CN109614877A - There is the pedestrian's attribute recognition approach blocked in low resolution monitoring scene - Google Patents
There is the pedestrian's attribute recognition approach blocked in low resolution monitoring scene Download PDFInfo
- Publication number
- CN109614877A CN109614877A CN201811370580.8A CN201811370580A CN109614877A CN 109614877 A CN109614877 A CN 109614877A CN 201811370580 A CN201811370580 A CN 201811370580A CN 109614877 A CN109614877 A CN 109614877A
- Authority
- CN
- China
- Prior art keywords
- image
- pedestrian
- blocked
- attribute
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention discloses there is the pedestrian's attribute recognition approach blocked in a kind of low resolution monitoring scene, first with the shelter in metric learning removal pedestrian image;Transverse cuts are carried out to the image after reparation, corresponding image block is respectively labeled as pedestrian's " head-and-shoulder area ", " upper body part " and " lower body part ";The feature of every block of image of the attribute corresponding part to be identified is extracted, and is 16-bin histogram by each character representation;Finally using 16-bin histogram as the feature vector of image, and the feature vector of image is inputted into trained SVM classifier and obtains recognition result.Classifying quality of the present invention is more preferable, and classification accuracy is higher.
Description
Technical field
The invention belongs to the feature identification techniques of pedestrian, and there is the pedestrian blocked to belong to specially in low resolution monitoring scene
Property recognition methods.
Background technique
Pedestrian is the main research object in monitoring scene.The essential attribute of identification pedestrian has very meaning.For
Facilitate the inquiry and retrieval tasks in conventional monitoring systems, it is necessary to artificially mark the essential attribute of pedestrian, then be sense
The pedestrian of interest inputs attribute tags.But under huge monitoring data, this work, which is not able to satisfy, marks pedestrian's attribute
Needs.Marking pedestrian's attribute automatically in monitoring by computer vision technique is a kind of effective method.
Pedestrian's attribute recognition approach is the algorithm that pedestrian's attribute in image is obtained by the processing to input picture, the present invention
Relate generally to Image Acquisition, image preprocessing, image enhancement, attributes extraction, attributive classification scheduling algorithm step.Wherein image enhancement
It is two critical issues with attributive classification.
It has possessed some special knowledge in humanized identification scene of being expert at present, if Layne et al. is in " Person re-
Identification in identification by attributes " with SVM algorithm to pedestrian's attribute, Deng et al.
Pedestrian's database (PETA) is introduced in " Pedestrian attribute recognition at far distance " to knowledge
The optimization of other algorithm.But they lack to the Study of recognition under low resolution figure field scape, so transporting under actual monitored scene
Centainly limited to existing.
In addition, to also possessing some special knowledge in the gender attribute identification of pedestrian, as Sun et al. utilizes genetic algorithm to the portion of face
Divide characteristic point as characteristic of division, the classifier by building artificial neural network as identification;There are also the parts two that Luo is proposed
Meta schema (LBP) feature solves multi-orientation Face identification;Amit Jain extracts face characteristic using independent component analysis (ICA),
And SVM classifier is combined to identify.Equally, they do not consider the application under monitoring scene.
Appearance attribute is mainly reflected in the understanding to pedestrian's clothes and the important content of pedestrian's Attribute Recognition.Pass through
Whether the understanding to clothes appearance can provide semantic attribute, the color including clothes, style, if wear glasses, and carry
Packaging.In recent years, many researchers are effectively recognized by the understanding to pedestrian's clothes, the background information of combining environmental
The basic appearance of pedestrian image.Gallagher and Chen et al. propose a kind of calculation that dress form is practised from one group of image middle school
Method come solve the problems, such as how from image divide clothes region.But likewise, if pedestrian image is blocked, clothes
Region is more difficult to divide.
Summary of the invention
It is an object of the invention to propose to have the pedestrian Attribute Recognition side blocked in a kind of low resolution monitoring scene
Method.
