CN103677274B - A kind of interaction method and system based on active vision - Google Patents
A kind of interaction method and system based on active vision Download PDFInfo
- Publication number
- CN103677274B CN103677274B CN201310724304.8A CN201310724304A CN103677274B CN 103677274 B CN103677274 B CN 103677274B CN 201310724304 A CN201310724304 A CN 201310724304A CN 103677274 B CN103677274 B CN 103677274B
- Authority
- CN
- China
- Prior art keywords
- skin
- colour
- training
- color
- prior probability
- 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
Abstract
The invention discloses a kind of interaction method and system based on active vision, its method is that projection arrangement will need interactive image projection to arbitrary plane, and use the operation information of the filming apparatus collection user interactive picture to being projected out, the information of shooting is transferred to processing means be analyzed processing simultaneously, realizes man-machine interaction with the operation obtaining user corresponding.The processing means of the present invention uses the mode of off-line training so that detection efficiency is improved.When training data, it is proposed that a kind of new training mechanism, decrease substantial amounts of manual operation.When detection, have employed the model of real-time update, the colour of skin of the most current complexion model only several frames in front is relevant, so can effectively remove the illumination impact on the colour of skin, and effectively improve efficiency of algorithm, fully achieve real-time requirement.
Description
Technical field
The present invention relates to technical field of computer vision, more particularly, to a kind of interaction based on active vision
Projecting method and system.
Then the picture to required interaction performs corresponding user operation, and by projection device to projection
Region is watched for user；
Background technology
Although the multimodal human-computer interaction technology of integrated use vision, audition, sense of touch, olfactory sensation, the sense of taste etc. is more come
Be applied manyly, but, both hands as action important in virtual reality system and perception relational model,
It still plays the most replaceable effect in virtual reality system.At present, touch system is as a kind of up-to-date electricity
Brain input equipment, it is man-machine interaction mode the simplest, convenient, a kind of.It give many matchmakers
Body, with brand-new looks, is extremely attractive brand-new multimedia interactive equipment.Along with the progress of science and technology, projection
The use of instrument is more and more easier, and any one plane can be become a display screen by it, and projector widely should
For training conference, classroom instruction, cinema etc..
At present, camera and projection have gradually entered into the life of ordinary citizen, and carry out hands by projection and camera
Automatically the identification of gesture becomes current study hotspot, by the automatic identification of gesture, has reached more preferable man-machine friendship
Mutually so that the use of projection is convenient.The projection interactive system being currently based on vision is substantially based on auxiliary
Light projection or based on finger tip location, but not strong in current existing scheme mostly real-time, and robustness is the highest.
Patent 200910190517.0 discloses a kind of projection interactive method based on finger, and hands is passed through in this invention
The information such as CF in video file, extract the profile of finger, and record the movement locus of finger,
Then by the movement locus of finger and the instruction contrast in the instruction database pre-defined, affiliated operation rail is judged
Which kind of operational order mark belongs to, and reaches the purpose of man-machine interaction.Patent 200910197516.9 discloses one
Finger identification method in interactive demonstration system, for interactive demonstration system based on video camera projection
In determined the operation behavior of user by the identification of finger.This invention is based on image processing techniques, according to
User paint operation time finger gesture feature, with its geometric space positional information for mainly identifying information, to taking the photograph
As the image of head shooting is analyzed and processes, thus identify finger.In interactive projection system, in order to right
Gesture is identified, and first has to segmentation from picture and obtains hand region.And at present about the segmentation of hand region
An always difficult point.In projection interactive system, due to the irradiation of projector light, the arm of people may
Presenting different colors, be also possible to comprise staff in projected picture, this brings difficulty all to segmentation arm.
Summary of the invention
For overcoming at least one defect (not enough) described in above-mentioned prior art, present invention firstly provides a kind of knowledge
Other efficiency and the high interaction method based on active vision of accuracy.
A further object of the present invention is to propose a kind of interactive projection system based on active vision.
