CN109190496A - A kind of monocular static gesture identification method based on multi-feature fusion - Google Patents
A kind of monocular static gesture identification method based on multi-feature fusion Download PDFInfo
- Publication number
- CN109190496A CN109190496A CN201810900949.5A CN201810900949A CN109190496A CN 109190496 A CN109190496 A CN 109190496A CN 201810900949 A CN201810900949 A CN 201810900949A CN 109190496 A CN109190496 A CN 109190496A
- Authority
- CN
- China
- Prior art keywords
- gesture
- image
- hand
- value
- 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.)
- Pending
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
- G06V40/107—Static hand or arm
- G06V40/113—Recognition of static hand signs
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (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)
Abstract
The invention discloses a kind of monocular static gesture identification method based on multi-feature fusion, step includes: images of gestures acquisition, includes the RGB image of gesture with monocular cam acquisition；Image preprocessing, skin color segmentation is carried out using human body complexion information, using the geometrical characteristic of Morphological scale-space and combination hand, hand is separated with complex background, the positioning centre of the palm is operated by range conversion and removes arm regions existing for hand, obtains gesture bianry image；Gesture feature extracts, and calculates the perimeter and area ratio, Hu square and Fourier descriptor feature of gesture, constitutes gesture feature vector；Gesture identification, input gesture feature vector training BP neural network, realizes static gesture classification.The geometrical characteristic of present invention combination Skin Color Information and hand realizes accurate Hand Gesture Segmentation under monocular vision using Morphological scale-space and range conversion operation；By combination various gestures feature and training BP neural network, the high gesture classifier of strong robustness, accuracy rate is obtained.
Description
Technical field
The present invention relates to field of image recognition, and in particular to a kind of monocular static gesture based on multi-feature fusion identification side
Method.
Background technique
Gesture as it is a kind of from however intuitive interactive mode, gradually development is the research heat of field of human-computer interaction
Point, and be widely used in terms of somatic sensation television game, robot control, computer.Compared to the hand based on data glove
Gesture identification technology, the Gesture Recognition of view-based access control model has many advantages, such as low for equipment requirements, interaction nature, and becomes gesture and know
Other main way.
Hand Gesture Segmentation is the key link in the gesture identification of view-based access control model, the influential effect of segmentation feature extraction, into
And influence gesture classification result.In the static gesture identification method based on monocular vision, due to the influence of complex background environment,
The result of Hand Gesture Segmentation is not satisfactory.With the appearance of kinect camera, depth information is used for gesture and separates with complex background
Research in.It due to kinect camera higher cost, is not widely used, therefore such gesture identification method is unable to get popularization
Using.The gesture feature that existing monocular static gesture identification method uses is relatively simple, leads to the robust of gesture recognition system
Property is weaker, and recognition accuracy is not high.It is therefore proposed that a kind of can accurately divide under complex background and identify that the monocular of gesture is quiet
State gesture identification method is current letter problem to be solved.
Summary of the invention
The purpose of the present invention is to solve drawbacks described above in the prior art, provide a kind of list based on multi-feature fusion
Mesh static gesture identification method.
The purpose of the present invention can be reached by adopting the following technical scheme that:
A kind of monocular static gesture identification method based on multi-feature fusion, the recognition methods include:
Images of gestures acquisition step includes the RGB image of gesture with monocular cam acquisition；
Image preprocessing step, using human body complexion information carry out skin color segmentation, by image the colour of skin and class colour of skin area
Domain extracts, and using the geometrical characteristic of Morphological scale-space and combination hand, hand is separated with complex background, passes through distance and becomes
It changes the operation positioning centre of the palm and removes arm regions existing for hand, obtain gesture bianry image；
Gesture feature extraction step calculates the perimeter and area ratio, Hu square and Fourier descriptor feature of gesture, constitutes
Gesture feature vector；
Gesture identification step passes through training BP nerve using the gesture feature vector of extraction as the input of BP neural network
The classification of network implementations static gesture.
Further, the image preprocessing step includes:
The image of input is carried out the conversion of color space by skin color segmentation, is divided by Chroma threshold by the skin in image
Color and class area of skin color extract, and obtain bianry image；
Morphological scale-space is carried out Morphological scale-space to the bianry image after skin color segmentation, is first operated using opening operation, then adopt
It is operated with closed operation, eliminates the isolated noise in image；
The segmentation of hand geometry realizes that face and hand are separated with complex background, calculates remaining two connected regions
The shape complexity C in domain extracts hand bianry image by judging with threshold value T；
By the arm removal in hand bianry image, gesture bianry image is obtained.
