CN109543546B - Gait age estimation method based on depth sequence distribution regression - Google Patents
Gait age estimation method based on depth sequence distribution regression Download PDFInfo
- Publication number
- CN109543546B CN109543546B CN201811255447.8A CN201811255447A CN109543546B CN 109543546 B CN109543546 B CN 109543546B CN 201811255447 A CN201811255447 A CN 201811255447A CN 109543546 B CN109543546 B CN 109543546B
- Authority
- CN
- China
- Prior art keywords
- gait
- age
- distribution
- global
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of computer vision and machine learning, and particularly relates to a gait age estimation method based on depth sequence distribution regression. The invention comprises the following steps: the method comprises the steps of preprocessing a gait video sequence, shooting the preprocessed video sequence into a gait energy image, and dividing the gait energy image into three parts, namely a head part, a body part and a lower body part which are not overlapped; performing data division by using a sequential regression method, and converting the K classification problem into a series of two classification problems for the ordered K classification problem, namely training K-1 two classifiers; extracting features by using a global convolutional neural network and a local convolutional neural network, namely extracting global gait features and three local gait features by using a global network and three sub-networks respectively; the network model is trained using an order distribution loss function, wherein the loss function is designed to take into account cross-entropy losses and distribution losses. The invention can better optimize the network and obviously improve the accuracy of gait age prediction.
Description
Technical Field
The invention belongs to the technical field of computer vision and video safety monitoring, and particularly relates to a gait age estimation method based on deep learning and sequence distribution regression.
Background
Age estimation is one of the hot problems in the field of computer vision, especially in the field of video surveillance. Age estimation methods can be divided into two categories, namely age estimation methods based on human faces and age estimation methods based on gait video sequences. The invention is based on the age estimation of gait, a section of pedestrian gait video sequence is given, and the specific age of a target (gait) is required to be predicted by utilizing a machine learning technology or a computer vision method. Gait age estimation methods do not work much, and currently existing methods can be divided into three categories: regression-based methods, classification-based methods, and prevalence learning-based methods. The following are some references to these three types of processes:
[1]Smola,A.J.,and Scho
B.2004.A tutorial on support vector regression.Statistics and computing 14(3):199–222.
[2]Rasmussen,C.E.2004.Gaussian processes in machine learning.In Advanced Lectures on Machine Learning.63–71.
[3]Lu,J.,and Tan,Y.-P.2010a.Gait-based human age estimation.IEEE Transactions on Information Forensics and Security 5(4):761–770.
[4]Lu,J.,and Tan,Y.-P.2010b.Ordinary preserving manifold analysis for human age estimation.In IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops,90–95.
[5]Lu,J.,and Tan,Y.-P.2013.Ordinary preserving manifold analysis for human age and head pose estimation.IEEE Transactions on Human-Machine Systems 43(2):249–258.
[6]Xu,C.；Makihara,Y.；Ogi,G.；Li,X.；Yagi,Y.；and Lu,J.2017.The OU-ISIR gait database comprising the large population dataset with age and performance evaluation of age estimation.IPSJ Transactions on Computer Vision and Applications 9(1):24.
[7]Li,X.；Makihara,Y.；Xu,C.；Yagi,Y.；and Ren,M.2018.Gait-based human age estimation using age group-dependent manifold learning and regression.Multimedia Tools and Applications 1–22.。
regression-based methods. The early stage gait age estimation mainly uses GEI characteristic diagram to extract manual characteristics by LBP, HOG and other methods, and then uses mature regression method, such as SVR 1 or GPR 2, to make age regression prediction for gait. The method is simple and direct, but has poor effect, and the feature extraction and the model training are separated. Recent work on regression-based methods, [7] proposed an age group classification-based method and an intra-group regression method (ASSOLPP). After training the age group classifier, regression was done within each age group using KernelSVM.
A classification based approach. In addition to regarding the age estimation problem as regression, there are also people regarding classification problems, regarding each age as a category. For example, a method for binary encoding age is proposed in [3], and the age estimation problem after encoding becomes a typical classification problem. Considering the inherent relationship between gender and age, the method learns gender as a one-dimensional splice into age codes.
