored, remove the impact of illumination etc., equilibrium figures image brightness, impact when reducing interference to feature extraction, improves the precision of feature extraction and the accuracy of retrieval result.
Embodiment 3
The present embodiment provides a kind of people search's system, as it is shown on figure 3, include the deploying troops on garrison duty camera 10 in multiple some positions and processor 20;
Camera 10, comprises the image of personnel (pedestrian etc.) for Real-time Collection, for instance, camera has the function of Intelligent candid, it is possible to automatically extract the pedestrian in monitoring video, it is thus achieved that pedestrian's picture that these take off out; By Real-time Collection to image be sent to processor;
Processor 20, including real-time processing device 201 and people search's device 202;
Real-time processing device 201, for receive camera send Real-time Collection to image; Image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and personnel's lower part of the body region above the waist; Extract polytype characteristic information of image in personnel whole body region, personnel region above the waist and personnel's lower part of the body region respectively; Polytype characteristic information of image in personnel whole body region, personnel region above the waist and personnel's lower part of the body region is stored in data base; Or store in data base after converting polytype characteristic information to semantic feature. Polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature etc.
People search's device 202, for receiving the image comprising personnel to be searched; Image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist; Extract polytype characteristic information of image in personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively; Calculate the similarity between individual features information in each type of characteristic information in each region and data base respectively; Weighted value shared by default similarity computes weighted, and obtains the fusion similarity of image in personnel whole body region, personnel region above the waist and/or the personnel's lower part of the body region that in data base, each personnel's image is corresponding respectively; Obtain the Search Results of personnel's image in the data base according to fusion sequencing of similarity respectively. Or receive the semantic feature reflecting personnel to be searched, carry out specific search according in all semantic features that store in data base of semantic feature received, finally search out, with personnel targets to be searched, there is the image of identical semantic feature.
Above-noted persons search for system, achieve the accurate search to the target under different monitoring point position, web client can also be passed through the target searched be shown, user can be allowed to navigate to from the image of magnanimity within extremely short time information such as time place that target once occurred, greatly improve work efficiency.
Obviously, above-described embodiment is only for clearly demonstrating example, and is not the restriction to embodiment. For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description. Here without also cannot all of embodiment be given exhaustive. And the apparent change thus extended out or variation are still among the protection domain of the invention.
Claims (13)
1. people search's method, it is characterised in that comprise the following steps:
Receive the image comprising personnel to be searched;
Described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively;
Calculating the similarity between individual features information in each type of characteristic information in each region and data base respectively, described data base is for storing polytype characteristic information of image in the polytype characteristic information and personnel's lower part of the body region including image in polytype characteristic information of image in the personnel whole body region that all personnel's image is corresponding, personnel region above the waist;
Weighted value shared by default described similarity computes weighted, and obtains the fusion similarity of image in described personnel whole body region corresponding to each personnel's image in described data base, personnel region above the waist and/or personnel's lower part of the body region respectively;
Obtain the Search Results of personnel's image in the described data base according to described fusion sequencing of similarity respectively.
2. people search's method according to claim 1, it is characterised in that described described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist include:
Judge whether whether described personnel to be searched proportion in the picture incomplete in the picture less than threshold value and described personnel to be searched respectively;
When described personnel to be searched proportion in the picture is not full-time in the picture less than threshold value or described personnel to be searched, adopt GRABCUT algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
When described personnel to be searched proportion in the picture is full-time in the picture more than or equal to threshold value or described personnel to be searched, adopt DDN algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist.
3. people search's method according to claim 1 and 2, it is characterized in that, described polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature.
4. people search's method according to claim 3, it is characterised in that described extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively and include:
When extracting multidimensional Dominant Color Features, respectively the color of image in each region is carried out rough segmentation and essence point, obtain rough segmentation color histogram and essence point color histogram respectively;
When extracting multichannel multiple color spaces assemblage characteristic, respectively image in each region is carried out piecemeal, add up the color histogram of every piece of each color space of subgraph respectively, each color histogram is carried out comprehensively acquisition Color rectangular histogram;
When extracting HOG feature, respectively image in each region is carried out piecemeal, add up gradient orientation histogram and the color histogram of every piece of subgraph respectively, undertaken described gradient orientation histogram and color histogram comprehensively obtaining colored gradient orientation histogram.