Further, the image by input carries out the conversion of color space, is divided by Chroma threshold by image
In the colour of skin and class area of skin color extract, it is as follows to obtain bianry image process:
The image of input is transformed into YCr ' Cb ' color space, conversion formula by rgb color space by convert color spaces
It is as follows:
Y=0.299 × r+0.587 × g+0.114 × b
Wherein r, g, b are respectively image three components of the red, green, blue in rgb color space, and y, cr', cb' are respectively
Luminance component, red chrominance component, chroma blue component of the image in YCr ' Cb ' color space；
Chroma threshold segmentation, if two chromatic components of pixel meet the colour of skin in the threshold value model of cr' and cb' component simultaneously
When enclosing, enable the pixel value be 1, otherwise value be 0, thus by image the colour of skin and class area of skin color extract, obtain
To bianry image.
Further, hand geometry segmentation realizes that face and hand are separated with complex background, calculates surplus
Under two connected regions shape complexity C, by with threshold value T judge, extract hand bianry image process it is as follows:
Area screening calculates the area of different connected regions in image, it is larger to extract area using eight connectivity distinguished number
Two connected regions, realize separating for face and hand and complex background；
Shape complexity threshold decision calculates the shape complexity C of remaining two connected regions, if the shape of connected region
Shape complexity C value is greater than threshold value T, then it is assumed that the region is non-hand region and removes, to obtain hand bianry image.
Further, the arm by hand bianry image removes, and it is as follows to obtain gesture bianry image process:
Centre of the palm positioning, is operated using range conversion, calculates minimum range of the hand pixel from hand boundary, and distance value takes
For preimage vegetarian refreshments value, except remaining region value of hand is 0, the maximum picture of value in the image obtained after range conversion operation
Vegetarian refreshments is the centre of the palm, and corresponding value is R0；
Palm cutting will be less than R at a distance from the centre of the palm1Pixel value be 0, to remove palm area, wherein R1
=1.35 × R0；
Threshold method judges that arm whether there is, and positions the pixel P of maximum value in image, and corresponding value is Pvalue,
Calculate Pvalue/R0If value is greater than threshold value T1, then P point region is arm regions, continues next removal arm behaviour
Make, arm regions are otherwise not present in hand bianry image, go to gesture feature extraction step；
Arm is removed, using eight connectivity distinguished number, removes P point region；
XOR operation finally obtains gesture bianry image using the XOR operation between image.
Further, the gesture feature extraction step includes:
The 7 invariant moments of gesture are calculated, Hu moment characteristics are constituted；
Using eight connectivity distinguished number, the perimeter and area of gesture area are calculated, calculates the ratio of perimeter and area；
Calculate the Fourier descriptor feature of gesture profile；
Combine Hu moment characteristics, perimeter and area ratio and Fourier descriptor feature, constitute 18 dimension gesture feature to
Amount.
Further, the Fourier descriptor characteristic procedure of the calculating gesture profile is as follows:
By the coordinate { (x of gesture contour edgek,yk) complex representation is used, constitute sequence of complex numbers { ck, ckIt is expressed as follows:
ck=xk+iyk, k=0,1,2 ..., N-1；
To discrete series { ckMake Fourier transformation, Fourier coefficient sequence { C (u) } is obtained, formula is as follows:
10 Fourier coefficients that u=1 starts are extracted, to its modulus value and are normalized, Fourier descriptor feature is constituted.