Realize technical solution of the invention are as follows: there is the pedestrian Attribute Recognition side blocked in low resolution monitoring scene
Method, specific steps are as follows:
Step 1, utilization measure study carry out reparation operation to the pedestrian image being blocked, and remove blocking in pedestrian image
Object;
Step 2 carries out transverse cuts to the image after reparation, and corresponding image block is respectively labeled as pedestrian " head and shoulder portion
Point ", " upper body part " and " lower body part ";
Step 3, the determination attribute to be identified extract the feature of every block of image of the attribute corresponding part, specific features packet
It includes: color characteristic, LBP feature, Gabor filter feature, Schmid filter characteristic, and be 16-bin by each character representation
Histogram;
Step 4, using the 16-bin histogram in step 3 as the feature vector of image, and it is the feature vector of image is defeated
Enter trained SVM classifier and obtains recognition result.
Preferably, utilization measure learns to carry out repairing the specific steps of operation to the pedestrian image being blocked in step 1 are as follows:
Step 1-1, the complete pedestrian image data set P of the information not being blocked is given;
Step 1-2, the given pedestrian image I being blocked and tag image I are blocked part, in pedestrian image data set P
It is middle that the pedestrian image T most like with the non-sheltering part of image I is found using hash algorithm, with pedestrian image T completion pedestrian image I
Be blocked part.
Preferably, it is found and the non-sheltering part of image I in pedestrian image data set P using hash algorithm in step 1-2
The specific steps of most like pedestrian image T are as follows:
Step 1-2-1, basis blocks label and cuts out the non-shield portions I1 of pedestrian image I;
Step 1-2-2, by the non-shield portions I1 gray processing of image and N*N size is normalized to, is denoted as image G;
Step 1-2-3, the average gray a for calculating all pixels in image G, by the gray scale of pixel each in image G
Value is compared with average value a, if the gray value of the pixel is greater than average value a, which is denoted as ' 1 ', if should
The gray value of pixel be less than average value a be then denoted as ' 0 ', by this by non-shield portions I1 be converted into 01 in the form of character string, note
For the Hash fingerprint of non-shield portions I1;
Step 1-2-4, the image in data set P in each image with the non-shield portions I1 corresponding position of image is cut out
P1, and the Hash fingerprint of image P1 is obtained using step 1-2-2~step 1-2-3 method;
Step 1-2-5, it is respectively compared each image in the Hash fingerprint and image set P1 of the non-shield portions I1 of image
The similarity size of Hash fingerprint finds out pedestrian image T most like with image I in data set P.
Preferably, in step 1-2-3 the Hash fingerprint of image I in data set P the Hash fingerprint of each image it is similar
Degree is " Hamming distance ", and " Hamming distance " is defined as the kinds of characters of corresponding position in compare two isometric character strings
Number, the more few then two images of number are more similar.
Preferably, the value range of N is 8~12.
Preferably, the corresponding relationship of the attribute and body part to be identified in step 3 is not: attribute " glasses " and " hair " belong to
In " head-and-shoulder area ";Attribute " V neck ", " trade mark " and " knapsack " belongs to " upper body part ", and attribute " shorts " and " jeans " belong to
" lower body part " belongs to entire pedestrian image for global property " male ", " women ".
Preferably, the kernel function of SVM classifier described in step 4 is histogram intersection core, the trained SVM classifier
For the optimum classifier for using the feature vector for the image set respective attributes not being blocked to be trained as training set.
Compared with prior art, the present invention its remarkable advantage are as follows: (1) classifying quality of the present invention is more preferable, and classification accuracy is more
It is high;(2) present invention has stronger robustness when facing the pedestrian image of low resolution;(3) by the present invention in that with measurement
Study can restore the image being blocked and carry out attributive classification, and recognition effect is more preferable.
Further detailed description is done to the present invention with reference to the accompanying drawing.
Detailed description of the invention
Fig. 1 is the overall procedure in monitoring scene under low resolution of the invention with the pedestrian's attribute recognition approach blocked
Figure.
Fig. 2 is rgb space image and each channel image.
Fig. 3 is YCbCr space image and each channel image.
Fig. 4 is HSV space image and each channel image.
Fig. 5 is image LBP feature, a left side be RGB image, in be gray level image, the right side be LBP map.