5. the people search's method according to claim 3 or 4, it is characterised in that in the described each type of characteristic information calculating each region respectively and data base, the similarity between individual features information includes:
Tanimoto distance metric method is adopted to carry out Similarity Measure for CEDD feature, CSD feature, multichannel multiple color spaces assemblage characteristic, FCTH feature;
Euclidean distance measure is adopted to carry out Similarity Measure for CLD feature, CSURF feature;
1 normal form distance metric method is adopted to carry out Similarity Measure for SCD feature;
The Point matching method of minimizing in neighborhood is adopted to carry out Similarity Measure for HOG feature;
COS distance similarity calculating method is adopted to carry out Similarity Measure for multidimensional Dominant Color Features.
6. the people search's method according to any one of claim 1-5, it is characterised in that described image is carried out dividing processing described, it is thus achieved that before the step in personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist, also include:
Described image is carried out white balance process, it is thus achieved that the image after white balance process.
7. people search's device, it is characterised in that including:
Receive unit, for receiving the image comprising personnel to be searched;
Cutting unit, for carrying out dividing processing to described image, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Feature extraction unit, for extracting polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively;
First computing unit, for calculating in each type of characteristic information in each region and data base the similarity between individual features information respectively, described data base is for storing polytype characteristic information of image in the polytype characteristic information and personnel's lower part of the body region including image in polytype characteristic information of image in the personnel whole body region that all personnel's image is corresponding, personnel region above the waist;
Second computing unit, compute weighted for weighted value shared by the described similarity preset, obtain the fusion similarity of image in described personnel whole body region corresponding to each personnel's image in described data base, personnel region above the waist and/or personnel's lower part of the body region respectively;
Search Results obtains unit, for obtaining the Search Results of personnel's image in the described data base according to described fusion sequencing of similarity respectively.
8. people search's device according to claim 7, it is characterised in that described cutting unit includes:
Judging unit, for judging whether whether described personnel to be searched proportion in the picture incomplete in the picture less than threshold value and described personnel to be searched respectively;
First segmentation result obtains unit, for when described personnel to be searched proportion in the picture not full-time in the picture less than threshold value or described personnel to be searched, adopt GRABCUT algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist;
Second segmentation result obtains unit, for when described personnel to be searched proportion in the picture full-time in the picture more than or equal to threshold value or described personnel to be searched, adopt DDN algorithm that described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist.
9. the people search's device according to claim 7 or 8, it is characterized in that, described polytype characteristic information includes multidimensional Dominant Color Features, multichannel multiple color spaces assemblage characteristic, CLD feature, CSD feature, SCD feature, HOG feature, CSURF feature, CEDD feature and FCTH feature.
10. people search's device according to claim 9, it is characterised in that described feature extraction unit includes:
Multidimensional Dominant Color Features extraction unit, for when extracting multidimensional Dominant Color Features, the color of image in each region carrying out rough segmentation and essence point respectively, obtains rough segmentation color histogram and essence point color histogram respectively;
Multichannel multiple color spaces assemblage characteristic extraction unit, for when extracting multichannel multiple color spaces assemblage characteristic, respectively image in each region is carried out piecemeal, add up the color histogram of every piece of each color space of subgraph respectively, each color histogram is carried out comprehensively acquisition Color rectangular histogram;
HOG feature extraction unit, for when extracting HOG feature, respectively image in each region is carried out piecemeal, add up gradient orientation histogram and the color histogram of every piece of subgraph respectively, undertaken described gradient orientation histogram and color histogram comprehensively obtaining colored gradient orientation histogram.
11. the people search's device according to claim 9 or 10, it is characterised in that described first computing unit includes:
First computation subunit, for adopting Tanimoto distance metric method to carry out Similarity Measure for CEDD feature, CSD feature, multichannel multiple color spaces assemblage characteristic, FCTH feature;
Second computation subunit, for adopting Euclidean distance measure to carry out Similarity Measure for CLD feature, CSURF feature;
3rd computation subunit, for adopting 1 normal form distance metric method to carry out Similarity Measure for SCD feature;
4th computation subunit, for adopting the Point matching method of minimizing in neighborhood to carry out Similarity Measure for HOG feature;
5th computation subunit, for adopting COS distance similarity calculating method to carry out Similarity Measure for multidimensional Dominant Color Features.
12. the people search's device according to any one of claim 7-11, it is characterised in that before described cutting unit, also include:
White balance processing unit, for carrying out white balance process to described image, it is thus achieved that the image after white balance process.
13. people search's system, it is characterised in that include cloth and be placed on camera and the processor of multiple some position;
Described camera, comprises the image of personnel for Real-time Collection; By Real-time Collection to image be sent to described processor;
Described processor, including real-time processing device and people search's device;
Described real-time processing device, for receive described camera send Real-time Collection to image; Described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and personnel's lower part of the body region above the waist; Extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and personnel's lower part of the body region respectively; Polytype characteristic information of image in described personnel whole body region, personnel region above the waist and personnel's lower part of the body region is stored in data base;
Described people search's device, for receiving the image comprising personnel to be searched; Described image is carried out dividing processing, it is thus achieved that personnel whole body region, personnel region and/or personnel's lower part of the body region above the waist; Extract polytype characteristic information of image in described personnel whole body region, personnel region above the waist and/or personnel's lower part of the body region respectively; Calculate the similarity between individual features information in each type of characteristic information in each region and data base respectively; Weighted value shared by default described similarity computes weighted, and obtains the fusion similarity of image in described personnel whole body region corresponding to each personnel's image in described data base, personnel region above the waist and/or personnel's lower part of the body region respectively; Obtain the Search Results of personnel's image in the described data base according to described fusion sequencing of similarity respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510964095.3A CN105630906A (en) | 2015-12-21 | 2015-12-21 | Person searching method, apparatus and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510964095.3A CN105630906A (en) | 2015-12-21 | 2015-12-21 | Person searching method, apparatus and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105630906A true CN105630906A (en) | 2016-06-01 |
Family
ID=56045839
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510964095.3A Pending CN105630906A (en) | 2015-12-21 | 2015-12-21 | Person searching method, apparatus and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105630906A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169817A (en) * | 2017-04-20 | 2017-09-15 | 温州市鹿城区中津先进科技研究院 | Electric business platform exhibiting pictures colouring information acquisition method |
| CN107239780A (en) * | 2017-04-29 | 2017-10-10 | 安徽慧视金瞳科技有限公司 | A kind of image matching method of multiple features fusion |
| CN107330360A (en) * | 2017-05-23 | 2017-11-07 | 深圳市深网视界科技有限公司 | A kind of pedestrian's clothing colour recognition, pedestrian retrieval method and device |