Methods based on popular learning. [4] [5] apply age manifold learning techniques on GEI to find low-dimensional representations (OPLDA, OPMFA) that capture the intrinsic data distribution and geometry of GEI.
Regardless of the methods based on regression, classification, or learning prevalence, the order relationship between ages is not considered, and thus such methods are not very effective. Moreover, the existing methods separate feature extraction and model training, and no deep learning-based end-to-end method exists.
Disclosure of Invention
The invention aims to provide a gait age estimation method with high gait age prediction accuracy.
The gait age estimation method provided by the invention is based on a deep learning and sequence distribution regression method, is called a gait age estimation method based on deep sequence distribution regression and is recorded as ODR-GLCNN; the method comprises the following specific steps:
(1) Gait video sequence preprocessing
Firstly, pedestrian detection and alignment are carried out, then segmentation of a target and a background is carried out, then a preprocessed video sequence is shot into a gait energy image, the gait energy image is marked as GEI, and the GEI is segmented into three parts, namely a head part, a body part and a lower body part, which are not overlapped;
prior to entering the network, the GEI image is adjusted to a 128 x 88 uniform size; the sizes of the head, body and lower body to be divided and not overlapped are (22 × 88), (48 × 88) and (58 × 88), respectively.
(2) Data partitioning for sequence regression methods
For the ordered K classification problem, the order regression method generally converts the K classification problem into a series of two classification problems, namely training K-1 two classifiers;
for the kth classifier, the training set D is subdivided into two subsets, with the sample age label greater than or equal to r k Is a positive sample, less than r k Is a negative sample;
where r is k Is the grade division parameter of the kth classifier, and the total number of the grade divisions is K-1; in gait age estimation, we divide the age from 2 to 90 years into 88 divisions; specifically, r 1 ＝3,r 2 ＝4,r 3 ＝5,…,,r 87 ＝89,r 88 =90; after training in K-1 classifiers, a test sample x is given i The corresponding predicted age is calculated as follows:
wherein f is k (x i ) Is the output of the kth second classification, ranging from (0, 1), [.]The true and false operation is shown, the internal condition is 1 if the internal condition is satisfied, and otherwise, the internal condition is 0.
(3) Feature extraction using global and local convolutional neural networks
Extracting global gait features and three local gait features (namely the gait features of the head, the upper body and the lower body) by using a global network and three sub-networks respectively;
specifically, as in the network structure shown in fig. 1, a GEI picture with a pixel size of 128 × 88 × 1 is used as an input of the global network; then, the GEI is cut into a head part, an upper body part and a lower body part with the pixel sizes of (22 × 88), (48 × 88) and (58 × 88), which are used as the input of three sub-networks;
here, the global network is two layers, and the sizes of convolution kernels are (7, 7), (7, 7); the sub-office network has two layers, and the convolution kernels of the three sub-networks have the same size, namely (7, 7) and (5, 7); after the three sub-networks are convoluted by two layers, splicing the three feature maps in a high dimension, and then splicing the three feature maps with the feature map output by the global network in a channel dimension; then, a roll base layer and three full connection layers are connected, and finally, K-1 dimensional binary classification results are output; wherein the convolution kernel size of the last roll base layer is (5,5); the dimensionality of the first two fully-connected layers is 1024, and the output of the last fully-connected layer is K-1.
(4) Training network models using order distribution loss functions
In the design of the loss function, not only the cross entropy loss but also the distribution loss are considered;
(a) The Cross Entropy Loss function (Cross-Encopy Loss) is:
where N is the number of samples, K is the number of categories,
the true value of the kth secondary class representing the ith sample,
is the corresponding predicted value;
(b) The EMD distribution Loss function (Earth Mover's Distance Loss) is:
wherein CDF is a cumulative distribution function;
(c) The overall loss function is:
The gait age estimation method provided by the invention is based on end-to-end sequence distribution regression of global and local convolutional neural networks. Two problems are mainly solved: (1) A global network is used for learning the global characteristics of gait GEI, and three sub-networks are used for respectively learning the local characteristics of the head part, the body part and the lower body part of the gait, so that more representative characteristics can be learned; (2) On the basis of sequence regression, considering the distribution loss between two output classes, thereby providing a sequence distribution regression method: not only the sequence relationship between gait ages but also the distribution relationship between sequences are considered. Therefore, the invention can better optimize the network and greatly improve the prediction performance of the model.
Drawings
FIG. 1: the invention discloses an ODR-GLCNN network structure.
FIG. 2: OULP-Age data set sample presentation.
FIG. 3: statistical distribution of Age and gender of the OULP-Age dataset.
FIG. 4: CS comparison of the present method with other methods on the OULP-Age dataset.
FIG. 5: gait age prediction sample presentation (inside brackets are prediction results).
Detailed Description
Having described the algorithmic principles and the specific steps of the present invention, the following demonstrates the test effect of the present invention on the current largest gait Age estimation data set (OULP-Age).
OULP-Age is a gray-scale gait energy map (GEI) dataset consisting of 63846 images of 128 x 88 pixels of GEI, with male and female specimens 31093 and 32753, respectively, spanning ages 2 to 90. Fig. 2 shows a GEI example image of some males and females of the OULP-Age dataset. Fig. 3 shows the statistical distribution of the data set samples over age and gender. According to a large number of previous experimental settings, half of the data set is used as a training set, the other half is used as a test set, and the age and gender distribution of the test set in the training set are basically similar.
The experiment used the mean absolute error MAE (mean absolute error) and the cumulative score CS (cumulative score) as indicators of the performance of age estimation. The MAE is defined by the formula:
wherein N represents the number of test samples, y i And y i ' indicates the real age and the predicted age of the ith sample, respectively. The definition formula of CS is:
wherein N is k The number of samples with absolute value error between the predicted value and the true value smaller than k, and N is the total number of test samples.
MAE represents the accuracy of the algorithm, and CS represents the stability and generalization of the algorithm.
Experimental example 1: performance comparison of algorithms on OULP-Age datasets
This experiment compared the Age estimation performance of the different models on the OULP-Age gait dataset. We compared existing methods such as SVR, MLG, GPR, OPLDA, OPMFA and ASSOLPP. Meanwhile, in order to embody the effectiveness of the method of the invention, several conventional deep learning methods are compared, such as CNN + Euclidean, CNN + Cross Engine Loss, VGG16+ Mean-Variance Loss. Table 1 and fig. 4 show the results of a comparison of the method of the present invention and other methods on an OULP-Age dataset. It can be seen that the method of the present invention achieves the best performance in both the MAE and CS indices.
Table 1: performance comparison of algorithm in OULP-Age dataset with other methods
| Method | MAE | CS(k＝5) |
| SVR | 7.66 | 41.40％ |
| MLG | 10.98 | 43.40％ |
| OPLDA | 8.45 | 36.50％ |
| OPMFA | 9.08 | 34.70％ |
| GPR | 7.30 | 43.60％ |
| ASSOLPP | 6.78 | 53.00％ |
| CNN+Euclidean | 6.96 | 51.13％ |
| CNN+Cross-Entropy | 5.40 | 65.64％ |
| VGG16+Mean-Variance | 5.59 | 60.46％ |
| ODR-GLCNN(Ours) | 5.13 | 66.95％ |
。
Experimental example 2: performance comparison of GL-CNN feature extraction network provided by the invention with common CNN and VGG16 methods on OULP-Age data sets
Table 2 shows the comparison results of using different network extraction features, where the loss functions are all the order distribution losses (ordering distribution loss) extracted by using the method. It can be seen that the global and local convolutional neural networks (GL-CNN) proposed by the present invention perform better in gait age estimation than the simple CNN and the more numerous VGGs 16 used. While the method of the present invention is longer in test time than simple CNN, it is much faster than VGG 16.