To achieve these goals, technical scheme is as follows:
A kind of interaction method based on active vision, projection arrangement will need interactive image projection to arbitrarily
Plane, and use the operation information of the filming apparatus collection user interactive picture to being projected out, simultaneously by shooting
Information is transferred to processing means and is analyzed processing, and realizes man-machine interaction with the operation obtaining user corresponding,
The process of described processing means Treatment Analysis information is: a selected color space to illumination-insensitive,
Use Bayes' assessment to extract hand region, and this hand region be tracked and the location of fingertip location,
Judge whether finger contacts with projection screen by range finding again, to determine whether to carry out man-machine interaction；
Using Bayes' assessment to extract hand region, its concrete mode is:
By off-line training acquisition probability P (s), P (c), P (c | s) and P (s | c), wherein P (s) represents training
During the prior probability of the colour of skin, P (c) represents the prior probability of each color during training, P (c | s) table
Show the prior probability that pixel is the colour of skin that pixel value is c, and P (s | c) represent that pixel value is c's after training
Pixel is the probability of the colour of skin；
P (s | c)=P (c | s) P (s)/P (c) (1)
Hysteresis threshold T is obtained during off-line trainingmax, Tmax=P (s | c)；It is to obtain according to probability distribution graph
, namely probability distribution is interval；
Adaptive method is used to carry out Face Detection to extract hand region, according to the colour of skin district of nearest w frame identification
Territory judges the area of skin color point detected at present, and the prior probability of nearest w frame is Pw(s), Pw(c), Pw(c | s),
Wherein PwS () represents the prior probability of the colour of skin, P in nearest w framewC () represents each color in nearest w frame
Prior probability, Pw(c | s) represent that in nearest w frame, pixel value is the prior probability that pixel is the colour of skin of c, separately examines
The probability P that survey is area of skin color ' (s | c) be:
P'(s | and c)=γ P (s | c)+(1-γ) Pw(s|c) (2)
Wherein Pw(s | c) obtained by formula (1), i.e. Pw(s | c)=Pw(c|s)Pw(s)/PwC (), γ is one
Control coefrficient, this control coefrficient is relevant with the training set of training stage；
Work as P'(s | c) ＞ Tmax, detect whether the most at present into area of skin color, so far to extract hand region, otherwise
It is not belonging to area of skin color.
Having a lot about hand region extracting method at present, the most frequently used having is based on hand color, based on hand
Shape and the method demarcated based on color and physics.The color space that Face Detection is used at present mainly has
The most several, RGB, normalized RGB, HSV, YCrCb, YUV etc..In the method, carry out
During Face Detection, use color space to illumination-insensitive, thus so that the detection robustness of the colour of skin
The highest.After a given color space, the way that the simplest judgement colour of skin which color is made up of is right
Selected space adds a restrictive condition, i.e. afterwards threshold value Tmax.This hysteresis threshold TmaxDrawn by experience,
As at given a series of area of skin color image, so can be obtained by the distribution of area of skin color.
This method uses Bayesian learning method based on off-line when extracting hand region, uses non-pattra leaves subsequently
This framework is tracked.With existing algorithm is compared, this method has the advantage that (1) passes through off-line
The mode of study obtains the distribution of the colour of skin, and so can be greatly increased detection is speed.(2) have employed in real time
The complexion model of change, i.e. this complexion model only several frames in front are relevant, therefore, although do not have the colour of skin mould of complexity
Type, under the light conditions even in change, also can robust, effectively identify area of skin color.(3) side
The efficiency of algorithm that method is used is the highest, can reach real-time process.
A kind of system being applied to described interaction method based on active vision, including for needing mutually
Dynamic image projection is to the projection arrangement of arbitrary plane, for gathering the operation of user's interactive picture to being projected out
The filming apparatus of information and carry out the processing means of information process analysis,
The process of described processing means Treatment Analysis information is: a selected color space to illumination-insensitive,
Use Bayes' assessment to extract hand region, and this hand region be tracked and the location of fingertip location,
Judge whether finger contacts with projection screen by range finding again, to determine whether to carry out man-machine interaction；
Using Bayes' assessment to extract hand region, its concrete mode is:
By off-line training acquisition probability P (s), P (c), P (c | s) and P (s | c), wherein P (s) represents training
During the prior probability of the colour of skin, P (c) represents the prior probability of each color during training, P (c | s) table
Show the prior probability that pixel is the colour of skin that pixel value is c, and P (s | c) represent that pixel value is c's after training
Pixel is the probability of the colour of skin；
P (s | c)=P (c | s) P (s)/P (c) (1)
Hysteresis threshold T is obtained during off-line trainingmax, Tmax=P (s | c)；
Adaptive method is used to carry out Face Detection to extract hand region, according to the colour of skin district of nearest w frame identification
Territory judges the area of skin color point detected at present, and the prior probability of nearest w frame is Pw(s), Pw(c), Pw(c | s),
Wherein PwS () represents the prior probability of the colour of skin, P in nearest w framewC () represents each color in nearest w frame
Prior probability, Pw(c | s) represent that in nearest w frame, pixel value is the prior probability that pixel is the colour of skin of c, separately examines
The probability P that survey is area of skin color ' (s | c) be:
P'(s | and c)=γ P (s | c)+(1-γ) Pw(s|c) (2)
Wherein Pw(s | c) obtained by formula (1), i.e. Pw(s | c)=Pw(c|s)Pw(s)/PwC (), γ is one
Control coefrficient, this control coefrficient is relevant with the training set of training stage；
Work as P'(s | c) ＞ Tmax, detect whether the most at present into area of skin color, so far to extract hand region, otherwise
It is not belonging to area of skin color.