Fig. 6 is the characteristics of image map that Gabor filter is extracted under different parameters
Fig. 7 is the characteristics of image map that Schmid filter extracts under different parameters
Specific embodiment
There are the pedestrian's attribute recognition approach blocked, specific steps in low resolution monitoring scene are as follows:
Step 1, utilization measure study carry out reparation operation to the pedestrian image being blocked, and remove blocking in pedestrian image
Object, specific steps are as follows:
Step 1-1, the complete pedestrian image data set P of the information not being blocked is given;
Step 1-2, the given pedestrian image I being blocked and tag image I are blocked part, in pedestrian image data set P
It is middle that the pedestrian image T most like with the non-sheltering part of image I is found using hash algorithm, with pedestrian image T completion pedestrian image I
Be blocked part.
In further embodiment, found and image I in pedestrian image data set P using hash algorithm in step 1-2
The specific steps of the most like pedestrian image T of non-sheltering part are as follows:
Step 1-2-1, basis blocks label and cuts out the non-shield portions I1 of pedestrian image I;
Step 1-2-2, by the non-shield portions I1 gray processing of image and N*N size is normalized to, is denoted as image G;
Step 1-2-3, the average gray a for calculating all pixels in image G, by the gray scale of pixel each in image G
Value is compared with average value a, if the gray value of the pixel is greater than average value a, which is denoted as ' 1 ', if should
The gray value of pixel be less than average value a be then denoted as ' 0 ', by this by non-shield portions I1 be converted into 01 in the form of character string, note
For the Hash fingerprint of non-shield portions I1;
Step 1-2-4, the image in data set P in each image with the non-shield portions I1 corresponding position of image is cut out
P1, and the Hash fingerprint of image P1 is obtained using step 1-2-2~step 1-2-3 method;
Step 1-2-5, it is respectively compared each image in the Hash fingerprint and image set P1 of the non-shield portions I1 of image
The similarity size of Hash fingerprint finds out pedestrian image T most like with image I in image P1.
In certain embodiments, the Hash fingerprint of image I and the similarity of the Hash fingerprint of each image in data set P are
" Hamming distance ", " Hamming distance " are defined as of the kinds of characters of corresponding position in compare two isometric character strings
Number, the more few then two images of number are more similar.
Step 2 carries out transverse cuts to the image after reparation, and corresponding image block is respectively labeled as pedestrian " head and shoulder portion
Point ", " upper body part " and " lower body part ";In certain embodiments, pedestrian image is laterally averagely cut into 10 pieces, label
It is 1~10 piece, is labeled as pedestrian " head-and-shoulder area " for the 1st piece to the 3rd piece, the 3rd piece to the 7th piece is labeled as pedestrian's " upper body
Point ", the 6th piece to the 10th piece is labeled as pedestrian " lower body part ";
Step 3, the determination attribute to be identified extract the feature of every block of image of the attribute corresponding part, specific features packet
It includes: color characteristic, LBP feature, Gabor filter feature, Schmid filter characteristic, and be 16-bin by each character representation
Histogram;In certain embodiments, the corresponding relationship of the attribute and body part to be identified are as follows: attribute " glasses " and " hair " belong to
In " head-and-shoulder area ";Attribute " V neck ", " trade mark " and " knapsack " belongs to " upper body part ", and attribute " shorts " and " jeans " belong to
" lower body part " belongs to entire pedestrian image for global property " male ", " women ".
Step 4, using the 16-bin histogram in step 3 as the feature vector of image, and it is the feature vector of image is defeated
Enter trained SVM classifier and obtains recognition result.
In certain embodiments, the training process of trained SVM classifier are as follows:
Transverse cuts are carried out to each image for the image set not being blocked, corresponding image block is respectively labeled as pedestrian
" head-and-shoulder area ", " upper body part " and " lower body part ";The feature of every block of image of part corresponding to respective attributes is extracted, it will
Each character representation is 16-bin histogram as the feature vector of training set and test set image training SVM classifier.Make
The classification results that SVM classifier is assessed with 5 folding cross validations, using the optimal model of classifying quality as optimal svm classifier
Device.
Embodiment
As shown in Figure 1, having the pedestrian's attribute recognition approach blocked, specific steps in low resolution monitoring scene are as follows:
Step 1, utilization measure study carry out reparation operation to the pedestrian image being blocked, and remove blocking in pedestrian image
Object, specific steps are as follows:
Step 1-1, the complete pedestrian image data set P of the information not being blocked is given;
Step 1-2, the given pedestrian image I being blocked and tag image I are blocked part, cut out according to label is blocked
The non-shield portions I1 of pedestrian image I;
By the non-shield portions I1 gray processing of image and N*N size is normalized to, is denoted as image G;
The average gray a for calculating all pixels in image G, by the gray value of pixel each in image G and average value a
It compares, if the gray value of the pixel is greater than average value a, which is denoted as ' 1 ', if the gray scale of the pixel
Value be less than average value a be then denoted as ' 0 ', by this by non-shield portions I1 be converted into 01 in the form of character string, be denoted as non-shield portions
The Hash fingerprint of I1;
The image P1 in data set P in each image with the non-shield portions I1 corresponding position of image is cut out, and utilizes step
The method of rapid 1-2-2~step 1-2-3 obtains the Hash fingerprint of image P1;
Step 1-2-5, it is respectively compared each image in the Hash fingerprint and image set P1 of the non-shield portions I1 of image
The similarity size of Hash fingerprint finds out pedestrian image T most like with image I in image P1.
It is blocked part with the part corresponding position image completion pedestrian image I that is blocked in pedestrian image T with image I.
Step 2 cuts the image after reparation, and pedestrian image is laterally averagely cut into 10 pieces, is labeled as 1~10
Block, we are labeled as pedestrian " head-and-shoulder area " for the 1st piece to the 3rd piece, and the 3rd piece to the 7th piece is labeled as pedestrian " upper body part ", the
6 pieces to the 10th piece are labeled as pedestrian " lower body part ";Different property distributions in different body parts, attribute " glasses " and
" hair " belongs to " head-and-shoulder area ", and attribute " V neck ", " trade mark " and " knapsack " belongs to " upper body part ", attribute " shorts " and " cowboy
Trousers " belong to " lower body part ", belong to entire pedestrian image for global property " male ", " women ".
Step 3, the determination attribute to be identified extract the feature of every block of image of the attribute corresponding part, specific features packet
It includes: color characteristic, LBP feature, Gabor filter feature, Schmid filter characteristic, and be 16-bin by each character representation
Histogram;
As shown in Figure 2, Figure 3, Figure 4, it is known that different colours space difference channel has different color characteristics.LBP feature
It is as shown in Figure 5 to extract image local textural characteristics.The parameter selection of Gabor filter is as shown in table 1, Gabor selected by 1 parameter of table
The image that filter extracts is as shown in Figure 6.The parameter selection of Schmid filter is as shown in table 2, the filter of Gabor selected by 1 parameter of table
The image that wave device extracts is as shown in Figure 7.
Table 1
Table 2
Step 4, using the 16-bin histogram in step 3 as the feature vector of image, and it is the feature vector of image is defeated
Enter trained SVM classifier and obtain recognition result, the kernel function of the SVM classifier is histogram intersection core.
Table 3
The present embodiment the Sub Data Set VIPeR respective attributes of PETA data set feature vector as the attributive classification device
Training set and test set.VIPeR data set details is as shown in table 3.The present embodiment is by the feature vector of training set and test set
It is mapped to higher dimensional space by histogram intersection kernel function, is classified with SVM classifier.
Histogram intersection kernel function is defined as:In formula, its training set sample of n
Number, (x (i), y (i) are training set sample.
The present embodiment is that classifier is respectively trained in each attribute, i.e., cuts to data set sample, by pedestrian image cross
To being averagely cut into 10 pieces, it is labeled as 1~10 piece, is labeled as pedestrian " head-and-shoulder area " for the 1st piece to the 3rd piece, the 3rd piece to the 7th
Block is labeled as pedestrian " upper body part ", and the 6th piece to the 10th piece is labeled as pedestrian " lower body part ", extracts the every of attribute corresponding part
The feature of block image, specific features include: color characteristic, LBP feature, Gabor filter feature, Schmid filter characteristic,
It is 16-bin histogram as the feature vector of image using each character representation, for the training of unbalanced attribute, using random
Down-sampling handles unbalanced data, carrys out assessment models quality using 5 folding cross validations, final to save the optimal mould of classifying quality
Type is as optimal SVM classifier.Eight attributes are selected to carry out the performance of testing attribute classification in the present embodiment.For each category
Property, the present embodiment carries out 10 samplings to training data, and using the average value of ten results as final classification result.This implementation
Example is compared with the method proposed in " Person re-identification by attributes ", and comparison result is known
Other accuracy is as shown in table 5.
Table 5
Claims (7)
1. having the pedestrian's attribute recognition approach blocked in low resolution monitoring scene, which is characterized in that specific steps are as follows:
Step 1, utilization measure study carry out reparation operation to the pedestrian image being blocked, and remove the shelter in pedestrian image;
Step 2, to after reparation image carry out transverse cuts, by corresponding image block be respectively labeled as pedestrian's " head-and-shoulder area ", " on
Body part " and " lower body part ";
Step 3, the determination attribute to be identified, extract the feature of every block of image of the attribute corresponding part, and specific features include: face
Color characteristic, LBP feature, Gabor filter feature, Schmid filter characteristic, and be 16-bin histogram by each character representation
Figure;
Step 4, using the 16-bin histogram in step 3 as the feature vector of image, and by the feature vector of image input instruct
The SVM classifier perfected obtains recognition result.
2. having the pedestrian's attribute recognition approach blocked, feature in low resolution monitoring scene according to claim 1
It is, utilization measure study repair to the pedestrian image being blocked the specific steps of operation in step 1 are as follows:
Step 1-1, the complete pedestrian image data set P of the information not being blocked is given;
Step 1-2, the given pedestrian image I being blocked and tag image I are blocked part, make in pedestrian image data set P
The pedestrian image T most like with the non-sheltering part of image I is found with hash algorithm, is hidden with pedestrian image T completion pedestrian image I
Stopper point.
3. having the pedestrian's attribute recognition approach blocked, feature in low resolution monitoring scene according to claim 2
It is, finds the row most like with the non-sheltering part of image I in step 1-2 using hash algorithm in pedestrian image data set P
The specific steps of people's image T are as follows:
Step 1-2-1, basis blocks label and cuts out the non-shield portions I1 of pedestrian image I;
Step 1-2-2, by the non-shield portions I1 gray processing of image and N*N size is normalized to, is denoted as image G;
Step 1-2-3, calculate image G in all pixels average gray a, by the gray value of pixel each in image G with
Average value a is compared, if the gray value of the pixel is greater than average value a, which is denoted as ' 1 ', if the pixel
Point gray value be less than average value a be then denoted as ' 0 ', by this by non-shield portions I1 be converted into 01 in the form of character string, be denoted as not
The Hash fingerprint of shield portions I1;
Step 1-2-4, the image construction figure in data set P in each image with the non-shield portions I1 corresponding position of image is cut out
Image set P1, and the Hash fingerprint of image P1 is obtained using step 1-2-2~step 1-2-3 method;
Step 1-2-5, the Hash of the Hash fingerprint and each image in image set P1 that are respectively compared the non-shield portions I1 of image refers to
The similarity size of line finds out pedestrian image T most like with image I in data set P.
4. having the pedestrian's attribute recognition approach blocked, feature in low resolution monitoring scene according to claim 3
It is, the similarity of the Hash fingerprint of image I and the Hash fingerprint of each image in data set P is " Hamming distance in step 1-2-3
From ", " Hamming distance " is defined as the number of the kinds of characters of corresponding position in compare two isometric character strings, and number is got over
At least two images are more similar.
5. having the pedestrian's attribute recognition approach blocked, feature in low resolution monitoring scene according to claim 3
It is, the value range of N is 8~12.
6. having the pedestrian's attribute recognition approach blocked, feature in low resolution monitoring scene according to claim 1
Be, the corresponding relationship of the attribute and body part to be identified in step 3 not: attribute " glasses " and " hair " belong to " head and shoulder portion
Point ";Attribute " V neck ", " trade mark " and " knapsack " belongs to " upper body part ", and attribute " shorts " and " jeans " belong to " lower part of the body portion
Point ", entire pedestrian image is belonged to for global property " male ", " women ".
7. having the pedestrian's attribute recognition approach blocked, feature in low resolution monitoring scene according to claim 1
Be, the kernel function of SVM classifier described in step 4 is histogram intersection core, the trained SVM classifier be using not by
The optimum classifier that the feature vector for the image set respective attributes blocked is trained as training set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811370580.8A CN109614877B (en) | 2018-11-17 | 2018-11-17 | Method for identifying attribute of pedestrian with shielding in low-resolution monitoring scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811370580.8A CN109614877B (en) | 2018-11-17 | 2018-11-17 | Method for identifying attribute of pedestrian with shielding in low-resolution monitoring scene |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109614877A true CN109614877A (en) | 2019-04-12 |
CN109614877B CN109614877B (en) | 2022-09-27 |
Family
ID=66003431
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811370580.8A Active CN109614877B (en) | 2018-11-17 | 2018-11-17 | Method for identifying attribute of pedestrian with shielding in low-resolution monitoring scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109614877B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472514A (en) * | 2019-07-22 | 2019-11-19 | 电子科技大学 | A kind of adaptive vehicle algorithm of target detection model and its construction method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106127164A (en) * | 2016-06-29 | 2016-11-16 | 北京智芯原动科技有限公司 | The pedestrian detection method with convolutional neural networks and device is detected based on significance |
CN108388850A (en) * | 2018-02-08 | 2018-08-10 | 天津大学 | A kind of human motion recognition method based on k arest neighbors and micro-Doppler feature |
-
2018
- 2018-11-17 CN CN201811370580.8A patent/CN109614877B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106127164A (en) * | 2016-06-29 | 2016-11-16 | 北京智芯原动科技有限公司 | The pedestrian detection method with convolutional neural networks and device is detected based on significance |
CN108388850A (en) * | 2018-02-08 | 2018-08-10 | 天津大学 | A kind of human motion recognition method based on k arest neighbors and micro-Doppler feature |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472514A (en) * | 2019-07-22 | 2019-11-19 | 电子科技大学 | A kind of adaptive vehicle algorithm of target detection model and its construction method |
CN110472514B (en) * | 2019-07-22 | 2022-05-31 | 电子科技大学 | Adaptive vehicle target detection algorithm model and construction method thereof |
Also Published As
Publication number | Publication date |
---|---|
CN109614877B (en) | 2022-09-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
WO2018072233A1 (en) | Method and system for vehicle tag detection and recognition based on selective search algorithm | |
CN102012939B (en) | Method for automatically tagging animation scenes for matching through comprehensively utilizing overall color feature and local invariant features | |
CN102096823A (en) | Face detection method based on Gaussian model and minimum mean-square deviation | |
CN109583482A (en) | A kind of infrared human body target image identification method based on multiple features fusion Yu multicore transfer learning | |
US20110142335A1 (en) | Image Comparison System and Method | |
CN108898125A (en) | One kind being based on embedded human face identification and management system | |
CN104036247A (en) | Facial feature based face racial classification method | |
CN111126240B (en) | Three-channel feature fusion face recognition method | |
WO2021082168A1 (en) | Method for matching specific target object in scene image | |
CN103093240A (en) | Calligraphy character identifying method | |
CN102332086A (en) | Facial identification method based on dual threshold local binary pattern | |
CN105138968A (en) | Face authentication method and device | |
CN106021603A (en) | Garment image retrieval method based on segmentation and feature matching | |
CN105550657A (en) | Key point based improved SIFT human face feature extraction method | |
CN108171223A (en) | A kind of face identification method and system based on multi-model multichannel | |
CN103218609A (en) | Multi-pose face recognition method based on hidden least square regression and device thereof | |
CN104636755A (en) | Face beauty evaluation method based on deep learning | |
CN101853397A (en) | Bionic human face detection method based on human visual characteristics | |
CN104299003A (en) | Gait recognition method based on similar rule Gaussian kernel function classifier | |
CN110427821A (en) | A kind of method for detecting human face and system based on lightweight convolutional neural networks | |
CN104598885A (en) | Method for detecting and locating text sign in street view image | |
CN103020614B (en) | Based on the human motion identification method that space-time interest points detects | |
CN106599785A (en) | Method and device for building human body 3D feature identity information database | |
CN104794693A (en) | Human image optimization method capable of automatically detecting mask in human face key areas |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |