CN104866825B - A kind of sign language video frame sequence classification method based on Hu square - Google Patents
A kind of sign language video frame sequence classification method based on Hu square Download PDFInfo
- Publication number
- CN104866825B CN104866825B CN201510254121.3A CN201510254121A CN104866825B CN 104866825 B CN104866825 B CN 104866825B CN 201510254121 A CN201510254121 A CN 201510254121A CN 104866825 B CN104866825 B CN 104866825B
- Authority
- CN
- China
- Prior art keywords
- frame
- image
- sign language
- label
- sequence
- 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
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
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
Abstract
The present invention discloses a kind of sign language video frame sequence classification method based on Hu square, comprising the following steps: step 1: obtaining sign language video to be identified；Step 2: frame sampling being carried out to the sign language video, obtains frame sequence；Step 3: the frame sequence of color image is converted to the frame sequence of bianry image；Step 4: by hand and background segment；Step 5: the Hu square of the hand images after extracting segmentation obtains the feature vector of hand images；Step 6: calculating the Euclidean distance between each frame and the feature vector of former frame, adjudicate, paste digital label, carry out preliminary classification.Step 7: frame sequence being carried out according to label and is further classified, transitional movement frame, the sorted sequence label of Sequential output are removed.Computation complexity of the present invention is lower, and has good robustness for the rotation of sign language image, zooming and panning, may be directly applied to sign language video identifying system.
Description
Technical field
The present invention relates to the field of video processing of computer vision, more particularly to a kind of sign language video frame based on Hu square
Sequence classification method
Background technique
Influence with computer to modern society is growing, and human-computer interaction is just becoming in people's daily life increasingly
Important a part.Most of man-machine interaction mode of today is the input tool relatively simple by keyboard and mouse etc., is made
It is then fast and easily more compared with them with the input mode of Body Languages.Sign Language Recognition is as in field of intelligent man-machine interaction
An important component, it is the purpose is to provide a kind of efficiently and accurately mechanism by computer, sign language interpreter is written
Sheet or voice.This technology not only makes the exchange between deaf person and Subjects With Normal Hearing become more convenient, but also in human-computer interaction
Also there are positive meaning and application value in field.
Traditional Sign Language Recognition includes the Sign Language Recognition of Sign Language Recognition and view-based access control model based on data glove.Wherein, base
In the Sign Language Recognition of data glove, user needs to wear data glove, and computer obtains the movement of human hand by data glove
Information is simultaneously handled in real time, and by fructufy when shows to realize human-computer interaction.And the Sign Language Recognition of view-based access control model is then
The hand images that people is obtained by digital photographing apparatus, are further handled the image of acquisition, and then identify corresponding hand
Language information.Generally speaking, the advantages of method based on data glove is that input data is few and resolution height, but data glove is set
Standby expensive and wearing is thick and heavy, is unfavorable for natural human-computer interaction.And method based on computer vision then may be implemented more certainly
Right human-computer interaction, while price is lower, input cost is small, but this method is primarily present following two large problems:
(1) Hand Gesture Segmentation in the case where monocular vision under complex background is very difficult, since background is various each
Sample, environmental factor is unpredictable, and not only unmature theoretical be used as is instructed, but also the realization of existing method is difficult, calculates again
Miscellaneous degree is high, and effect is also not highly desirable.Common solution is to increase limitation at present, such as using black or the wall of white,
Dark clothes etc. simplify background, or the gloves etc. for requiring manpower to wear special color emphasize prospect, to simplify hand region and background
The division in region.
(2) during the progress of sign language, hand can irregular generation during sign language is realized as non-rigid object
Deformation, therefore, a critical issue is just how to carry out the classification of sequence of frames of video, identification in the identification of dynamic sign language video
The semanteme of the dynamic sign language sequence out.Conventional track algorithm tracks extremely easy loss in the motion profile of track record hand
Target, and be difficult to give for change again.
Geometric moment was by Hu (Visual pattern recognition by moment invariants) in 1962
It is proposed that there is translation, rotation and scale invariability.Hu using second order and third central moment construct seven not bending moment M1~
M7, they can keep translation, scaling and invariable rotary under the conditions of row graph, and related definition is also used by the present invention.It is real
On border, in picture in the identification process of object, only M1 and M2 invariance is kept relatively good, other several not bending moments
Bring application condition is big.
Summary of the invention
In view of the foregoing, it is necessary to a kind of correct recognition rata that can be improved sign language video be provided, there is preferable Shandong
Stick is effectively improved the classification method of the sign language video frame sequence of sign language video recognition effect.