Compared with prior art, technical solution of the present invention provides the benefit that: the present invention uses the side of off-line training
Formula so that detection efficiency is improved.When training data, it is proposed that a kind of new training mechanism, decrease
Substantial amounts of manual operation.When detection, have employed the model of real-time update, the most current complexion model only front
The colour of skin of several frames is relevant, so can effectively remove the illumination impact on the colour of skin.And the calculation that this patent proposes
Method efficiency is the highest, fully achieves real-time requirement.
Accompanying drawing explanation
Fig. 1 is the interactive projection system schematic diagram of the present invention.
Fig. 2 is the flow chart of processing means of the present invention analyzing and processing information.
Fig. 3 is for using triangulation measuring principle figure.
Detailed description of the invention
Accompanying drawing being merely cited for property explanation, it is impossible to be interpreted as the restriction to this patent；
In order to the present embodiment is more preferably described, some parts of accompanying drawing have omission, zoom in or out, and do not represent reality
The size of border product；
To those skilled in the art, in accompanying drawing, some known features and explanation thereof may be omitted is to manage
Solve.
It is illustrated in figure 1 the optical projection system schematic diagram of the present invention, including projector, two cameras, calculating equipment.
The main effect of projector is exactly that picture is thrown into arbitrary plane.The effect of camera is the picture that shooting projects away
Face, is then communicated in calculating equipment.The data of collected by camera are mainly analyzed by calculating equipment, its point
The flow chart that analysis processes is as shown in Figure 2.
First, it is from complex background, extract staff.Extracting staff from complex background is exactly from entire image
The middle staff part by correspondence extracts, and it relates to the segmentation of image and asks staff area judging two
Topic.Segmentation image typically belongs to be the feature extraction of low level, mainly make use of the geological information of staff, color
Information and movable information.Wherein, geological information includes the shape of staff, profile etc.；Movable information refers to
The movement locus of staff.The most accurate location fingertip location that is extracted as in staff region is laid a good foundation, generally
The methods such as grey relevant dynamic matrix, edge detection operator method, calculus of finite differences can be used to realize.In the present invention, in order to
Remove the impact that projection ray irradiates, use method based on Bayesian Estimation when arm extracts.Finding hands
After the position in portion, start to keep the tracking to hand region, simultaneously need to position finger tip accurately.And about
The localization method of finger tip also has a lot, has special marking method, edge analysis, Hough circle transformation etc..
Finally, need calculating camera to the distance of finger, judge whether finger contacts with projection screen.This is
The main method using triangulation of system, i.e. based on binocular vision three-dimensional rebuilding method calculates this distance.
In the present embodiment, the method for " Bayesian Estimation " is used to carry out skin cluster:
Detection method about hand region has a lot at present, and the most frequently used having is based on hand color, based on hands
Portion's shape and the method demarcated based on color and physics.It is to illumination that current algorithm exists maximum problem
The most sensitive, secondly it is exactly the efficiency of algorithm.The color space that at present Face Detection is used mainly have with
Under several, RGB, normalized RGB, HSV, YCrCb, YUV etc..When carrying out Face Detection,
The color space to illumination-insensitive can be paid the utmost attention to, thus so that the detection robustness of the colour of skin compares
High.After a given color space, the way that the simplest judgement colour of skin which color is made up of is to selected
Space add restrictive condition, i.e. a hysteresis threshold.This hysteresis threshold is drawn by experience, as given one
The area of skin color image of series, so can be obtained by the distribution of area of skin color.The present embodiment uses based on off-line
Bayesian learning method detect area of skin color, use non-Bayesian frame to be tracked subsequently.With existing
Algorithm compare, the present invention has an advantages below: (1) obtain by the way of off-line learning the colour of skin point
Cloth, so can be greatly increased detection is speed.(2) have employed the complexion model of real-time change, i.e. this skin
The color model only several frames in front are relevant, therefore, although do not have the complexion model of complexity, even in the illumination of change
In the case of, also can robust, effectively identify area of skin color.(3) efficiency of algorithm of the present invention
The highest, real-time process can be reached.
1.1 Face Detection
Face Detection mainly includes following components: (a) estimates certain pixel and belong to the probability of the colour of skin；
B () obtains hysteresis threshold T according to probability distribution graphmax；Face Detection have employed bayes method, mainly wraps
Include the off-line training process of iteration and adaptive detection process.
A. training and testing mechanism
Firstly the need of the given a series of picture including area of skin color, and need manual choosing in picture
Going out flesh tone portion, employing color space is YUV4:2:2.The when that but the present embodiment specifically being trained and identifies,
Do not use Y-component, mainly have following two reason: (1) Y-component is relevant with the brightness of pixel,
Therefore remove Y-component, can effectively reduce the illumination impact on detection.(2) after removing Y-component, phase
For YUV, the dimension of image reduces, so the efficiency of whole process can be greatly promoted.
Assuming that a pixel is that (x, y), its pixel value is that (x, y), training process mainly calculates following several c=c to I
Individual value: prior probability P (s) of (1) colour of skin；(2) prior probability that each color occurs in training data
P(c)；(3) pixel value is the prior probability P that pixel is the colour of skin (c | s) of c；Through training after, pixel
The probability P that pixel is the colour of skin (s | c) that value is c can be obtained by bayesian criterion:
P (s | c)=P (c | s) P (s)/P (c) (1)
I.e. can determine that hysteresis threshold Tmax, Tmax=P (s | c), in being embodied as, when a certain pixel belongs to skin
The probability of color is more than hysteresis threshold Tmax, then this point is the colour of skin.
A. the off-line training after simplifying
Training mainly completes in the case of off-line, asks so it does not interferes with online detection efficiency etc.
Topic.But, the training data that will obtain abundance is a job the most time-consuming.In order to solve this problem,
The present embodiment uses a kind of adaptive training process.First, it is trained with one group of little data set,
Then carry out the identification of area of skin color by substantial amounts of data and hysteresis threshold, and real-time renewal priori is general
Rate P (s), P (c) and P (c | s).The most continuous update after threshold value by the area of skin color in picture and
Non-area of skin color is separately.If classifier training obtains the result of mistake, then need manpower intervention to correct this
Mistake, but the method still can complete the work required for major part.If it is intended to more accurate result, can
To input more training data, if the result of training has reached demand, can deconditioning immediately.
B. the self-adapting detecting colour of skin
Even if having employed UV model, in the case of illumination is continually changing, still can obtain some mistakes
Recognition result.In order to solve this problem, need according to the area of skin color of former frame identifications judge current
Area of skin color point.Therefore, this patent have employed two groups of prior probabilitys: i.e. off-line training prior probability P (s), P (c)
With P (c | s).The prior probability P of w frame recentlyw(s), Pw(c), Pw(c|s).The priori of the most nearest w frame
Probability can reflect current colour of skin state, and also can preferably adapt to current light conditions.The colour of skin is then
Defined by following formula:
P'(s | and c)=γ P (s | c)+(1-γ) Pw(s|c) (2)
Wherein Pw(s | c) and Pw(s | c) all can be obtained by (1) formula, the priori of the most whole training set is general
Rate and the prior probability of nearest w frame.γ is then a control coefrficient, and this control coefrficient is with the training of detection-phase
Collect relevant.Work as P'(s | c) ＞ Tmax, detect whether the most at present into area of skin color, so far to extract hand region,
Otherwise it is not belonging to area of skin color.
1.2 colours of skin are followed the tracks of
Owing to being tracked just for hand region, i.e. monotrack, the present embodiment have employed Camshift
Algorithm, what this algorithm was the most highly developed at present applies in motion tracking and image segmentation, and it is by video
All frame numbers make meanShift computing, and meanShift algorithm is the gradient ascent algorithm of a variable step, and will
The size of the result of previous frame, i.e. search window and center, as at the beginning of next frame meanshift algorithm search window
Initial value.With this, iteration continues, so that it may realizes the tracking to target.Algorithmic procedure is as follows:
(1) search window is initialized；(2) color probability distribution of search window is calculated；(3) meanShift is run
Algorithm, it is thus achieved that the size and location of search window；(4) in next frame video image, the initial of step (3) is used
Value reinitializes the size and location of search window, then jumps to step (2) and proceed.
1.3 find finger tip
After taking out arm, task below is intended to find finger tip position in the picture, the method finding finger tip
Have a variety of, such as the approximation K curvature by calculating profile, take extreme value according to K curvature at finger tip point and obtain finger
Point position. the present invention is by following steps searching finger tip:
1) find out largest contours and fill this profile, thus having obtained not having noisy arm foreground image；
2) convex closure of profile is calculated；
3) calculate arm position of centre of gravity, convex closure is found out several candidate points of maximum curvature；
4) using candidate point farthest for distance position of centre of gravity as finger tip.
1.4 range finding
After finding fingertip location, need the distance obtaining finger tip to camera.According to solid geometry principle, first
The demarcation of projector and video camera is carried out during the work to be done.Accurately and simple calibration process be projector-
Video camera constitutes active vision system and carries out the key point of three-dimensional measurement, and calibrating camera method becomes the most very much
Ripe, the most commonly used is the scaling method of Zhang Zhengyou.Calibrating camera has only to use a plane chessboard
Lattice.As long as finding the projection just can be in the hope of as the corresponding relation between the two-dimensional points of three-dimensional coordinate point and projection picture
Solve inner parameter and the external parameter of projector.
Following steps are used to carry out labeling projection instrument in the present embodiment:
S1. calibrating camera；
S2. prepare a blank, blank posts a papery gridiron pattern；
S3. control projector projects to go out a gridiron pattern and be radiated on blank；
Extract two tessellated angle points the most respectively；
S5. blank place plane is calculated according to papery X-comers；
S6. the video camera demarcated is used to calculate the three-dimensional coordinate of projection X-comers；
The original image of the projection S7. combining angle point three-dimensional coordinate point and projector calculates outside the internal participation of projector
Portion's parameter；
The inside that can obtain projector through above step participates in external parameter.Just triangulation can be utilized
Method calculates the finger tip distance to camera, as shown in Figure 3.Wherein P is observation station, Ol、OrFor camera, T
Being the distance between two cameras, Z is two cameras vertical dimensions to observation station, and f is the focal length of camera, xr
And xlRepresent observation station horizontal coordinate in the picture respectively.Represent two camera shootings respectively
The horizontal coordinate of picture centre.
Utilize similar triangles can be easy to derive Z value, as shown in Figure 3:
Utilize the principle of triangulation, it is also possible to calculate the distance of camera distance projection screen easily, thus
The distance of finger distance screen can be extrapolated, if the distance of finger distance screen is less than a certain specific threshold,
It is considered as click event to occur.By finger tip position in screen, and geometric calibration before, permissible
At cursor of mouse location to finger tip, and analog mouse clicks on event.Realize interpersonal alternately, thus reach
Any one projection plane is become the purpose of a touch screen.
The present invention uses the method for " Bayesian Estimation " to separate arm, and motion tracking system is joined interaction
In optical projection system so that the efficiency of algorithm is improved.One new data training mechanism is proposed again；And
During Face Detection, real-time renewal complexion model so that robustness and the efficiency of algorithm are all greatly improved.Simultaneously
The present invention is by the checking of test of many times, and result shows that this system processing speed is fast, reliable and stable.
Obviously, the above embodiment of the present invention is only for clearly demonstrating example of the present invention, and not
It it is the restriction to embodiments of the present invention.For those of ordinary skill in the field, in described above
On the basis of can also make other changes in different forms.Here without also cannot be to all of enforcement
Mode gives exhaustive.All any amendment, equivalent and improvement made within the spirit and principles in the present invention
Deng, within should be included in the protection domain of the claims in the present invention.
Claims (10)
1. an interaction method based on active vision, projection arrangement arrives needing interactive image projection
Arbitrary plane, and use the operation information of the filming apparatus collection user interactive picture to being projected out, will clap simultaneously
The information taken the photograph is transferred to processing means and is analyzed processing, and realizes man-machine interaction with the operation obtaining user corresponding,
It is characterized in that, the process of described processing means Treatment Analysis information is: select one to illumination-insensitive
Color space, use Bayes' assessment to extract hand region, and this hand region be tracked and finger tip
The location of position, then judge whether finger contacts with projection screen by range finding, man-machine to determine whether to carry out
Alternately；
Using Bayes' assessment to extract hand region, its concrete mode is:
By off-line training acquisition probability P (s), P (c), P (c | s) and P (s | c), wherein P (s) represents training
During the prior probability of the colour of skin, P (c) represents the prior probability of each color during training, P (c | s) table
Show the prior probability that pixel is the colour of skin that pixel value is c, and P (s | c) represent that pixel value is c's after training
Pixel is the probability of the colour of skin；
P (s | c)=P (c | s) P (s)/P (c) (1)
Hysteresis threshold T is obtained during off-line trainingmax, Tmax=P (s | c)；
Adaptive method is used to carry out Face Detection to extract hand region, according to the colour of skin district of nearest w frame identification
Territory judges the area of skin color point detected at present, and the prior probability of nearest w frame is Pw(s), Pw(c), Pw(c | s),
Wherein PwS () represents the prior probability of the colour of skin, P in nearest w framewC () represents each color in nearest w frame
Prior probability, Pw(c | s) represent that in nearest w frame, pixel value is the prior probability that pixel is the colour of skin of c, separately examines
The probability P that survey is area of skin color ' (s | c) be:
P'(s | and c)=γ P (s | c)+(1-γ) Pw(s|c) (2)
Wherein Pw(s | c) obtained by formula (1), i.e. Pw(s | c)=Pw(c|s)Pw(s)/PwC (), γ is one
Control coefrficient, this control coefrficient is relevant with the training set of training stage；
Work as P'(s | c) ＞ Tmax, detect whether the most at present into area of skin color, so far to extract hand region, otherwise
It is not belonging to area of skin color.
Interaction method based on active vision the most according to claim 1, described color space is adopted
Use YUV4:2:2.
Interaction method based on active vision the most according to claim 2, described YUV4:2:2
Color space does not use Y-component.
Interaction method based on active vision the most according to claim 1, described off-line training
Detailed process is: selects one group of a small amount of data to be trained, then uses mass data and hysteresis threshold to enter
The identification of row area of skin color, and real-time renewal prior probability P (s), P (c) and P (c | s), the most constantly
By the hysteresis threshold after updating, the area of skin color in picture and non-area of skin color are separated.
Interaction method based on active vision the most according to claim 1, described to hand region
The implementation being tracked is: use Camshift algorithm, will the result of previous frame, i.e. search window
Size and center, as the initial value of next frame meanshift algorithm search window；With this, iteration continues, it is achieved
Tracking to target.
The most according to claim 5, interaction method based on active vision, described enter hand region
The specific implementation of line trace is:
1) search window is initialized；
2) color probability distribution of search window is calculated；
3) meanShift algorithm is run, it is thus achieved that the size and location of search window；
4) by step 3 in next frame video image) initial value reinitialize size and the position of search window
Put, then jump to step 2) proceed.
Interaction method based on active vision the most according to claim 1, finger tip position, described location
The mode put is:
A) find out largest contours and fill this profile, to have obtained not having noisy arm foreground image；
B) convex closure of profile is calculated；
C) calculate arm position of centre of gravity, convex closure is found out several candidate points of maximum curvature；
D) using candidate point farthest for distance position of centre of gravity as finger tip.
Interaction method based on active vision the most according to claim 1, described range finding is to measure
Finger tip to the distance of camera, concrete mode is: labeling projection instrument and video camera, the wherein scaling method of projector
For:
S1. calibrating camera；
S2. prepare a blank, blank posts a papery gridiron pattern；
S3. control projector projects to go out a gridiron pattern and be radiated on blank；
Extract two tessellated angle points the most respectively；
S5. blank place plane is calculated according to papery X-comers；
S6. the video camera demarcated is used to calculate the three-dimensional coordinate of projection X-comers；
S7. the original image combining the three-dimensional coordinate point of angle point and the projection of projector calculates the internal participation of projector
External parameter；
S8. the method utilizing triangulation calculates finger tip distance Z to camera；When distance Z is less than a certain specific
Threshold value, then it is assumed that have click event to occur.
9. the interaction method based on active vision being applied to described in any one of claim 1 to 8
System, including for the interactive image projection projection arrangement to arbitrary plane will be needed, is used for gathering user
To the filming apparatus of the operation information of the interactive picture being projected out and the processing means that carries out information process analysis,
It is characterized in that, the process of described processing means Treatment Analysis information is: select one to illumination-insensitive
Color space, use Bayes' assessment to extract hand region, and this hand region be tracked and finger tip
The location of position, then judge whether finger contacts with projection screen by range finding, man-machine to determine whether to carry out
Alternately；
Using Bayes' assessment to extract hand region, its concrete mode is:
By off-line training acquisition probability P (s), P (c), P (c | s) and P (s | c), wherein P (s) represents training
During the prior probability of the colour of skin, P (c) represents the prior probability of each color during training, P (c | s) table
Show the prior probability that pixel is the colour of skin that pixel value is c, and P (s | c) represent that pixel value is c's after training
Pixel is the probability of the colour of skin；
P (s | c)=P (c | s) P (s)/P (c) (1)
Hysteresis threshold T is obtained during off-line trainingmax, Tmax=P (s | c)；
Adaptive method is used to carry out Face Detection to extract hand region, according to the colour of skin district of nearest w frame identification
Territory judges the area of skin color point detected at present, and the prior probability of nearest w frame is Pw(s), Pw(c), Pw(c | s),
Wherein PwS () represents the prior probability of the colour of skin, P in nearest w framewC () represents each color in nearest w frame
Prior probability, Pw(c | s) represent that in nearest w frame, pixel value is the prior probability that pixel is the colour of skin of c, separately examines
The probability P that survey is area of skin color ' (s | c) be:
P'(s | and c)=γ P (s | c)+(1-γ) Pw(s|c) (2)
Wherein Pw(s | c) obtained by formula (1), i.e. Pw(s | c)=Pw(c|s)Pw(s)/PwC (), γ is one
Control coefrficient, this control coefrficient is relevant with the training set of training stage；
Work as P'(s | c) ＞ Tmax, detect whether the most at present into area of skin color, so far to extract hand region, otherwise
It is not belonging to area of skin color.
System the most according to claim 9, it is characterised in that described filming apparatus is two cameras.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310724304.8A CN103677274B (en) | 2013-12-24 | 2013-12-24 | A kind of interaction method and system based on active vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310724304.8A CN103677274B (en) | 2013-12-24 | 2013-12-24 | A kind of interaction method and system based on active vision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103677274A CN103677274A (en) | 2014-03-26 |
| CN103677274B true CN103677274B (en) | 2016-08-24 |
Family
ID=50315082
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310724304.8A Active CN103677274B (en) | 2013-12-24 | 2013-12-24 | A kind of interaction method and system based on active vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103677274B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104978012B (en) | 2014-04-03 | 2018-03-16 | 华为技术有限公司 | One kind points to exchange method, apparatus and system |
| CN104202547B (en) * | 2014-08-27 | 2017-10-10 | 广东威创视讯科技股份有限公司 | Method, projection interactive approach and its system of target object are extracted in projected picture |
| CN104331158B (en) * | 2014-10-29 | 2018-05-25 | 山东大学 | The man-machine interaction method and device of a kind of gesture control |
| CN104778460B (en) * | 2015-04-23 | 2018-05-04 | 福州大学 | A kind of monocular gesture identification method under complex background and illumination |
| CN105371784A (en) * | 2015-12-24 | 2016-03-02 | 吉林大学 | Machine vision based holographic man-machine interaction system for automotive inspection |
| CN105869166B (en) * | 2016-03-29 | 2018-07-10 | 北方工业大学 | A kind of human motion recognition method and system based on binocular vision |
| CN106204604B (en) * | 2016-04-29 | 2019-04-02 | 北京仁光科技有限公司 | Project touch control display apparatus and its exchange method |
| CN107343184A (en) * | 2016-05-03 | 2017-11-10 | 中兴通讯股份有限公司 | Projecting apparatus processing method, device and terminal |
| CN106991417A (en) * | 2017-04-25 | 2017-07-28 | 华南理工大学 | A kind of visual projection's interactive system and exchange method based on pattern-recognition |
| JP6898021B2 (en) * | 2018-03-07 | 2021-07-07 | Ｎｅｃソリューションイノベータ株式会社 | Operation input device, operation input method, and program |
| CN109190357B (en) * | 2018-08-30 | 2021-08-06 | 袁精侠 | Gesture verification code implementation method for man-machine verification by only utilizing cache resources |
| CN109410274B (en) * | 2018-10-08 | 2022-03-15 | 武汉工程大学 | Method for positioning typical non-cooperative target key points in real time under high frame frequency condition |
| CN109934105B (en) * | 2019-01-30 | 2022-12-16 | 华南理工大学 | Virtual elevator interaction system and method based on deep learning |
| CN109683719B (en) * | 2019-01-30 | 2021-10-22 | 华南理工大学 | Visual projection interaction method based on YOLOv3 |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101430760A (en) * | 2008-11-18 | 2009-05-13 | 北方工业大学 | Human face super-resolution processing method based on linear and Bayesian probability mixed model |
| CN102298694A (en) * | 2011-06-21 | 2011-12-28 | 广东爱科数字科技有限公司 | Man-machine interaction identification system applied to remote information service |
-
2013
- 2013-12-24 CN CN201310724304.8A patent/CN103677274B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101430760A (en) * | 2008-11-18 | 2009-05-13 | 北方工业大学 | Human face super-resolution processing method based on linear and Bayesian probability mixed model |
| CN102298694A (en) * | 2011-06-21 | 2011-12-28 | 广东爱科数字科技有限公司 | Man-machine interaction identification system applied to remote information service |
Non-Patent Citations (3)
| Title |
|---|
| A Bayesian skin/non-skin color classifier using non-parametric density estimation;Chai D，etc;《proceedings of the 2003 international symposium on》;20030628;全文 * |
| 一种肤色干扰下的变形手势跟踪方法;刘玉进等;《计算机工程与应用》;20091211(第35期);全文 * |
| 基于机器视觉的运动目标检测技术的研究;王兴杰;《中国优秀硕士学位论文全文数据库》;20130715;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103677274A (en) | 2014-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103677274B (en) | A kind of interaction method and system based on active vision | |
| US8401248B1 (en) | Method and system for measuring emotional and attentional response to dynamic digital media content | |
| CN103941866B (en) | Three-dimensional gesture recognizing method based on Kinect depth image | |
| CN104639840B (en) | Image processing apparatus and image processing method | |
| CN108446585A (en) | Method for tracking target, device, computer equipment and storage medium | |
| CN109684925B (en) | Depth image-based human face living body detection method and device | |
| CN111539273A (en) | Traffic video background modeling method and system | |
| US20120287122A1 (en) | Virtual apparel fitting system and method | |
| Niu et al. | Tactic analysis based on real-world ball trajectory in soccer video | |
| CN102831382A (en) | Face tracking apparatus and method | |
| Tan et al. | Dynamic hand gesture recognition using motion trajectories and key frames | |
| KR20110013200A (en) | Identifying method of human attitude and apparatus of the same | |
| CN108198201A (en) | A kind of multi-object tracking method, terminal device and storage medium | |
| CN103598870A (en) | Optometry method based on depth-image gesture recognition | |
| CN102831439A (en) | Gesture tracking method and gesture tracking system | |
| CN104202547A (en) | Method for extracting target object in projection picture, projection interaction method and system thereof | |
| CN109145803A (en) | Gesture identification method and device, electronic equipment, computer readable storage medium | |
| CN104821010A (en) | Binocular-vision-based real-time extraction method and system for three-dimensional hand information | |
| CN111161313A (en) | Multi-target tracking method and device in video stream | |
| CN109934127A (en) | Pedestrian's recognition and tracking method based on video image and wireless signal | |
| CN111400423B (en) | Smart city CIM three-dimensional vehicle pose modeling system based on multi-view geometry | |
| CN113435336A (en) | Running intelligent timing system and method based on artificial intelligence | |
| EP4136570A1 (en) | Artificial intelligence and computer vision powered driving-performance assessment | |
| CN110517285B (en) | Large-scene minimum target tracking based on motion estimation ME-CNN network | |
| CN103761011A (en) | Method, system and computing device of virtual touch screen |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CP03 | Change of name, title or address |
Address after: Kezhu road high tech Industrial Development Zone, Guangzhou city of Guangdong Province, No. 233 510670 Patentee after: Wei Chong group Limited by Share Ltd Address before: 510663 No. 6, color road, hi tech Industrial Development Zone, Guangdong, Guangzhou, China Patentee before: Guangdong Weichuangshixun Science and Technology Co., Ltd. |
|
| CP03 | Change of name, title or address |