| CN107506738A (en) * | 2017-08-30 | 2017-12-22 | 深圳云天励飞技术有限公司 | Feature extracting method, image-recognizing method, device and electronic equipment |
| CN108229493A (en) * | 2017-04-10 | 2018-06-29 | 商汤集团有限公司 | Banknote validation method, apparatus and electronic equipment |
| CN108629233A (en) * | 2017-03-20 | 2018-10-09 | 华为技术有限公司 | A kind of pedestrian retrieval method and device |
| CN108664526A (en) * | 2017-04-01 | 2018-10-16 | 华为技术有限公司 | The method and apparatus of retrieval |
| CN108694347A (en) * | 2017-04-06 | 2018-10-23 | 北京旷视科技有限公司 | Image processing method and device |
| CN108733679A (en) * | 2017-04-14 | 2018-11-02 | 华为技术有限公司 | A kind of pedestrian retrieval method, device and system |
| CN108764350A (en) * | 2018-05-30 | 2018-11-06 | 苏州科达科技股份有限公司 | Target identification method, device and electronic equipment |
| CN109766865A (en) * | 2019-01-19 | 2019-05-17 | 郑州天迈科技股份有限公司 | A kind of watershed, multizone local fusion and signature tracking passenger flow statistical method |
| CN109815874A (en) * | 2019-01-17 | 2019-05-28 | 苏州科达科技股份有限公司 | A kind of personnel identity recognition methods, device, equipment and readable storage medium storing program for executing |
| CN110322472A (en) * | 2018-03-30 | 2019-10-11 | 华为技术有限公司 | A kind of multi-object tracking method and terminal device |
| CN112287769A (en) * | 2020-10-09 | 2021-01-29 | 江汉大学 | Face detection method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102332034A (en) * | 2011-10-21 | 2012-01-25 | 中国科学院计算技术研究所 | Portrait picture retrieval method and device |
| CN103049446A (en) * | 2011-10-13 | 2013-04-17 | 中国移动通信集团公司 | Image retrieving method and device |
| CN103997624A (en) * | 2014-05-21 | 2014-08-20 | 江苏大学 | Overlapped domain dual-camera target tracking system and method |
| CN103996046A (en) * | 2014-06-11 | 2014-08-20 | 北京邮电大学 | Personnel recognition method based on multi-visual-feature fusion |
-
2015
- 2015-12-21 CN CN201510964095.3A patent/CN105630906A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049446A (en) * | 2011-10-13 | 2013-04-17 | 中国移动通信集团公司 | Image retrieving method and device |
| CN102332034A (en) * | 2011-10-21 | 2012-01-25 | 中国科学院计算技术研究所 | Portrait picture retrieval method and device |
| CN103997624A (en) * | 2014-05-21 | 2014-08-20 | 江苏大学 | Overlapped domain dual-camera target tracking system and method |
| CN103996046A (en) * | 2014-06-11 | 2014-08-20 | 北京邮电大学 | Personnel recognition method based on multi-visual-feature fusion |
Non-Patent Citations (6)
| Title |
|---|
| BHARATH HARIHARAN等: "Hypercolumns for object segmentation and fine-grained localization", 《2015 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| KEZE WANG等: "3D Human Activity Recognition with Reconfigurable Convolutional Neural Networks", 《PROCEEDINGS OF THE 22ND ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
| PING LUO等: "Pedestrian Parsing via Deep Decompositional Network", 《2013 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| WANLI OUYANG等: "Multi-source Deep Learning for Human Pose Estimation", 《2014 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| XINGYU ZENG等: "Multi-Stage Contextual Deep Learning for Pedestrian Detection", 《2013 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| 杨杰等: "《视频目标检测和跟踪及其应用》", 31 August 2012 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11250250B2 (en) | 2017-03-20 | 2022-02-15 | Huawei Technologies Co., Ltd. | Pedestrian retrieval method and apparatus |
| CN108629233A (en) * | 2017-03-20 | 2018-10-09 | 华为技术有限公司 | A kind of pedestrian retrieval method and device |
| CN108664526B (en) * | 2017-04-01 | 2022-04-29 | 华为技术有限公司 | Retrieval method and device |
| CN108664526A (en) * | 2017-04-01 | 2018-10-16 | 华为技术有限公司 | The method and apparatus of retrieval |
| CN108694347A (en) * | 2017-04-06 | 2018-10-23 | 北京旷视科技有限公司 | Image processing method and device |
| CN108694347B (en) * | 2017-04-06 | 2022-07-12 | 北京旷视科技有限公司 | Image processing method and device |
| CN108229493A (en) * | 2017-04-10 | 2018-06-29 | 商汤集团有限公司 | Banknote validation method, apparatus and electronic equipment |
| CN108733679A (en) * | 2017-04-14 | 2018-11-02 | 华为技术有限公司 | A kind of pedestrian retrieval method, device and system |
| CN108733679B (en) * | 2017-04-14 | 2021-10-26 | 华为技术有限公司 | Pedestrian retrieval method, device and system |
| CN107169817A (en) * | 2017-04-20 | 2017-09-15 | 温州市鹿城区中津先进科技研究院 | Electric business platform exhibiting pictures colouring information acquisition method |
| CN107239780A (en) * | 2017-04-29 | 2017-10-10 | 安徽慧视金瞳科技有限公司 | A kind of image matching method of multiple features fusion |
| CN107330360A (en) * | 2017-05-23 | 2017-11-07 | 深圳市深网视界科技有限公司 | A kind of pedestrian's clothing colour recognition, pedestrian retrieval method and device |
| CN107506738A (en) * | 2017-08-30 | 2017-12-22 | 深圳云天励飞技术有限公司 | Feature extracting method, image-recognizing method, device and electronic equipment |
| CN110322472A (en) * | 2018-03-30 | 2019-10-11 | 华为技术有限公司 | A kind of multi-object tracking method and terminal device |
| CN108764350A (en) * | 2018-05-30 | 2018-11-06 | 苏州科达科技股份有限公司 | Target identification method, device and electronic equipment |
| CN109815874A (en) * | 2019-01-17 | 2019-05-28 | 苏州科达科技股份有限公司 | A kind of personnel identity recognition methods, device, equipment and readable storage medium storing program for executing |
| CN109766865A (en) * | 2019-01-19 | 2019-05-17 | 郑州天迈科技股份有限公司 | A kind of watershed, multizone local fusion and signature tracking passenger flow statistical method |
| CN112287769A (en) * | 2020-10-09 | 2021-01-29 | 江汉大学 | Face detection method, device, equipment and storage medium |
| CN112287769B (en) * | 2020-10-09 | 2024-03-12 | 江汉大学 | Face detection method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105630906A (en) | Person searching method, apparatus and system | |
| CN105023008B (en) | The pedestrian of view-based access control model conspicuousness and multiple features recognition methods again | |
| CN103186904B (en) | Picture contour extraction method and device | |
| CN104966085B (en) | A kind of remote sensing images region of interest area detecting method based on the fusion of more notable features | |
| CN103839065B (en) | Extraction method for dynamic crowd gathering characteristics | |
| CN103577475B (en) | A kind of picture mechanized classification method, image processing method and its device | |
| CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
| CN103632158B (en) | Forest fire prevention monitor method and forest fire prevention monitor system | |
| CN104732200B (en) | A kind of recognition methods of skin type and skin problem | |
| CN107066972B (en) | Natural scene Method for text detection based on multichannel extremal region | |
| CN106447646A (en) | Quality blind evaluation method for unmanned aerial vehicle image | |
| CN104715251B (en) | A kind of well-marked target detection method based on histogram linear fit | |
| CN104408733B (en) | Object random walk-based visual saliency detection method and system for remote sensing image | |
| CN105426828A (en) | Face detection method, face detection device and face detection system | |
| CN104715244A (en) | Multi-viewing-angle face detection method based on skin color segmentation and machine learning | |
| CN105049790A (en) | Video monitoring system image acquisition method and apparatus | |
| CN107038416A (en) | A kind of pedestrian detection method based on bianry image modified HOG features | |
| CN112001241B (en) | Micro-expression recognition method and system based on channel attention mechanism | |
| CN109191428A (en) | Full-reference image quality evaluating method based on masking textural characteristics | |
| CN109271932A (en) | Pedestrian based on color-match recognition methods again | |
| CN109344842A (en) | A kind of pedestrian's recognition methods again based on semantic region expression | |
| CN107818303A (en) | Unmanned plane oil-gas pipeline image automatic comparative analysis method, system and software memory | |
| CN110503017A (en) | Wisdom energy conservation occupancy detection system and method based on image procossing | |
| CN110298893A (en) | A kind of pedestrian wears the generation method and device of color identification model clothes | |
| CN106909883A (en) | A kind of modularization hand region detection method and device based on ROS |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160601 |