CN106055106A - Leap Motion-based advantage point detection and identification method - Google Patents

Abstract

The invention provides a Leap Motion-based advantage point detection and identification method, and belongs to the field of human-computer interaction of computer systems. The method comprises the steps of firstly obtaining a gesture vertex, and building a gesture library to validate an improved non-parameter control advantage point detection algorithm; giving all points between a first point and a tail point and connecting the first point and the tail point to form a line segment; finding the point furthest to the line segment from the given points and judging whether the point is greater than epsilon or not, if so, reserving the point, and or else, rejecting the point; and repeating the method and finally obtaining the optimized broken line. The parameter epsilon is obtained by self-adaption of the improved non-parameter control advantage point detection algorithm. The algorithm is applied to a Leap Motion motion-sensing controller and more gestures can be expanded. The Leap Motion-based advantage point detection and identification method has good self-adaptivity and accuracy on gesture identification; the gestures can be more accurately identified; and the method has extensive applications in human-computer interaction application.

Description

A kind of Dominant point detection recognition methods based on Leap Motion

Technical field

A kind of Dominant point detection recognition methods based on Leap Motion belongs to the field of human-computer interaction of computer system.

Background technology

Along with human-computer interaction technology application is constantly brought forth new ideas and develops, virtual reality technology greatly strengthens Museum Exhibit The form of expression and interaction capabilities, wherein the application with Leap motion is relatively broad, but the gesture used is only limitted to Leap Gesture in Motion kit.Therefore to solve this problem, need gesture summit is preferably optimized, use advantage The method of some detection obtains more accurate summit.

Dominant point detection algorithm is studied by many scholars in recent years, and Dominant point detection algorithm has a lot, according to making Method can be divided into dynamic programming, split-run, combined method, digital line method, breakpoint compression method and calculate curvature Method.All use a preset parameter ε to carry out the choice at control point in these algorithms, be referred to as state modulator Dominant point detection and calculate Method, determines the choice of this point by the side-play amount judging current point and 2 line segments being linked to be of head and the tail.But state modulator is excellent Gesture point detection algorithm needs to set in advance parameter, and does not has the most adaptive for the parameter that the line segment of different length is fixing Ying Xing.

Summary of the invention

Therefore according to problem above, the present invention proposes the Dominant point detection algorithm that a kind of nonparametric controls, it is therefore intended that will The present invention is used in the middle of the Virtual Antique gesture guide to visitors in museum, accurately identifies the gesture of operator, improves man-machine interaction.

To achieve these goals, this invention takes following technical scheme:

One, gesture summit is obtained.Use Leap Motion to obtain hand images, the hand the five fingers or any one finger are regarded as Summit also records data.

Two, the nonparametric control Dominant point detection algorithm that gesture library is improved is set up with checking.Set up vertical fold line, water respectively Flat-folded line and square gesture.

Three, the nonparametric using a kind of improvement controls Dominant point detection algorithm and carries out gesture summit accepting or rejecting optimization.First, The institute being given between first point and tail point a little, and connects first point and tail point, forms line segment.Then, find in the point be given Find from this line segment point furthest, it is judged that whether this point, more than ε, if setting up, retaining this point, otherwise casting out.Repeat This method, and finally give the broken line after optimization.Wherein parameter ε is obtained by the nonparametric control Dominant point detection algorithm self adaptation improved Out.

Four, the summit of optimization is coupled together mate with the gesture in gesture library, complete to identify.

Compared with existing detection method, present invention have the advantage that (1) nonparametric controls Dominant point detection algorithm pair In complex gesture, there is more preferable resolution；(2) are due to the property of there are differences between operator, nonparametric as shown in drawings Control Dominant point detection algorithm can more gesture yardstick be identified, and recognition effect is accurate；(3) nonparametric controls excellent The noise that gesture point detection algorithm produces for equipment all has good robustness under different gesture yardsticks.As shown in drawings.

Accompanying drawing explanation

Fig. 1 system flow chart.

Fig. 2 vertical fold line gesture identification schematic diagram.

Fig. 3 horizontal fold line gesture identification schematic diagram.

Fig. 4 square gesture identification schematic diagram.

The nonparametric that Fig. 5 improves controls the denoising schematic diagram of Dominant point detection algorithm.

Detailed description of the invention

3 the invention will be further described below in conjunction with the accompanying drawings:

1. obtain summit

Obtained the hand images of each frame by Leap Motion motion sensing control device, it can identify the most accurately and sell Refer to towards and the normal vector of hand so that each details of hand perfectly presents.We choose in each frame The mean place of finger tips as gesture summit.

2. summit processes

First dynamic gesture being carried out summit processing procedure, all summits recorded due to Leap Motion are whole Needed to be manually placed in the middle of the identification region of Leap Motion before doing gesture for gesture summit, such as operator, moving The parameter value ε that the summit produced during Dong calculates then needs to be cast out.The most rule of thumb we need front 2 tops Point is cast out.So, after removing front 2 gesture summits, the calculating of parameter can be more efficient.Now summit is referred to as { P₁, P₂,…P_k}(k∈N⁺)

3. parameter calculates

Connect P₁P_kObtain line segment l, computed range s=| P₁P_k|, and slope m, and calculate ε.Concrete grammar is as follows:

Nonparametric RDP detection algorithm key issue to be solved based on Leap Motion is automatically to choose suitably Parameter value ε.We carry out the change of adaptive control parameter according to the angle between length, slope and the line segment of each line segment Change.And according to image processing

X'=round (x)；Y'=round (y) (1)

X, y are respectively abscissa and the vertical coordinate on summit, round it, use P ' (x ', y ') approximate true summit P (x, y)。

x’,y’∈Z (2)

Z is shaping real number.

X'=x+ Δ x；Y'=y+ Δ y (3)

-0.5≤Δx≤0.5,-0.5≤Δy≤0.5 (4)

Δx_k,Δy_kIt is respectively summit P_kAbscissa and the side-play amount of vertical coordinate.Definition P₁P₂Slope be m, P₁′P₂′ Slope be m '.:

The sine value calculated by original point is used to estimate with by the angular difference approximating the sine value that point calculates Error:

(6) formula is substituted in (7) and can obtain:

Assume that s and t is as follows:

Substitute into (8) to obtain:

From formula (4), the maximum of Δ x+ Δ y is 1；

From formula (9), | (x₂-x₁)/s | with | (y₂-y₁)/s | both less than equal to 1, s is always more thanSo Thus draw

N is positive integer: (n ∈ N+)

When | Δ x₂-Δx₁|=| Δ y₂-Δy₁| when=1There is maximum, therefore can obtain t by formula (5) Big value:

Thus obtain maximum

Because t_max≤ 1, formula (14) can be write as:

The upper limit representing rate of increase isDuring the iteration each time of RDP algorithm, utilize formula (14) try to achieve

4. parameter processing

In order to promote the robustness of the nonparametric RDP detection algorithm noise to being produced by Leap Motion, in non-ginseng In the iterative process each time of number RDP algorithms, we add the choice link of parameter value, rule of thumb, if later parameter ε Less than previous parameter ε three/cast out first.Be we have found that by such process and produce by Leap Motion Noise there is good robustness, as shown in red circle in Fig. 6, the part in circle be then intended to detection during need house The part gone because it and be not belonging to the part of gesture.

5. calculate the ultimate range on summit

By P₁P_kAfter drawing line segment l and calculating ε, we calculate P respectively₁To P_kBetween all summits to line segment l away from From, and find summit P farthest for distance line segment l_max, its ultimate range is d_max.If d_max≤ ε, then P_maxCast out, otherwise then protect Stay P_max, and connect P₁P_maxObtain line segment l_(1,max), connect P_maxP_kObtain line segment l_(max,k).

Calculate again according to 3. parameters, 4. parameter processing and 5. three steps of ultimate range calculating summit are iterated Operation, until by all summit { P₁,P₂,…P_kProcessed, it is finally completed the detection on all summits.

6. coupling gesture library

Summit after optimizing is attached according to sequencing, generates gesture graph, and enters with the gesture in gesture library Row coupling.It is shown experimentally that the gesture using nonparametric RDP detection algorithm based on Leap Motion to detect compares RDP Algorithm has more preferable robustness.

Advantage: RDP algorithm has good gesture identification effect, but under the scene that this artificial abortion in museum is big, often One operator suffers from different operating habits, and the gesture size drawn differs.But RDP algorithm be a kind of parameter fix Detection method, does not has the multiple dimensioned property of gesture identification, is not therefore suitable in the scene that this artificial abortion in museum is bigger. RDP algorithm is the detection method that a kind of parameter is fixing, does not has the multiple dimensioned property of gesture identification, proposed a kind of base Nonparametric RDP detection algorithm in Leap Motion has more preferable resolution for complex gesture；Compared to parameter RDP detection algorithm, it has multiple dimensioned property, and the gesture for different operators suffers from more preferable adaptability；Meanwhile, non- Parameter RDP detection algorithm all has good robust for the noise produced by the shake of operator's hand under different gesture yardsticks Property, the highest to the gesture motion code requirement of operator, operate more friendly.

Claims (1)

1. a Dominant point detection recognition methods based on Leap Motion, it is characterised in that comprise the following steps:

1) summit is obtained

Obtained the hand images of each frame by Leap Motion motion sensing control device, choose putting down of the finger tips in each frame All positions are as gesture summit；

2) summit processes

First dynamic gesture being carried out summit processing procedure, after removing front 2 gesture summits, remaining summit is referred to as { P₁, P₂,…P_k}；

3) parameter calculates

Connect P₁P_kObtain the first line segment, computed range s=| P₁P_k|, and slope m, and calculate ε；Specific as follows:

X'=round (x)；Y'=round (y) (1)

X, y are respectively abscissa and the vertical coordinate on summit, in (1) formula round it, use the summit P ' (x ', y ') after rounding near Vraisemblance summit P (x, y)；

x’,y’∈Z (2)

Z is shaping real number；

X'=x+ Δ x；Y'=y+ Δ y (3)

-0.5≤Δx≤0.5,-0.5≤Δy≤0.5 (4)

Δx_k,Δy_kIt is respectively summit P_kAbscissa and the side-play amount of vertical coordinate；Definition P₁P₂Slope be m, P₁′P₂' slope For m '；

$m=tan\mathrm{\φ}=\frac{y_2-y_1}{x_2-x_1}---\left(5\right)$

$\begin{array}{c}{m}^{\′}=\frac{{y_2}^{\′}-{y_1}^{\′}}{{x_2}^{\′}-{x_1}^{\′}}=\\ \left(m+\frac{{\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1}{x_2-x_1}\right)/\left(1+\frac{{\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1}{x_2-x_1}\right)\end{array}---\left(6\right)$

The angular difference using the sine value that calculated by original point and the sine value calculated by approximation point carrys out estimation difference:

$\begin{array}{c}\∂\mathrm{\φ}=|{\tan }^{-1}m-{\tan }^{-1}{m}^{\′}|=\\ \left|{\tan }^{-1}\frac{m-m^{\′}}{1+{\mathrm{mm}}^{\′}}\right|\end{array}---\left(7\right)$

(6) formula is substituted in (7) and obtains:

$\begin{array}{c}\∂\mathrm{\φ}=\left|{\tan }^{-1}\left(\frac{\left(1+\frac{{\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1}{x_2-x_1}\right)m-\left(m+\frac{{\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1}{x_2-x_1}\right)}{\left(1+\frac{{\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1}{x_2-x_1}\right)+m\left(m+\frac{{\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1}{x_2-x_1}\right)}\right)\right|\\ \left|{\tan }^{-1}\left(\frac{\left(\frac{{\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1}{x_2-x_1}\right)m-\left(\frac{{\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1}{x_2-x_1}\right)}{\left(1+m^2\right)+\left(\frac{{\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1}{x_2-x_1}\right)+m\left(\frac{{\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1}{x_2-x_1}\right)}\right)\right|\\ \left|{\tan }^{-1}\left(\frac{m\left({\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1\right)-\left({\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1\right)}{\left(1+m^2\right)\left(x_2-x_1\right)+\left({\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1\right)+m\left({\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1\right)}\right)\right|\end{array}---\left(8\right)$

Assume that s and t is as follows:

$s=\sqrt{{\left(x_2-x_1\right)}^2+{\left(y_2-y_1\right)}^2}---\left(9\right)$

$\begin{array}{c}t=\frac{\left({\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1\right)\left(x_2-x_1\right)}{s^2}+\\ \frac{\left({\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1\right)\left(y_2-y_1\right)}{s^2}\end{array}---\left(10\right)$

Substitute into (8) to obtain:

$\begin{array}{c}\∂\mathrm{\φ}=\\ \left|\begin{array}{c}{\tan }^{-1}\left(\left(\frac{x_2-x_1}{s^2}\right){\left(1+t\right)}^{-1}\right(m\left({\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1\right)-\\ \left({\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1\right)\left)\right)\end{array}\right|\end{array}---\left(11\right)$

From formula (4), the maximum of Δ x+ Δ y is 1；

From formula (9), | x₂-x₁)/s | with | (y₂-y₁)/s | both less than equal to 1, s is always more thanSoThus Go out

$\∂\mathrm{\φ}=\left|\begin{array}{c}{\tan }^{-1}\left(\left(\frac{x_2-x_1}{s^2}\right)\right(m\left({\mathrm{\Δx}}_2-{\mathrm{\Δx}}_1\right)\\ -\left({\mathrm{\Δy}}_2-{\mathrm{\Δy}}_1\right)\left)\right(\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{\left(-t\right)}^n\left)\right)\end{array}\right|---\left(12\right)$

When | Δ x₂-Δx₁|=| Δ y₂-Δy₁| when=1There is maximum, therefore obtain the maximum of t by formula (5):

$t_{\max }=\left(\frac{1}{s}\right)\left(\left|\cos \mathrm{\φ}\right|+\left|\sin \mathrm{\φ}\right|\right)---\left(13\right)$

Thus obtain maximum

$\∂{\mathrm{\φ}}_{\max }=\max \left({\tan }^{-1}\left(\frac{1}{s}\left(|\sin \mathrm{\φ}\±\cos \mathrm{\φ}|\right)\left(\left|\underset{n=0}{\overset{\mathrm{\∞}}{\mathrm{\Σ}}}{\left(-t_{\max }\right)}^n\right|\right)\right)\right)---\left(14\right)$

Because t_max≤ 1, formula (14) is write as:

$\begin{array}{c}\∂{\mathrm{\φ}}_{\max }=\max \left({\tan }^{-1}\right(\frac{1}{s}\left(|\sin \mathrm{\φ}\±\cos \mathrm{\φ}|\right)(1-\\ t_{\max }+t_{\max }^2\left)\right))+O(t_{\max }^3)\end{array}---\left(15\right)$

The upper limit representing rate of increase isDuring the iteration each time of RDP algorithm, try to achieve

4). parameter processing

Iterative process adds the choice link of parameter value each time, if later parameter ε is less than three points of previous parameter ε Cast out first；

5). calculate the ultimate range on summit

By P₁P_kAfter drawing the first line segment and calculating ε, calculate P respectively₁To P_kBetween all summits to the distance of the first line segment, And find the farthest summit P of distance the first line segment_max, its ultimate range is d_max；If d_max≤ ε, then P_maxCast out, otherwise then protect Stay P_max, and connect P₁P_maxObtain the first line segment_(1,max), connect P_maxP_kObtain the first line segment_(max,k)；

Repeat step 3)-5) until by all summit { P₁,P₂,…P_kProcessed, complete the detection on all summits.