To achieve the above object, the present invention provides a kind of sign language video frame sequence classification method based on Hu square, including with
Lower step:
Step 1: obtaining colored sign language video to be identified；
Step 2: frame sampling being carried out to the colour sign language video, obtains frame sequence；
Step 3: the frame sequence of color image is converted to the frame sequence of bianry image；
Step 4: by hand and background segment；
Step 5: M1 the and M2 square in the Hu square of the hand images sequence after extracting segmentation obtain the features of hand images to
Amount；
Further, bending moment M1 and M2 are not defined as follows:
M1=η20+η02
Wherein,
N and M is the height and width of image respectively, and f (x, y) is image function,Representative image center of gravity, definition
Are as follows:
Wherein,
Step 6: calculating the Euclidean distance between each frame hand images and the feature vector of former frame hand images
Sample_dis, and adaptive threshold judgement is carried out, digital label is pasted, preliminary classification is carried out；
Step 7: the classification of frame sequence essence being carried out according to label, removes transition frames, the sorted sequence label of Sequential output.
Further, image is mapped to characteristic vector space from pixel space with M1 and M2 square in the step 5, reduced
Operation complexity:
Wherein, fi(x, y) is the image function of the i-th frame,For corresponding feature vector.
Further, the Euclid of two frame image features vectors is adjudicated in the step 6 with adaptive threshold thre
Measurement carries out preliminary classification:
ρ function is defined as follows:
For the feature vector of the i-th frame image, LiFor the digital label of the i-th frame image, if L1=1；
Under the premise of the sampling interval that sample rate meets front cross frame is sufficiently small, the hand done in front cross frame can be approximatively thought
Language is all of a sort sign language, therefore can be setRealize that the stronger adaptive threshold of robustness determines.
Further, include: step 71 in the step 7, obtain the label of new frame sampling image；Step 72, pass through
The label of step 71 judges whether the frame is transition gesture motion；If so, the frame image is removed；If it is not, by the frame with it is same
The frame of label is classified as one kind；Step 73, according to frame sequence Sequential output classification results, classification results are indicated with digital label, together
The frame sequence of one digital label indicates the video clip sample frame set that they are same sign language movement；
Further, step 7 removes transition sign language frame using digital label, specifically includes:
Assuming that a total of N frame image of target video sequence, if to the digital label L of the i-th frame image (1 < i < N)iMeet:
Li=Li-1+1&&Li=Li+1- 1, then the frame is transition gesture motion, needs to remove the frame, it may be assumed that for arbitrary kth frame image,
As k >=i, L is enabledk=Lk-1；Wherein, LkFor the digital label of kth frame image.
Further, the step 3 is into the following steps are included: step 31, the frame sampling image that obtaining step 2 obtains；Step
Rapid 32, all pixels of the image are traversed, judge whether it may be the pixel of hand region, if so, being set to white
Color, if it is not, being set to black；Step 33, bianry image is exported.
Further, step 4 is the following steps are included: step 41, the frame bianry image that obtaining step 3 obtains；Time step 42,
All profiles of the image are gone through, judge whether it is the profile of hand region, if so, continuing step 43, if it is not, repeating step 42
Until traversal terminates to exit；Step 43, by the hand region and background segment；Step 44, the hand region image is exported.
Preferably, the sample rate Sample_rate in the step 2 is to take a frame every 1 second.
Compared with prior art, the invention has the advantages that and technical effect:
A kind of sign language video frame sequence classification method based on Hu square provided by the present invention, computation complexity is lower, and
The non-intrinsically safe change of the images such as rotation, the zooming and panning of sign language image caused by inevitable for external environment difference has
Good robustness.The present invention provides how to remove transition gesture motion in traditional frame sequence to influence this to sign language video identification
The simple solution of one problem.The present invention may be directly applied to sign language video identifying system, by providing sign language frame sequence
Sequence digital label, the sign language frame of same label are same class, as long as therefore taking a frame, progress letter from every one kind in order
Single static two-value sign language image recognition, can be completed the semantics recognition of sign language video.
Detailed description of the invention
Fig. 1 is the overview flow chart of the sign language video frame sequence classification of the invention based on Hu square；
Fig. 2 is the sub-process figure of step S103 in Fig. 1；
Fig. 3 is the sub-process figure of step S104 in Fig. 1；
Fig. 4 is the sub-process figure of step S107 in Fig. 1.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is described in detail.It should be appreciated that described herein, specific examples are only used to explain the present invention, is not used to limit
The fixed present invention, if it is existing to be that those skilled in the art can refer in addition, have the process not being described in detail especially or symbol below
Technology realize or understand.