Further, the BP neural network includes input layer, hidden layer and output layer, and input layer has d neuron,
It is determined by the dimension of gesture feature vector, output layer has s neuron, is determined by gesture species number, and hidden layer has q nerve
Member, the connection weight between h-th of neuron of i-th of neuron of input layer and hidden layer are vih, h-th of neuron of hidden layer with
Connection weight between j-th of neuron of output layer is whj, the threshold value of h-th of neuron of hidden layer isJ-th of output layer
The threshold value of neuron is θj。
Further, before the gesture identification step, further includes:
BP neural network training step inputs the gesture feature vector training BP neural network of training sample, and process is as follows:
Random initializtion weight and threshold value, the value range for initializing weight is [- 1,1], the value model of initial threshold value
It encloses for [- 0.5,0.5]；
Input the gesture feature vector (x of training sample1,x2,...,x18)；
Calculate the output data of each layer, wherein the BP neural network is using sigmoid function as each layer neuron
Activation primitive, formula is as follows:
The output valve of h-th of neuron of hidden layer is αh, calculation formula is as follows:
The output valve of j-th of neuron of output layer isCalculation formula is as follows:
Mean square error E is calculated, calculation formula is as follows:
Wherein (y1,y2,...,y8) be training sample class label；
Parameter updates, and when E is greater than setting error, updates the weight and threshold value of network, using gradient descent method with amendment
Current BP neural network；When E is less than setting error, deconditioning network obtains optimal model parameter.
Further, the calculation formula of the shape complexity C is as follows:
Wherein, A is the area of connected region, and p is the perimeter of connected region.
The present invention has the following advantages and effects with respect to the prior art:
(1) present invention utilizes the Extraction of Geometrical Features hand region of human body complexion information and hand, is grasped using range conversion
It removes existing arm regions, realizes being precisely separating for gesture and complex background；
(2) present invention realizes strong robustness, a standard by combining a variety of effective gesture feature training BP neural networks
The high gesture recognition system of true rate；
(3), identification low with equipment cost that the present invention is based on the static gesture identification methods of common monocular cam is accurately
The advantages that rate is high, easy to promote and utilize.
Detailed description of the invention
Fig. 1 is the flow chart of monocular static gesture identification method based on multi-feature fusion disclosed in the present invention；
Fig. 2 is image preprocessing stream in monocular static gesture identification method based on multi-feature fusion disclosed in the present invention
Cheng Tu；
Fig. 3 is that gesture feature extracts in monocular static gesture identification method based on multi-feature fusion disclosed in the present invention
Flow chart；
Fig. 4 is gesture identification process in monocular static gesture identification method based on multi-feature fusion disclosed in the present invention
Figure.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Embodiment
As shown in Figure 1, a kind of monocular static gesture identification method based on multi-feature fusion, process are as follows: gesture figure
As acquisition step, image preprocessing step, gesture feature extraction step and gesture identification step.
Wherein, S1, images of gestures acquisition step:
It include the RGB image of gesture using monocular cam acquisition, the monocular cam should be located at immediately ahead of human body
Position, face and hand are biggish two of area in all colours of skin and class area of skin color in the image of acquisition.
Wherein, S2, image preprocessing step:
As shown in Fig. 2, image preprocessing step process is as follows:
S201, skin color segmentation, detailed process is as follows:
The image of input is transformed into YCr ' Cb ' color space by rgb color space by S2011, convert color spaces, tool
Body conversion formula is as follows:
Y=0.299 × r+0.587 × g+0.114 × b
Wherein r, g, b are respectively image three components of the red, green, blue in rgb color space；Y, cr', cb' are respectively
Luminance component, red chrominance component, chroma blue component of the image in YCr ' Cb ' color space.
S2012, Chroma threshold segmentation, if two chromatic components of pixel meet the colour of skin in cr' and cb' component simultaneously
When threshold range, enable the pixel value be 1, otherwise value be 0, thus by image the colour of skin and class area of skin color extract
Come, obtains a secondary bianry image.
S202, Morphological scale-space, detailed process is as follows:
Morphological scale-space is carried out to the bianry image after skin color segmentation, is first operated using opening operation, then is grasped using closed operation
Make, a large amount of isolated noise in image can be eliminated.
S203, the segmentation of hand geometry, detailed process is as follows:
S2031, area screening calculate the area of different connected regions in image, extract face using eight connectivity distinguished number
Biggish two connected regions of product realize separating for face and hand and complex background.
S2032, shape complexity threshold decision calculate the shape complexity of remaining two connected regions, if connected region
The shape complexity C value in domain is greater than threshold value T, then it is assumed that the region is non-hand region and removes, to obtain hand two-value
Image, the present invention take T=0.3 best, shape complexity C's specific formula is as follows:
Wherein, A is the area of connected region, and p is the perimeter of connected region.
S204, arm removal, specific practice are as follows:
S2041, centre of the palm positioning, are operated using range conversion, calculate minimum range of the hand pixel from hand boundary, away from
Replace preimage vegetarian refreshments value from value, except remaining region value of hand is 0, value is most in obtained image after range conversion operation
Big pixel is the centre of the palm, and corresponding value is R0。
S2042, palm cutting, will be less than R at a distance from the centre of the palm1Pixel value be 0, to remove palm area,
The present invention takes R1=1.35 × R0Most preferably.
S2043, threshold method judge that arm whether there is, and position the pixel P of maximum value in image, and corresponding value is
Pvalue, calculate Pvalue/R0If value is greater than threshold value T1, then P point region is arm regions, need to carry out removing arm behaviour
Make, arm regions are otherwise not present in hand bianry image, can be directly used for extracting gesture feature, the present invention takes T1=0.35 most
It is good.
S2044, arm is removed, using eight connectivity distinguished number, removes P point region.
S2045, XOR operation finally obtain gesture bianry image using the XOR operation between image.
Wherein, S3, gesture feature extraction step:
As shown in figure 3, extracting gesture feature to gesture bianry image, comprising the following steps:
S301, the 7 invariant moments for calculating gesture constitute Hu moment characteristics.
S302, using eight connectivity distinguished number, calculate the perimeter and area of gesture area, calculate the ratio of perimeter and area
Value.
S303, the Fourier descriptor feature for calculating gesture profile, specific practice are as follows:
S3031, the coordinate { (x by gesture contour edgek,yk) complex representation is used, constitute sequence of complex numbers { ck, ckSpecifically
It is expressed as follows:
ck=xk+iyk, k=0,1,2 ..., N-1；
S3032, to discrete series { ckMake Fourier transformation, Fourier coefficient sequence { C (u) } is obtained, specific formula is such as
Under:
S3033,10 Fourier coefficients that u=1 starts are extracted, to its modulus value and normalized, constitute Fourier's description
Subcharacter.
S304, the ratio and Fourier descriptor feature for combining Hu moment characteristics, perimeter and area, the gesture for constituting 18 dimensions are special
Levy vector.
Wherein, S4, gesture identification step:
The present invention is realized using the gesture feature vector of extraction as the input of BP neural network by training BP neural network
Static gesture classification.The BP neural network includes input layer, hidden layer and output layer, and input layer has d neuron, by hand
The dimension of gesture feature vector determines that output layer has s neuron, determined by gesture species number, and hidden layer has q neuron, defeated
Entering the connection weight between h-th of neuron of i-th of neuron of layer and hidden layer is vih, h-th of neuron of hidden layer and output
Connection weight between j-th of neuron of layer is whj, the threshold value of h-th of neuron of hidden layer isJ-th of nerve of output layer
The threshold value of member is θj.The present invention chooses d=18, q=10, s=8, and the present invention realizes the classification of 8 kinds of static gestures.
As shown in figure 4, before gesture identification step, further includes:
BP neural network training step inputs the gesture feature vector training BP neural network of training sample, detailed process
It is as follows:
I. random initializtion weight and threshold value, the value range for initializing weight is [- 1,1], the value of initial threshold value
Range is [- 0.5,0.5].
Ii. the gesture feature vector (x of training sample is inputted1,x2,...,x18)。
Iii. the output data of each layer is calculated:
BP neural network of the present invention is all made of activation primitive of the sigmoid function as each layer neuron, and specific formula is such as
Under:
The output valve of h-th of neuron of hidden layer is αh, calculation formula is as follows:
The output valve of j-th of neuron of output layer isCalculation formula is as follows:
Iv. mean square error E is calculated, specific formula is as follows:
Wherein (y1,y2,...,y8) be training sample class label.
V. parameter updates, and when E is greater than setting error, updates the weight and threshold value of network, using gradient descent method to repair
Proper preceding BP neural network；When E is less than setting error, deconditioning network obtains optimal model parameter.
Sorting phase carries out gesture classification with trained BP neural network model, and specific practice is as follows: input is to be sorted
The gesture feature vector of sample, calculates the output data of each layer, obtains the output valve of network, the as classification results of gesture.
In conclusion present embodiment discloses a kind of monocular static gesture identification method based on multi-feature fusion, the party
Method utilizes the Extraction of Geometrical Features hand region of human body complexion information and hand, removes existing arm using range conversion operation
Region, realization gesture and complex background are precisely separating.This method is by combining a variety of effective gesture feature training BP nerves
Network realizes the high gesture recognition system of strong robustness, an accuracy rate.In addition, this method is real using common monocular cam
Existing static gesture identification, has many advantages, such as that equipment cost is low, recognition accuracy is high, easy to promote and utilize.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (10)
1. a kind of monocular static gesture identification method based on multi-feature fusion, which is characterized in that the recognition methods includes:
Images of gestures acquisition step includes the RGB image of gesture with monocular cam acquisition；
Image preprocessing step, using human body complexion information carry out skin color segmentation, by image the colour of skin and class area of skin color mention
It takes out, using the geometrical characteristic of Morphological scale-space and combination hand, hand is separated with complex background, is grasped by range conversion
Make the positioning centre of the palm and remove arm regions existing for hand, obtains gesture bianry image；
Gesture feature extraction step calculates the perimeter and area ratio, Hu square and Fourier descriptor feature of gesture, constitutes gesture
Feature vector；
Gesture identification step passes through training BP neural network using the gesture feature vector of extraction as the input of BP neural network
Realize static gesture classification.
2. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 1, which is characterized in that
The image preprocessing step includes:
The image of input is carried out the conversion of color space by skin color segmentation, by Chroma threshold segmentation by image the colour of skin and
Class area of skin color extracts, and obtains bianry image；
Morphological scale-space is carried out Morphological scale-space to the bianry image after skin color segmentation, is first operated using opening operation, then used and close
Arithmetic operation eliminates the isolated noise in image；
The segmentation of hand geometry realizes that face and hand are separated with complex background, calculates remaining two connected regions
Shape complexity C extracts hand bianry image by judging with threshold value T；
By the arm removal in hand bianry image, gesture bianry image is obtained.
3. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 2, which is characterized in that
The described image by input carries out the conversion of color space, by Chroma threshold segmentation by image the colour of skin and class colour of skin area
Domain extracts, and it is as follows to obtain bianry image process:
The image of input is transformed into YCr ' Cb ' color space by rgb color space by convert color spaces, and conversion formula is as follows:
Y=0.299 × r+0.587 × g+0.114 × b
Wherein r, g, b are respectively image three components of the red, green, blue in rgb color space, and y, cr', cb' are respectively image
Luminance component, red chrominance component, chroma blue component in YCr ' Cb ' color space；
Chroma threshold segmentation, if two chromatic components of pixel meet the colour of skin in the threshold range of cr' and cb' component simultaneously
When, enable the pixel value be 1, otherwise value be 0, thus by image the colour of skin and class area of skin color extract, obtain
Bianry image.
4. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 2, which is characterized in that
The hand geometry segmentation, realizes that face and hand are separated with complex background, calculates remaining two connected regions
Shape complexity C, by with threshold value T judge, extract hand bianry image process it is as follows:
Area screening calculates the area of different connected regions in image, extracts area biggish two using eight connectivity distinguished number
A connected region, realization face and hand are separated with complex background；
Shape complexity threshold decision calculates the shape complexity C of remaining two connected regions, if the shape of connected region is multiple
Miscellaneous degree C value is greater than threshold value T, then it is assumed that the region is non-hand region and removes, to obtain hand bianry image.
5. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 2, which is characterized in that
Described removes the arm in hand bianry image, and it is as follows to obtain gesture bianry image process:
Centre of the palm positioning, is operated using range conversion, calculates minimum range of the hand pixel from hand boundary, and distance value replaces former
Pixel value, except remaining region value of hand is 0, the maximum pixel of value in the image obtained after range conversion operation
For the centre of the palm, corresponding value is R0；
Palm cutting will be less than R at a distance from the centre of the palm1Pixel value be 0, to remove palm area, wherein R1=
1.35×R0；
Threshold method judges that arm whether there is, and positions the pixel P of maximum value in image, and corresponding value is Pvalue, calculate
Pvalue/R0If value is greater than threshold value T1, then P point region is arm regions, continue next removal arm operation,
Otherwise arm regions are not present in hand bianry image, go to gesture feature extraction step；
Arm is removed, using eight connectivity distinguished number, removes P point region；
XOR operation finally obtains gesture bianry image using the XOR operation between image.
6. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 1, which is characterized in that
The gesture feature extraction step includes:
The 7 invariant moments of gesture are calculated, Hu moment characteristics are constituted；
Using eight connectivity distinguished number, the perimeter and area of gesture area are calculated, calculates the ratio of perimeter and area；
Calculate the Fourier descriptor feature of gesture profile；
The ratio and Fourier descriptor feature of Hu moment characteristics, perimeter and area are combined, the gesture feature vector of 18 dimensions is constituted.
7. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 6, which is characterized in that
The Fourier descriptor characteristic procedure of the calculating gesture profile is as follows:
By the coordinate { (x of gesture contour edgek, yk) complex representation is used, constitute sequence of complex numbers { ck, ckIt is expressed as follows:
ck=xk+iyk, k=0,1,2 ..., N-1；
To discrete series { ckMake Fourier transformation, Fourier coefficient sequence { C (u) } is obtained, formula is as follows:
10 Fourier coefficients that u=1 starts are extracted, to its modulus value and are normalized, Fourier descriptor feature is constituted.
8. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 1, which is characterized in that
The BP neural network includes input layer, hidden layer and output layer, and input layer has d neuron, by gesture feature vector
Dimension determines that output layer has s neuron, determined by gesture species number, and hidden layer has q neuron, i-th of nerve of input layer
Connection weight between member and h-th of neuron of hidden layer is vih, h-th of neuron of hidden layer and j-th of neuron of output layer
Between connection weight be whj, the threshold value of h-th of neuron of hidden layer isThe threshold value of j-th of neuron of output layer is θj。
9. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 8, which is characterized in that
Before the gesture identification step, further includes:
BP neural network training step inputs the gesture feature vector training BP neural network of training sample, and process is as follows:
Random initializtion weight and threshold value, the value range for initializing weight is [- 1,1], and the value range of initial threshold value is
[-0.5,0.5]；
Input the gesture feature vector (x of training sample1,x2,...,x18)；
Calculate the output data of each layer, wherein the BP neural network swashing as each layer neuron using sigmoid function
Function living, formula are as follows:
The output valve of h-th of neuron of hidden layer is αh, calculation formula is as follows:
The output valve of j-th of neuron of output layer isCalculation formula is as follows:
Mean square error E is calculated, calculation formula is as follows:
Wherein (y1,y2,...,y8) be training sample class label；
Parameter updates, and when E is greater than setting error, the weight and threshold value of network are updated using gradient descent method, current to correct
BP neural network；When E is less than setting error, deconditioning network obtains optimal model parameter.
10. a kind of monocular static gesture identification method based on multi-feature fusion according to claim 4, feature exist
In the calculation formula of the shape complexity C is as follows:
Wherein, A is the area of connected region, and p is the perimeter of connected region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810900949.5A CN109190496A (en) | 2018-08-09 | 2018-08-09 | A kind of monocular static gesture identification method based on multi-feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810900949.5A CN109190496A (en) | 2018-08-09 | 2018-08-09 | A kind of monocular static gesture identification method based on multi-feature fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109190496A true CN109190496A (en) | 2019-01-11 |
Family
ID=64921162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810900949.5A Pending CN109190496A (en) | 2018-08-09 | 2018-08-09 | A kind of monocular static gesture identification method based on multi-feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109190496A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109934159A (en) * | 2019-03-11 | 2019-06-25 | 西安邮电大学 | A kind of gesture identification method of multiple features fusion |
| CN110796033A (en) * | 2019-10-12 | 2020-02-14 | 江苏科技大学 | Static gesture recognition method based on bounding box model |
| CN111160194A (en) * | 2019-12-23 | 2020-05-15 | 浙江理工大学 | Static gesture image recognition method based on multi-feature fusion |
| CN111258430A (en) * | 2020-01-21 | 2020-06-09 | 哈尔滨拓博科技有限公司 | Desktop interaction system based on monocular gesture control |
| CN111339970A (en) * | 2020-03-02 | 2020-06-26 | 上海化学工业区公共管廊有限公司 | Smoking behavior detection method suitable for public environment |
| WO2020173024A1 (en) * | 2019-02-26 | 2020-09-03 | 南京邮电大学 | Multi-gesture precise segmentation method for smart home scenario |
| CN111901681A (en) * | 2020-05-04 | 2020-11-06 | 东南大学 | Intelligent television control device and method based on face recognition and gesture recognition |
| WO2020237519A1 (en) * | 2019-05-29 | 2020-12-03 | 深圳大学 | Identification method, apparatus and device, and storage medium |
| CN112034981A (en) * | 2020-08-20 | 2020-12-04 | 深圳创维－Rgb电子有限公司 | Display terminal control method, display terminal, and computer-readable storage medium |
| CN112068705A (en) * | 2020-09-15 | 2020-12-11 | 山东建筑大学 | Bionic robot fish interaction control method and system based on gesture recognition |
| CN112232217A (en) * | 2020-10-16 | 2021-01-15 | 怀化新大地电脑有限公司 | Gesture recognition system |
| CN112906550A (en) * | 2021-02-09 | 2021-06-04 | 哈尔滨理工大学 | Static gesture recognition method based on watershed transformation |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106293057A (en) * | 2016-07-20 | 2017-01-04 | 西安中科比奇创新科技有限责任公司 | Gesture identification method based on BP neutral net |
| CN106503619A (en) * | 2016-09-23 | 2017-03-15 | 南京理工大学 | Gesture identification method based on BP neural network |
| CN106503651A (en) * | 2016-10-21 | 2017-03-15 | 上海未来伙伴机器人有限公司 | A kind of extracting method of images of gestures and system |
| CN107133562A (en) * | 2017-03-17 | 2017-09-05 | 华南理工大学 | A kind of gesture identification method based on extreme learning machine |
| CN108108648A (en) * | 2016-11-24 | 2018-06-01 | 广州映博智能科技有限公司 | A kind of new gesture recognition system device and method |
-
2018
- 2018-08-09 CN CN201810900949.5A patent/CN109190496A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106293057A (en) * | 2016-07-20 | 2017-01-04 | 西安中科比奇创新科技有限责任公司 | Gesture identification method based on BP neutral net |
| CN106503619A (en) * | 2016-09-23 | 2017-03-15 | 南京理工大学 | Gesture identification method based on BP neural network |
| CN106503651A (en) * | 2016-10-21 | 2017-03-15 | 上海未来伙伴机器人有限公司 | A kind of extracting method of images of gestures and system |
| CN108108648A (en) * | 2016-11-24 | 2018-06-01 | 广州映博智能科技有限公司 | A kind of new gesture recognition system device and method |
| CN107133562A (en) * | 2017-03-17 | 2017-09-05 | 华南理工大学 | A kind of gesture identification method based on extreme learning machine |
Non-Patent Citations (5)
| Title |
|---|
| 周龙: "基于神经网络方法的手势行为识别技术研究", 《中国优秀硕士论文全文数据库信息科技辑》 * |
| 徐常青等: "《数学实验与软件计算》", 31 January 2014 * |
| 曹建秋等: "基于改进YCrCb颜色空间的肤色分割", 《重庆交通大学学报自然科学版》 * |
| 杨帆等: "《精通图像处理经典算法（MATLAB版）》", 30 April 2014 * |
| 赵小川等: "《MATLAB数字图像处理-从仿真到C/C++代码的自动生成》", 30 September 2015 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2021517281A (en) * | 2019-02-26 | 2021-07-15 | 南京郵電大学Ｎａｎｊｉｎｇ Ｕｎｉｖｅｒｓｉｔｙ Ｏｆ Ｐｏｓｔｓ Ａｎｄ Ｔｅｌｅｃｏｍｍｕｎｉｃａｔｉｏｎｓ | Multi-gesture fine division method for smart home scenes |
| WO2020173024A1 (en) * | 2019-02-26 | 2020-09-03 | 南京邮电大学 | Multi-gesture precise segmentation method for smart home scenario |
| CN109934159A (en) * | 2019-03-11 | 2019-06-25 | 西安邮电大学 | A kind of gesture identification method of multiple features fusion |
| WO2020237519A1 (en) * | 2019-05-29 | 2020-12-03 | 深圳大学 | Identification method, apparatus and device, and storage medium |
| CN110796033A (en) * | 2019-10-12 | 2020-02-14 | 江苏科技大学 | Static gesture recognition method based on bounding box model |
| CN110796033B (en) * | 2019-10-12 | 2023-07-28 | 江苏科技大学 | Static gesture recognition method based on bounding box model |
| CN111160194A (en) * | 2019-12-23 | 2020-05-15 | 浙江理工大学 | Static gesture image recognition method based on multi-feature fusion |
| CN111258430A (en) * | 2020-01-21 | 2020-06-09 | 哈尔滨拓博科技有限公司 | Desktop interaction system based on monocular gesture control |
| CN111339970A (en) * | 2020-03-02 | 2020-06-26 | 上海化学工业区公共管廊有限公司 | Smoking behavior detection method suitable for public environment |
| CN111339970B (en) * | 2020-03-02 | 2023-04-07 | 上海化学工业区公共管廊有限公司 | Smoking behavior detection method suitable for public environment |
| CN111901681B (en) * | 2020-05-04 | 2022-09-30 | 东南大学 | Intelligent television control device and method based on face recognition and gesture recognition |
| CN111901681A (en) * | 2020-05-04 | 2020-11-06 | 东南大学 | Intelligent television control device and method based on face recognition and gesture recognition |
| CN112034981A (en) * | 2020-08-20 | 2020-12-04 | 深圳创维－Rgb电子有限公司 | Display terminal control method, display terminal, and computer-readable storage medium |
| CN112068705A (en) * | 2020-09-15 | 2020-12-11 | 山东建筑大学 | Bionic robot fish interaction control method and system based on gesture recognition |
| CN112232217A (en) * | 2020-10-16 | 2021-01-15 | 怀化新大地电脑有限公司 | Gesture recognition system |
| CN112232217B (en) * | 2020-10-16 | 2022-08-02 | 怀化新大地电脑有限公司 | Gesture recognition system |
| CN112906550A (en) * | 2021-02-09 | 2021-06-04 | 哈尔滨理工大学 | Static gesture recognition method based on watershed transformation |
| CN112906550B (en) * | 2021-02-09 | 2022-07-19 | 哈尔滨理工大学 | Static gesture recognition method based on watershed transformation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109190496A (en) | A kind of monocular static gesture identification method based on multi-feature fusion | |
| Nikam et al. | Sign language recognition using image based hand gesture recognition techniques | |
| CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
| CN108875787A (en) | A kind of image-recognizing method and device, computer equipment and storage medium | |
| CN108256421A (en) | A kind of dynamic gesture sequence real-time identification method, system and device | |
| CN106778785B (en) | Construct the method for image Feature Selection Model and the method, apparatus of image recognition | |
| CN109558832A (en) | A kind of human body attitude detection method, device, equipment and storage medium | |
| Konwar et al. | An American sign language detection system using HSV color model and edge detection | |
| CN111274921B (en) | Method for recognizing human body behaviors by using gesture mask | |
| CN109214297A (en) | A kind of static gesture identification method of combination depth information and Skin Color Information | |
| Bilal et al. | A hybrid method using haar-like and skin-color algorithm for hand posture detection, recognition and tracking | |
| CN106446862A (en) | Face detection method and system | |
| CN111158491A (en) | Gesture recognition man-machine interaction method applied to vehicle-mounted HUD | |
| CN112906550B (en) | Static gesture recognition method based on watershed transformation | |
| Vishwakarma et al. | Simple and intelligent system to recognize the expression of speech-disabled person | |
| CN106097354A (en) | A kind of combining adaptive Gauss Face Detection and the hand images dividing method of region growing | |
| Barkoky et al. | Static hand gesture recognition of Persian sign numbers using thinning method | |
| CN111080670A (en) | Image extraction method, device, equipment and storage medium | |
| CN110046544A (en) | Digital gesture identification method based on convolutional neural networks | |
| CN110956099A (en) | Dynamic gesture instruction identification method | |
| CN111160194B (en) | Static gesture image recognition method based on multi-feature fusion | |
| CN103778430B (en) | Rapid face detection method based on combination between skin color segmentation and AdaBoost | |
| CN106909884A (en) | A kind of hand region detection method and device based on hierarchy and deformable part sub-model | |
| Lei et al. | A novel side face contour extraction algorithm for driving fatigue statue recognition | |
| Sokhib et al. | A combined method of skin-and depth-based hand gesture recognition. |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190111 |