Table 2: the GL-CNN method proposed in the present invention is compared with other methods
| Method | MAE | CS(k＝5) | Time(ms) |
| CNN | 5.35 | 65.66％ | 7.27× |
| VGG16 | 5.74 | 63.53％ | 21.9× |
| GL-CNN(Ours) | 5.13 | 66.95％ | 8.99× |
。
Experimental example 3: comparison of the proposed order Distribution Loss function (original Distribution Loss) with the traditional Euclidean Loss and cross entropy Loss method on the data set OULP-Age data set
Table 3 shows the effect of different loss functions on model performance. In this experiment, the order distribution loss (order distribution loss) proposed by the method is compared with the commonly used Euclidean loss (Euclidean loss) and cross-entropy loss (cross-entropy loss), and in order to compare the performances of the three loss functions, the feature extraction network used here is GL-CNN. The results in Table 3 show that the sequence distribution loss performance proposed by this method is better than that of the other two methods. The cross entropy loss is better than the Euclidean distance loss performance mainly because the cross entropy method considers the order relation among age labels, and the order distribution loss performance proposed by the method is better than the cross entropy loss because the method considers the order relation among the age labels and also considers the distribution relation among the two values output by the network. Thus, the performance is better than the other two.
Table 3: order distribution penalty presented in this invention is compared to other methods
| Method | MAE | CS(k＝5) | CS(k＝10) |
| Euclidean Loss | 6.73 | 52.95％ | 79.67％ |
| Cross-Entropy Loss | 5.24 | 65.96％ | 84.57％ |
| Ordinal Distribution Loss(Ours) | 5.13 | 66.95％ | 85.01％ |
。
Claims (3)
1. A gait age estimation method based on depth order distribution regression is characterized by comprising the following specific steps:
(1) Gait video sequence preprocessing
Firstly, pedestrian detection and alignment are carried out, then segmentation of a target and a background is carried out, and then a preprocessed video sequence is shot into a gait energy image which is marked as GEI; dividing the GEI into three parts of a head part, an upper body part and a lower body part which are not overlapped;
(2) Data partitioning for sequence regression methods
For the ordered K classification problem, converting the K classification problem into a series of two classification problems, namely training K-1 two classifiers;
(3) Feature extraction using global and local convolutional neural networks
Respectively extracting global gait features and three local gait features by using a global network and three sub-networks, wherein the three local gait features are the gait features of the head, the upper body and the lower body;
(4) Training network models using order distribution loss functions
The design of the loss function takes into account both cross entropy loss and distribution loss:
(a) The cross entropy loss function is:
where N is the number of samples, K is the number of categories,
the true value of the kth secondary class representing the ith sample,
is the corresponding predicted value;
(b) The EMD distribution loss function is:
wherein CDF is a cumulative distribution function;
(c) The overall loss function is:
in step (2), for the kth classifier, the training set D is subdivided into two subsets: sample age label greater than or equal to r k Is a positive sample, less than r k Is a negative sample, i.e.
Where r is k Is the grade division parameter of the kth classifier, and the total number of the grade divisions is K-1; after K-1 classifiers are trained, the test is givenTest sample x i The corresponding predicted age is calculated as follows:
wherein f is k (x i ) Is the output of the kth second classification, ranging from (0, 1), [.]The true and false operation is shown, the internal condition is 1 if the internal condition is satisfied, and otherwise, the internal condition is 0.
2. The gait age estimation method based on depth-order distribution regression according to claim 1, characterized in that in step (1), before being input into the network, the GEI images are adjusted to 128 x 88 uniform size; the sizes of the divided non-overlapping head, upper body, and lower body images are (22 × 88), (48 × 88), and (58 × 88), respectively.
3. The gait age estimation method based on depth-order distribution regression as claimed in claim 1, wherein in step (3), the process of extracting features by the global and local convolutional neural networks is as follows:
the GEI picture with the pixel size of 128 multiplied by 88 multiplied by 1 is used as the input of the global network; cutting the GEI into a head part, an upper body part and a lower body part with the pixel sizes of (22 × 88), (48 × 88) and (58 × 88) respectively as the input of three sub-networks;
wherein, the global feature network has two layers, and the sizes of convolution kernels are (7, 7) and (7, 7); the three sub-networks have two layers, and the convolution kernels of the three sub-networks have the same size and are respectively (7, 7), (5, 7); after the three sub-networks are convoluted by two layers, splicing the three feature maps in a high dimension, and then splicing the three feature maps with the feature map output by the global network in a channel dimension; then, a roll base layer and three full connection layers are connected, and finally, K-1 dimensional binary classification results are output; the convolution kernel size of the last layer of the volume base layer is (5, 5), the dimensionality of the first two full-connected layers is 1024, and the output of the last layer of the full-connected layer is K-1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811255447.8A CN109543546B (en) | 2018-10-26 | 2018-10-26 | Gait age estimation method based on depth sequence distribution regression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811255447.8A CN109543546B (en) | 2018-10-26 | 2018-10-26 | Gait age estimation method based on depth sequence distribution regression |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109543546A CN109543546A (en) | 2019-03-29 |
| CN109543546B true CN109543546B (en) | 2022-12-20 |
Family
ID=65845120
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811255447.8A Active CN109543546B (en) | 2018-10-26 | 2018-10-26 | Gait age estimation method based on depth sequence distribution regression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109543546B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112149449A (en) * | 2019-06-26 | 2020-12-29 | 北京华捷艾米科技有限公司 | Face attribute recognition method and system based on deep learning |
| CN112329716A (en) * | 2020-11-26 | 2021-02-05 | 重庆能源职业学院 | Pedestrian age group identification method based on gait characteristics |
| CN114882593A (en) * | 2022-05-18 | 2022-08-09 | 厦门市美亚柏科信息股份有限公司 | Robust space-time mixed gait feature learning method and system |
| CN115731517B (en) * | 2022-11-22 | 2024-02-20 | 南京邮电大学 | Crowded Crowd detection method based on crown-RetinaNet network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003204953A (en) * | 2002-01-16 | 2003-07-22 | Nippon Telegr & Teleph Corp <Ntt> | Gait pattern classification display |
| CN103294982A (en) * | 2012-02-24 | 2013-09-11 | 北京明日时尚信息技术有限公司 | Method and system for figure detection, body part positioning, age estimation and gender identification in picture of network |
| CN105975916A (en) * | 2016-04-28 | 2016-09-28 | 西安电子科技大学 | Age estimation method based on multi-output convolution neural network and ordered regression |
| GB201614425D0 (en) * | 2016-08-24 | 2016-10-05 | Fujitsu Ltd | Data collector |
| CN107085716A (en) * | 2017-05-24 | 2017-08-22 | 复旦大学 | Across the visual angle gait recognition method of confrontation network is generated based on multitask |
| CN107330412A (en) * | 2017-07-06 | 2017-11-07 | 湖北科技学院 | A kind of face age estimation method based on depth rarefaction representation |
| CN108171209A (en) * | 2018-01-18 | 2018-06-15 | 中科视拓（北京）科技有限公司 | A kind of face age estimation method that metric learning is carried out based on convolutional neural networks |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102982323B (en) * | 2012-12-19 | 2016-05-11 | 重庆信科设计有限公司 | Gait recognition method fast |
| US10448867B2 (en) * | 2014-09-05 | 2019-10-22 | Vision Service Plan | Wearable gait monitoring apparatus, systems, and related methods |
| CN106250818B (en) * | 2016-07-19 | 2019-03-26 | 南京大学 | A kind of total order keeps the face age estimation method of projection |
| JP6649323B2 (en) * | 2016-10-27 | 2020-02-19 | 株式会社アシックス | Gait analysis system and method |
| CN108573209A (en) * | 2018-02-28 | 2018-09-25 | 天眼智通(香港)有限公司 | A kind of age-sex's recognition methods of the single model multi output based on face and system |
-
2018
- 2018-10-26 CN CN201811255447.8A patent/CN109543546B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003204953A (en) * | 2002-01-16 | 2003-07-22 | Nippon Telegr & Teleph Corp <Ntt> | Gait pattern classification display |
| CN103294982A (en) * | 2012-02-24 | 2013-09-11 | 北京明日时尚信息技术有限公司 | Method and system for figure detection, body part positioning, age estimation and gender identification in picture of network |
| CN105975916A (en) * | 2016-04-28 | 2016-09-28 | 西安电子科技大学 | Age estimation method based on multi-output convolution neural network and ordered regression |
| GB201614425D0 (en) * | 2016-08-24 | 2016-10-05 | Fujitsu Ltd | Data collector |
| CN107085716A (en) * | 2017-05-24 | 2017-08-22 | 复旦大学 | Across the visual angle gait recognition method of confrontation network is generated based on multitask |
| CN107330412A (en) * | 2017-07-06 | 2017-11-07 | 湖北科技学院 | A kind of face age estimation method based on depth rarefaction representation |
| CN108171209A (en) * | 2018-01-18 | 2018-06-15 | 中科视拓（北京）科技有限公司 | A kind of face age estimation method that metric learning is carried out based on convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| 分而冶之的人脸图像年龄估计;刘玉坤等;《计算机工程与设计》;20171116(第11期);全文 * |
| 基于卷积神经网络的人脸年龄估计算法;周旺;《中国优秀硕士学位论文全文数据库 信息科技辑》;20170815;全文 * |
| 基于稀疏表示的自动年龄估计;李玲芝等;《中南大学学报(自然科学版)》;20150326(第03期);全文 * |
| 基于过程神经网络的步态模式自动分类;王斐等;《东北大学学报(自然科学版)》;20110415(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109543546A (en) | 2019-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109543546B (en) | Gait age estimation method based on depth sequence distribution regression | |
| CN108875608B (en) | Motor vehicle traffic signal identification method based on deep learning | |
| Lu et al. | Object detection based on SSD-ResNet | |
| CN110321967B (en) | Image classification improvement method based on convolutional neural network | |
| CN111274922B (en) | Pedestrian re-identification method and system based on multi-level deep learning network | |
| CN107330397B (en) | Pedestrian re-identification method based on large-interval relative distance measurement learning | |
| CN111738143B (en) | Pedestrian re-identification method based on expectation maximization | |
| CN105528595A (en) | Method for identifying and positioning power transmission line insulators in unmanned aerial vehicle aerial images | |
| CN107066559A (en) | A kind of method for searching three-dimension model based on deep learning | |
| CN104408469A (en) | Firework identification method and firework identification system based on deep learning of image | |
| CN104268593A (en) | Multiple-sparse-representation face recognition method for solving small sample size problem | |
| Champ et al. | A comparative study of fine-grained classification methods in the context of the LifeCLEF plant identification challenge 2015 | |
| CN105184772A (en) | Adaptive color image segmentation method based on super pixels | |
| Zhang et al. | Multi-features integration based hyperspectral videos tracker | |
| CN109165698A (en) | A kind of image classification recognition methods and its storage medium towards wisdom traffic | |
| CN110880010A (en) | Visual SLAM closed loop detection algorithm based on convolutional neural network | |
| CN112784754A (en) | Vehicle re-identification method, device, equipment and storage medium | |
| Lodh et al. | Flower recognition system based on color and GIST features | |
| Ticay-Rivas et al. | Pollen classification based on geometrical, descriptors and colour features using decorrelation stretching method | |
| Hrkać et al. | Deep learning architectures for tattoo detection and de-identification | |
| CN108154172A (en) | Image-recognizing method based on three decisions | |
| Kale et al. | Age, gender and ethnicity classification from face images with CNN-based features | |
| CN112766180B (en) | Pedestrian re-identification method based on feature fusion and multi-core learning | |
| Sun et al. | Deep learning based pedestrian detection | |
| Hrkac et al. | Tattoo detection for soft biometric de-identification based on convolutional neural networks |
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 |