As shown in Figure 1, being the flow chart of sign language video frame sequence classification method preferred embodiment of the present invention.Step S101,
Sign language video to be identified is obtained using digital photographing apparatus.In the present embodiment, it is contemplated that utilize current computer vision
Technology can not also directly extract careful free hand gestures characteristic information from two-dimensional space, in order to describe gesture feature subtly with area
Divide similar gesture word, using the calibration and detection at a secondary blue gloves supplemental characteristic position, this is not limiting upon interactive nature
Property.Digital photographing apparatus used by the present embodiment is the DSC-W730 digital camera of Sony company production, and valid pixel is
16100000, sampling rate is 29.97 frames/s, and the camera is to be located to do sign language person front and shoot forward when video recording.
Step S102 is that sample rate Sample_rate carries out frame sampling to the sign language video.In the present embodiment,
Sample_rate is set as taking 1 frame every 30 frames, i.e., about 1s acquires 1 frame.
Step S103 is converted to black and white binary image sequence as preprocessing part, by collected color image sequence,
As shown in Fig. 2, the step S103 further includes steps of step S201: obtaining new frame sampling image；Step
S202: traversing all pixels of the image, judges whether it may be the pixel of hand region；If so, being set to white
Color；If it is not, being set to black；Step S203: output bianry image.Specifically, the basic principle of binaryzation is to pass through gloves
Color the part of hand is set as white, remaining background is black, and specific decision criteria is as follows:
1. the rgb value of pixel pixel meets: B=Max (R, G, B)
2. the B value of pixel pixel meets: B > B_thre, preferably, B_thre=50
3. the RGB variance Var of pixel pixel meets: Var > Var_thre, preferably, Var_thre=350
Wherein,
Criterion 1 is because gloves are blue, and B value is the maximum value of rgb value；Criterion 2 is because meeting the picture of criterion 1
Vegetarian refreshments is not necessarily blue, and B value needs to be more than that certain threshold value just can be determined that as blue；Criterion 3 is because region is bright
Aobvious blue, so the variance of its rgb value can be more than certain threshold value.
The pixel pixel for only meeting above-mentioned 3 criterion simultaneously just can be determined that pixel for hand region, by it
Rgb value is all assigned a value of 255 (i.e. white), otherwise, its rgb value is all assigned a value of 0 (i.e. black).
It when there is blue object in shooting background, can also be binarized as white area, therefore, in bianry image not
All white areas are all hand region, and step S104 distinguishes hand region and other white areas, by hand and back
Scape segmentation.As shown in figure 3, the step S104 further includes steps of step S301: obtaining new frame bianry image；
Step S302: traversing all profiles of the image, judge its whether be hand region profile；If so, continuing step S303；If
It is no, step S302 is repeated until traversal terminates to exit；Step S303: by the hand region and background segment；Step S304: output
The hand region image.Specifically, the basic principle of segmentation is identified using the size of the minimum rectangle of enclosing region profile
Hand region out, specific decision criteria are as follows:
1. the length for surrounding the minimum rectangle of profile Contour meets: len_min < length < len_max, preferably
Ground, len_min=150, len_max=250
2. the width for surrounding the minimum rectangle of profile Contour meets: hgh_min < height < hgh_max, preferably
Ground, hgh_min=100, hgh_max=200
Only meet criterion 1 simultaneously and when criterion 2 just can be determined that as hand region, and with its minimum rectangle general of encirclement
The region segmentation comes out, the bianry image after output hand and background segment.
Since the data volume that image is included is very big, in order to effectively further analyze these image datas, such as Fig. 1 institute
Show, in step s105, with the opponent of the moment invariants Hu square with constant rate, translation invariance and rotational invariance
Portion's bianry image is described, and by the analysis of analysis one group of feature vector value of conversion of bianry image sequence, reduces answering for operation
Miscellaneous degree.In fact, the 7 invariant moments of Hu only have the ratio of M1 and M2 invariance holding in picture in the identification process of object
Preferably, other several not bending moment bring application conditions are big.Therefore only with M1 and M2 to the two-value hand after segmentation in S105
Portion's image is described.Specifically, bending moment M1 and M2 are not defined as follows:
M1=η20+η02
Wherein,
F (x, y) is image function,Representative image center of gravity, is defined as:
Wherein,
With not bending moment M1 and M2, the two-value hand images after each segmentation can be mapped as following feature vector,
Realize the reduction of computation complexity:
After image is mapped to characteristic vector space from pixel space, difference between each image can with they
Difference between feature vector is measured.And the difference between feature vector can be by two vectors in feature space
Distance characterizes.As shown in Figure 1, in step s 106, indicated with euclidean metric two vectors in feature space away from
From, be defined as follows shown in:
Assuming that there are two image Hu Character eigenvectorsThen their Euclid's degree
Amount are as follows:
After obtaining the euclidean metric between the feature vector of two images, need to set an adaptive threshold value,
To judge whether they belong to same class.If same class, identical label is sticked；If not same class, sticks different marks
Label.In fact, assume that the sign language done in the 1st second is all of a sort sign language completely, therefore threshold value takes front cross frame image
The euclidean metric of feature vector.Specific decision criteria is as follows:
Wherein,
For the feature vector of the i-th frame image, LiFor the label of the i-th frame image, label is represented with number in the present embodiment.
By step S106 available preliminary classification results.However, traditional frame sequence classification problem
One difficult point is: those frames excessively acted how are removed from collected frame sequence.It will be according to image in step S107
Label be that frame sequence carries out smart classification.Specifically, as shown in figure 4, the step S107 further includes steps of step
S401: the label of new frame sampling image is obtained；Step S402: judge whether the frame is transition gesture motion by the label；
If so, removing it；If it is not, the frame of itself and same label is classified as one kind.Step S403: classify according to frame sequence Sequential output
As a result.The specific decision criteria of the step S402 is as follows:
Assuming that a total of N frame image of target video sequence, if to the digital label L of the i-th frame image (1 < i < N)iMeet:
Li=Li-1+1&&Li=Li+1- 1, then the frame is transition gesture motion, needs to remove the frame, it may be assumed that for arbitrary kth frame image,
As k >=i, L is enabledk=Lk-1；Wherein, LkFor the digital label of kth frame image.If not satisfied, returning to step S401.
Finally, according to frame sequence Sequential output essence classification results.Classification results indicate with digital label, same digital label
Frame sequence indicate they be same sign language movement video clip sample frame set.
Claims (1)
1. a kind of sign language video frame sequence classification method based on Hu square, which comprises the following steps:
Step S101 obtains sign language video to be identified using digital photographing apparatus；Using a secondary blue gloves supplemental characteristic portion
The calibration and detection of position, this is not limiting upon interactive naturality；
Step S102 is that sample rate SAMPLE_RATE carries out frame sampling to the sign language video；SAMPLE_RATE is set as every 30
Frame takes 1 frame, i.e. 1S acquires 1 frame；
Step S103 is converted to black and white binary image sequence as preprocessing part, by collected color image sequence；It is described
Step S103 further includes steps of step S201: obtaining new frame sampling image；Step S202: the image institute is traversed
Some pixels judge whether it may be the pixel of hand region；If so, being set to white；If it is not, being set to black
Color；Step S203: output bianry image；Specifically, the basic principle of binaryzation is to be set the part of hand by the color of gloves
For white, remaining background is black, and specific decision criteria is as follows:
1. the rgb value of pixel pixel meets: B=Max (R, G, B)
2. the B value of pixel pixel meets: B > B_thre,
3. the RGB variance Var of pixel pixel meets: Var > Var_thre, wherein
Criterion 1 is because gloves are blue, and B value is the maximum value of rgb value；Criterion 2 is because meeting the pixel of criterion 1
It is not necessarily blue, B value needs to be more than that certain threshold value just can be determined that as blue；Criterion 3 is because region is apparent
Blue, so the variance of its rgb value can be more than certain threshold value；
The pixel pixel for only meeting above-mentioned 3 criterion simultaneously is just determined as the pixel of hand region, its rgb value is complete
Portion is assigned a value of 255, otherwise, its rgb value is all assigned a value of 0 i.e. black；
When there is blue object in shooting background, can also be binarized as white area, it is therefore, in bianry image and not all
White area be all hand region, step S104 distinguishes hand region and other white areas, by hand and background point
It cuts；The step S104 further includes steps of step S301: obtaining new frame bianry image；Step S302: traversal should
All profiles of image, judge its whether be hand region profile；If so, continuing step S303；If it is not, repeating step S302
Until traversal terminates to exit；Step S303: by the hand region and background segment；Step S304: the hand region image is exported；
Specifically, the basic principle of segmentation is that hand region is identified using the size of the minimum rectangle of enclosing region profile, specifically
Decision criteria is as follows:
1. the length for surrounding the minimum rectangle of profile Contour meets: len_min < length < len_max,
2. the width for surrounding the minimum rectangle of profile Contour meets: hgh_min < height < hgh_max,
Only meet criterion 1 simultaneously and when criterion 2 just can be determined that as hand region, and with its minimum rectangle of encirclement by the area
Regional partition comes out, the bianry image after output hand and background segment；
Since the data volume that image is included is very big, in order to effectively further analyze these image datas, in step s105,
With being retouched to hand bianry image for the moment invariants Hu square with constant rate, translation invariance and rotational invariance
It states, by the analysis of analysis one group of feature vector value of conversion of bianry image sequence, reduces the complexity of operation；In fact, right
In picture in the identification process of object, the 7 invariant moments of Hu only have M1 and M2 invariance to keep relatively good, and others are several not
Bending moment bring application condition is big；Therefore only the two-value hand images after segmentation are described with M1 and M2 in S105；Tool
Body, bending moment M1 and M2 are not defined as follows:
M1=η20+η02
Wherein,
F (x, y) is image function,Representative image center of gravity, is defined as:
Wherein,
With not bending moment M1 and M2, the two-value hand images after each segmentation are mapped as following feature vector, realizes and calculates
The reduction of complexity:
After image is mapped to characteristic vector space from pixel space, difference between each image can use their feature
Difference between vector is measured；And the difference between feature vector can pass through distance of two vectors in feature space
To characterize；In step s 106, distance of two vectors in feature space is indicated with euclidean metric, is defined as follows institute
Show:
Assuming that there are two image Hu Character eigenvectorsThen their euclidean metric are as follows:
After obtaining the euclidean metric between the feature vector of two images, need to set an adaptive threshold value, to sentence
Breaking, whether they belong to same class；If same class, identical label is sticked；If not same class, sticks different labels；It is real
On border, assume that the sign language done in the 1st second is all of a sort sign language completely, thus threshold value take front cross frame characteristics of image to
The euclidean metric of amount；Specific decision criteria is as follows:
Wherein,
For the feature vector of the i-th frame image, LiFor the label of the i-th frame image, label is represented with number；
A preliminary classification results have been obtained by step S106；However, a difficult point of traditional frame sequence classification problem
It is: how removes the frame of those transitional movements from collected frame sequence；To be according to the label of image in step S107
Frame sequence carries out smart classification；The step S107 further includes steps of step S401: obtaining new frame sampling image
Label；Step S402: judge whether the frame is transition gesture motion by the label；If so, removing it；If it is not, by its with
The frame of same label is classified as one kind；Step S403: according to frame sequence Sequential output classification results；The step S402's specifically sentences
It fixes then as follows:
Assuming that a total of N frame image of target video sequence, if to the digital label L of the i-th frame imageiMeet: Li=Li-1+1&&Li
=Li+1- 1,1<i<N, then the frame is transition gesture motion, needs to remove the frame, it may be assumed that for arbitrary kth frame image, as k>=i
When, enable Lk=Lk-1；Wherein, LkFor the digital label of kth frame image；If not satisfied, returning to step S401；
Finally, according to frame sequence Sequential output essence classification results；Classification results indicate with digital label, the frame of same digital label
Sequence indicates the video clip sample frame set that they are same sign language movement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510254121.3A CN104866825B (en) | 2015-05-17 | 2015-05-17 | A kind of sign language video frame sequence classification method based on Hu square |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510254121.3A CN104866825B (en) | 2015-05-17 | 2015-05-17 | A kind of sign language video frame sequence classification method based on Hu square |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104866825A CN104866825A (en) | 2015-08-26 |
| CN104866825B true CN104866825B (en) | 2019-01-29 |
Family
ID=53912646
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510254121.3A Active CN104866825B (en) | 2015-05-17 | 2015-05-17 | A kind of sign language video frame sequence classification method based on Hu square |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104866825B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113191288A (en) * | 2021-05-09 | 2021-07-30 | 刘明 | Artificial intelligence cloud platform system for sign language communication and intelligent medicine box |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010032733A (en) * | 2008-07-28 | 2010-02-12 | Asutemu:Kk | Finger language image generating system, server, terminal device, information processing method, and program |
| CN101661556A (en) * | 2009-09-25 | 2010-03-03 | 哈尔滨工业大学深圳研究生院 | Static gesture identification method based on vision |
| CN101719271A (en) * | 2009-11-05 | 2010-06-02 | 浙江传媒学院 | Video shot boundary detection method based on mixed projection function and support vector machine |
| CN101853071A (en) * | 2010-05-13 | 2010-10-06 | 重庆大学 | Gesture identification method and system based on visual sense |
| CN102289666A (en) * | 2011-09-02 | 2011-12-21 | 广东中大讯通软件科技有限公司 | Sign language identifying method based on median filter and Hu moment vector |
| CN103559498A (en) * | 2013-09-24 | 2014-02-05 | 北京环境特性研究所 | Rapid man and vehicle target classification method based on multi-feature fusion |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9001036B2 (en) * | 2007-12-20 | 2015-04-07 | University Of Central Florida Research Foundation, Inc. | Systems and methods of camera-based fingertip tracking |
-
2015
- 2015-05-17 CN CN201510254121.3A patent/CN104866825B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010032733A (en) * | 2008-07-28 | 2010-02-12 | Asutemu:Kk | Finger language image generating system, server, terminal device, information processing method, and program |
| CN101661556A (en) * | 2009-09-25 | 2010-03-03 | 哈尔滨工业大学深圳研究生院 | Static gesture identification method based on vision |
| CN101719271A (en) * | 2009-11-05 | 2010-06-02 | 浙江传媒学院 | Video shot boundary detection method based on mixed projection function and support vector machine |
| CN101853071A (en) * | 2010-05-13 | 2010-10-06 | 重庆大学 | Gesture identification method and system based on visual sense |
| CN102289666A (en) * | 2011-09-02 | 2011-12-21 | 广东中大讯通软件科技有限公司 | Sign language identifying method based on median filter and Hu moment vector |
| CN103559498A (en) * | 2013-09-24 | 2014-02-05 | 北京环境特性研究所 | Rapid man and vehicle target classification method based on multi-feature fusion |
Non-Patent Citations (1)
| Title |
|---|
| 基于中值滤波和Hu矩向量的手语识别;华斌等;《计算机工程与设计》;20110216;第32卷(第2期);第1，4节 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104866825A (en) | 2015-08-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108710865B (en) | Driver abnormal behavior detection method based on neural network | |
| CN109344701B (en) | Kinect-based dynamic gesture recognition method | |
| CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
| CN106022231A (en) | Multi-feature-fusion-based technical method for rapid detection of pedestrian | |
| CN111563452B (en) | Multi-human-body gesture detection and state discrimination method based on instance segmentation | |
| CN108647625A (en) | A kind of expression recognition method and device | |
| CN107480585B (en) | Target detection method based on DPM algorithm | |
| CN108960076B (en) | Ear recognition and tracking method based on convolutional neural network | |
| CN109886086B (en) | Pedestrian detection method based on HOG (histogram of oriented gradient) features and linear SVM (support vector machine) cascade classifier | |
| CN113252614B (en) | Transparency detection method based on machine vision | |
| CN107085729B (en) | Bayesian inference-based personnel detection result correction method | |
| Vishwakarma et al. | Simple and intelligent system to recognize the expression of speech-disabled person | |
| CN112101208A (en) | Feature series fusion gesture recognition method and device for elderly people | |
| Alksasbeh et al. | Smart hand gestures recognition using K-NN based algorithm for video annotation purposes | |
| Waqar et al. | Meter digit recognition via Faster R-CNN | |
| CN104866826A (en) | Static gesture language identification method based on KNN algorithm and pixel ratio gradient features | |
| Weerasekera et al. | Robust asl fingerspelling recognition using local binary patterns and geometric features | |
| Agrawal et al. | A Tutor for the hearing impaired (developed using Automatic Gesture Recognition) | |
| CN102663369B (en) | Human motion tracking method on basis of SURF (Speed Up Robust Feature) high efficiency matching kernel | |
| CN104866825B (en) | A kind of sign language video frame sequence classification method based on Hu square | |
| CN103824058A (en) | Face recognition system and method based on locally distributed linear embedding algorithm | |
| Işikdoğan et al. | Automatic recognition of Turkish fingerspelling | |
| Yamamoto et al. | Algorithm optimizations for low-complexity eye tracking | |
| CN109214279A (en) | Online human face expression pre-detection method and device based on video | |
| CN113139946A (en) | Shirt stain positioning device based on vision |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |