A method and device for recognizing dynamic gestures of pilots

Abstract

The present application provides a method and device for recognizing dynamic gestures of pilots, belonging to the technical field of aviation human-computer interaction. The method includes: receiving angle data of a dynamic gesture to be recognized sent by a data glove. The pilot wears the data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect angle data of finger joints at a preset frequency; detecting an instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture stream surface model according to the angle data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture stream surface model, determining a starting time point of the instantaneous gesture posture; determining a gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture. It has a relatively fast operation speed, high recognition accuracy and is easy to expand, and can recognize gestures in real time for recognizing dynamic gestures of pilots.

Description

Technical Field

[0001] This application belongs to the technical field of aviation human-computer interaction, and particularly relates to a method and device for recognizing dynamic gestures of pilots. Background Art

[0002] Aircraft operation is a highly complex task. From entering the cockpit to taking off, a pilot has to go through steps such as turning on the power and connecting to the ground power supply, starting the auxiliary power unit, making overhead panel settings, checking the navigation database, setting the route, configuring takeoff parameters, optimizing the fuel system, making the final corrections before taxiing and takeoff, and taking off. Each of these steps involves several to dozens of operations. Currently, pilots basically operate the aircraft through direct operations on the physical interfaces of the front panel and overhead panel of the aircraft cockpit. In the research and development of future aircraft cockpit interfaces, voice interaction, gesture interaction, and eye movement interaction have become hot research issues. Gesture is one of the main modalities for human information communication, which can originate from any body movement or state, such as hand gestures, foot gestures, and facial gestures. The main purpose of gesture recognition is to recognize human gestures through mathematical algorithms. It provides a richer modality than traditional text recognition and graphic recognition and is an important way for computers to understand human language. With the rapid development of human-computer interaction technology and artificial intelligence technology, gesture recognition technology has received high attention in many research and application fields.

[0003] The aircraft cockpit, as a special operating environment, poses severe challenges to existing gesture recognition technologies. The cockpit environment is relatively narrow, and the light changes greatly during actual flight operations, severely restricting the application of gesture recognition technologies based on optical sensing. Moreover, pilots have many actual task scenarios, and many tasks require pilots to use their hands for fine operations. Therefore, gesture recognition technology with real-time performance, accuracy, and easy expandability is the key technology for implementing pilot gesture recognition in the aircraft cockpit. Summary of the Invention

[0004] To solve the problem of pilot gesture recognition in related technologies, this application provides a method and device for recognizing dynamic gestures of pilots. The technical solutions are as follows:

[0005] In a first aspect, a method for recognizing dynamic gestures of pilots is provided. The method includes:

[0006] Receiving angle data of a dynamic gesture to be recognized sent by a data glove. The pilot wears a data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect angle data of finger joints at a preset frequency;

[0007] Detect the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture stream surface model according to the angle data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture stream surface model, determine the starting time point of the instantaneous gesture posture;

[0008] Determine the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture.

[0009] Further, the method further includes:

[0010] Receive the angle data of each specified dynamic gesture sent by the data glove;

[0011] Establish a gesture stream surface model corresponding to the specified dynamic gesture based on the angle data of each specified dynamic gesture.

[0012] Among them, the establishing a gesture stream surface model corresponding to the specified dynamic gesture based on the angle data of each specified dynamic gesture includes:

[0013] Establish a three-dimensional coordinate system, set the x-axis in the three-dimensional coordinate system as time, the y-axis as the angle data number, and the z-axis as the magnitude of the angle data. The x-axis indicates the dimension direction, and the y-axis indicates the longitude direction;

[0014] Take the established three-dimensional coordinate system as the gesture stream surface model of the specified dynamic gesture.

[0015] Further, after receiving the angle data of each specified dynamic gesture sent by the data glove, the method further includes:

[0016] Perform normalization processing on the angle data of each specified dynamic gesture.

[0017] Among them, the detecting the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture stream surface model according to the angle data of the dynamic gesture to be recognized, and when the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture stream surface model, determining the starting time point of the instantaneous gesture posture includes:

[0018] For each gesture stream surface model:

[0019] Use the mean shift algorithm to estimate the position of the low-frequency band interval in this gesture stream surface model, and determine the estimated position as the convergence position of the mean shift point. The range of the low-frequency band is 0-100 Hz;

[0020] Use the sliding time window algorithm to search forward along the positive x-axis from the convergence position of the mean shift point to find the starting point of the low-frequency band interval;

[0021] Take the starting point of the low-frequency band interval as the starting time point of the instantaneous gesture posture.

[0022] Among them, determining the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture includes:

[0023] Determine a time interval with a low-frequency band interval, and the starting point of the time interval is the starting time point of the instantaneous gesture posture;

[0024] Determine the angle data number within this time interval;

[0025] Input the angle data number within this time interval into the classifier, and determine the output of the classifier as the gesture type of the dynamic gesture to be recognized.

[0026] Furthermore, the method further includes:

[0027] Train the classifier according to the angle data numbers of each specified dynamic gesture. The input of the classifier is the angle data number of the dynamic gesture, and the output is the gesture type of the dynamic gesture.

[0028] Among them, the classifier records the previously recognized dynamic gesture, and the method further includes:

[0029] When the currently recognized dynamic gesture is the same as the previously recognized dynamic gesture recorded by the classifier, it is determined that no new gesture dynamics are detected, and no output operation is performed;

[0030] When the currently recognized dynamic gesture is different from the previously recognized dynamic gesture recorded by the classifier, it is determined that the currently recognized dynamic gesture is a new dynamic gesture, and the output operation is performed.

[0031] In a second aspect, a pilot dynamic gesture recognition device is provided, and the device includes:

[0032] A receiving module, configured to receive the angle data of the dynamic gesture to be recognized sent by a data glove. The pilot wears a data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect the angle data of the finger joints at a preset frequency;

[0033] A detection module, configured to detect the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture stream surface model according to the angle data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture stream surface model, determine the starting time point of the instantaneous gesture posture;

[0034] A determination module, configured to determine the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture.

[0035] According to the process of the pilot making gestures, considering the aspects of computing speed, real-time performance, and computing accuracy, the present application proposes a method and device for recognizing dynamic gestures based on instantaneous gesture postures. The present application uses a data glove as a gesture input device, which is equipped with sensors that can record the movement angles of different joints. The proposed gesture flow surface model is used to describe the input gesture data stream. The mean shift algorithm is used to find the starting time point of the instantaneous gesture posture, and based on this, the target gesture is further detected. It has a fast operation speed, high recognition accuracy, and is easy to expand, and can accurately recognize dynamic gestures in real time. Description of the Drawings

[0036] Figure 1 It is a flowchart of a method for recognizing dynamic gestures provided by the present application;

[0037] Figure 2 It is a flowchart of real-time instantaneous posture detection provided by the present application. Detailed Description of the Embodiment

[0038] The present application will be further described in detail below through specific embodiments and the accompanying drawings.

[0039] Step 110: Receive the angle data of the dynamic gesture to be recognized sent by the data glove.

[0040] The pilot wears a data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect the angle data of the finger joints at a preset frequency;

[0041] Step 120: Detect the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture flow surface model according to the angle data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture flow surface model, determine the starting time point of the instantaneous gesture posture;

[0042] Step 130: Determine the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture.

[0043] According to the process of the pilot making gestures, considering the aspects of computing speed, real-time performance, and computing accuracy, the present application proposes a method for recognizing dynamic gestures based on instantaneous gesture postures. Using a data glove as a gesture input device, which is equipped with sensors that can record the movement angles of different joints. The proposed gesture flow surface model is used to describe the input gesture data stream, find the starting time point of the instantaneous gesture posture, and based on this, the target gesture is further detected. It has a fast operation speed, high recognition accuracy, and is easy to expand, and can accurately recognize dynamic gestures in real time.

[0044] The present application provides another method for recognizing the pilot's dynamic gestures, and the method includes:

[0045] Step 210: Receive the angular data of each specified dynamic gesture sent by the data glove.

[0046] For example, as shown in Table 1, each specified dynamic gesture may include gesture types such as the number 1, OK, Good, etc.

[0047] Table 1 Specified Dynamic Gestures

[0048]

[0049] Step 220: Based on the angular data of each specified dynamic gesture, establish a gesture flow surface model corresponding to the specified dynamic gesture.

[0050] In this embodiment, after receiving the angular data of each specified dynamic gesture sent by the data glove, the angular data of each specified dynamic gesture may first be normalized. Then, based on the normalized angular data, a gesture flow surface model corresponding to the specified dynamic gesture is established. In this way, the recognition efficiency is higher and the recognition accuracy is higher.

[0051] In Step 220, establishing a gesture flow surface model corresponding to the specified dynamic gesture based on the angular data of each specified dynamic gesture includes:

[0052] Establish a three-dimensional coordinate system, set the x-axis in the three-dimensional coordinate system as time, the y-axis as the angular data number, and the z-axis as the magnitude of the angular data. The x-axis indicates the dimension direction, and the y-axis indicates the longitude direction;

[0053] Take the established three-dimensional coordinate system as the gesture flow surface model of the specified dynamic gesture.

[0054] It should be noted that Steps 210 to 220 only need to be executed once. After completion, they can be used every time a dynamic gesture is recognized subsequently.

[0055] Step 230: Receive the angular data of the dynamic gesture to be recognized sent by the data glove.

[0056] The pilot wears a data glove, and the data glove is provided with multiple sensors, and each sensor is used to collect the angular data of finger joints at a preset frequency.

[0057] Step 240: Detect the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture flow surface model according to the angular data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture flow surface model, determine the starting time point of the instantaneous gesture posture.

[0058] Step 240 may specifically include:

[0059] For each gesture flow surface model:

[0060] 1) Estimate the position of the low-frequency band interval in the gesture flow surface model using the mean shift algorithm, and determine the estimated position as the convergence position of the mean shift point. The range of the low-frequency band is 0 - 100 Hz.

[0061] Among them, the mean shift algorithm specifically includes the following content:

[0062] Define y l (t k ) ∈ [0, 1] as the sample feature point, l is the feature number (a total of N), t is the time, and k represents the current time point.

[0063] Roughly find the position of the low-frequency interval using the mean shift algorithm. Assume the current position is t k , and the next position drifted to from t k is denoted as m(t k ). The calculation formula of m(t k ) is:

[0064]

[0065] Among them, M 0.5 = [0.5m], m is the pre-defined mean shift window length, and w(t k ) is the weight corresponding to the time point t k .

[0066] In formula (1), use an empirical function with the derivative of the sample feature point with respect to latitude y i′ (t k ) as the independent variable as the weight function. First, calculate the derivative of each sample feature point with respect to latitude using the least squares method. The calculation formula is:

[0067]

[0068] Among them, n 0.5 = [0.5n], n is the pre-set length, l = 1…N, and N is the number of features.

[0069] After obtaining the derivative of each sample feature point with respect to latitude, calculate the weight of each time point. Define the indicator function I(x):

[0070]

[0071] Then the weight function w is defined as:

[0072]

[0073] Among them, N is the number of features.

[0074] It can be seen from the definition of the weight function that the smaller the weight function, the "flatter" the gesture flow surface at the corresponding time point. After moving to the next position, mean shift is performed again until the shift point converges (this position is denoted as t m ). Finally, a density function d(t m ) is calculated to describe the degree of "flatness".

[0075]

[0076] where M 0.25 = [0.25M], and M is the length of the mean shift window defined in advance.

[0077] If d(t m ) ≥ D, where D is a threshold set in advance, it is considered that the shift point has roughly moved to the position in the low-frequency band interval.

[0078] Take out N×l C eigenvalues, divide them into N groups, sort the data in each group, and then calculate:

[0079]

[0080] where l C = 1, 2, 3, …, N. The eigenvector C = (c1, c2, …, c N ) is obtained T .

[0081] Define the above method as a function E, with the independent variable being a certain time interval and the dependent variable being an N-dimensional eigenvector, that is: E(l c (t k )) = C.

[0082] If d(t m ) ≤ D, force the shift point to move forward and restart the iterative process of shift point movement.

[0083] In this embodiment, the gesture flow surface model needs to be Fourier-transformed first, and then 0 - 100 Hz is determined as the low-frequency band. The aforementioned gesture flow surface model is drawn based on time-domain signals, and the Fourier transform of common time-domain signals is as follows:

[0084]

[0085] where f(t) is the time-domain signal value, F(ω) is the frequency-domain signal value, ω represents frequency, and t represents time.

[0086] 2) Use the sliding time window algorithm to search forward along the positive x-axis from the mean shift point convergence position to find the starting point of the low-frequency band interval;

[0087] The above-mentioned sliding time window algorithm specifically includes:

[0088] Set the sliding time window as W(t k ), and define it as an interval with a length of l w . The sliding time window is defined with 2 attributes: Dens i , Dev i . Define the attribute of the sliding time window W(t k ) as a Cartesian product, that is, (Dens i , Dev i ).

[0089] Use the derivative of each sample feature point to calculate Dens i . First, according to the aforementioned mean shift algorithm, use the least squares method to calculate the derivatives of N sample feature points at a time point: y i′ (t k ), l = 1…N.

[0090] Then, calculate Dens i :

[0091]

[0092] Among them, l = 1, 2, …, N, and I(y' l (t k )) is the aforementioned indicator function.

[0093] When the pilot's gesture has no movement, Dens i = 0; when there is single finger joint movement, Dens i = 1. It is possible to judge whether the window W(t k ) is located in the low-frequency band interval according to Densi.

[0094] Then, introduce an attribute Dev i to improve the accuracy of detecting t a .

[0095]

[0096] Among them, x i is the normalized real-time collected data value, c i is the corresponding eigenvalue, and I1(x i , c i ) is the indicator function, which is specifically expressed as follows:

[0097]

[0098] When there are obvious changes in the finger joints, Dev i= 1, otherwise Dev i = 0.

[0099] After this algorithm starts, slide the window W forward from the point where drift converges. When sliding the window W forward, when the attribute Dev i or Dev i is equal to 1, it is considered that the window W has moved to the starting point t a of the low-frequency interval segment. At this time, moving forward a certain distance can obtain the time interval T = [t a , t b , where t b is the termination moment. Calculate the average number of sample feature points that satisfy the derivative y l′ (t k ) < 0.005 within this interval, and denote it as N LFP .

[0100]

[0101] Among them, N is the number of features.

[0102] If N LFP is greater than the pre-set threshold, which is set to 70% in this operation example, it is considered that the interval T contains the low-frequency interval segment. If this is the first time to reach this state after the end of the previous gesture, in this operation example, it means that the data stream in the corresponding window fluctuates, then it is considered that the previous gesture has ended. Then calculate R = E(T), output the feature vector R to the classifier for recognition, otherwise it is considered that the current gesture has not ended, no output is made, and the mean shift algorithm is re-executed; if N LFP is less than the current set threshold, the drift point is forced to move backward, and a new drift is performed to re-execute the mean shift algorithm.

[0103] 3) Use the starting point of the low-frequency segment interval as the starting time point of the instantaneous gesture posture.

[0104] Step 250, determine the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture.

[0105] In step 250, determining the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture includes:

[0106] Determine a time interval where there is a low-frequency segment interval, and the starting point of the time interval is the starting time point of the instantaneous gesture posture;

[0107] Determine the angle data number within this time interval;

[0108] Input the angle data number within this time interval into the classifier, and determine the output of the classifier as the gesture type of the dynamic gesture to be recognized.

[0109] Furthermore, the method further includes:

[0110] Training a classifier based on the angle data numbers of each specified dynamic gesture, where the input of the classifier is the angle data number of the dynamic gesture and the output is the gesture type of the dynamic gesture.

[0111] Optionally, the classifier records the previously recognized dynamic gesture, and the method may further include:

[0112] When the currently recognized dynamic gesture is the same as the previously recognized dynamic gesture recorded by the classifier, it is determined that no new gesture dynamics are detected, and the output operation is not performed;

[0113] When the currently recognized dynamic gesture is different from the previously recognized dynamic gesture recorded by the classifier, it is determined that the currently recognized dynamic gesture is a new dynamic gesture, and the output operation is performed.

[0114] Based on the process of the pilot making gestures, considering the computing speed, real-time performance, and computing accuracy, this application proposes a method for recognizing dynamic gestures based on instantaneous gesture postures. This application uses a data glove as the gesture input device, which is equipped with sensors that can record the movement angles of different joints. The proposed gesture flow surface model is used to describe the input gesture data stream. The mean shift algorithm is used to find the starting time point of the instantaneous gesture posture, and this is used to further detect the target gesture. It has a fast operation speed, high recognition accuracy, and is easy to expand, and can accurately recognize dynamic gestures in real time.

[0115] This application provides a pilot dynamic gesture recognition device, and the device includes:

[0116] A receiving module, configured to receive the angle data of the dynamic gesture to be recognized sent by the data glove. The pilot wears the data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect the angle data of the finger joints at a preset frequency;

[0117] A detection module, configured to detect the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture flow surface model according to the angle data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture flow surface model, the starting time point of the instantaneous gesture posture is determined;

[0118] A determination module, configured to determine the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture.

[0119] The recognition device of dynamic gestures based on instantaneous gesture postures proposed in this application uses a data glove as a gesture input device. Sensors are installed on it to record the movement angles of different joints. The proposed gesture flow surface model is used to describe the input gesture data stream, find the starting time point of the instantaneous gesture posture, and further detect the target gesture based on this. It has a fast operation speed, high recognition accuracy, and is easy to expand, and can accurately recognize dynamic gestures in real time.

[0120] During the gesture recognition process of this application, high recognition rate and high-efficiency recognition are achieved. For the proposed mean shift algorithm, the distance of drift is dynamically determined by the data within the window first, and the data segments of moving gestures can be skipped. When the drift converges, the "flatness" of the data stream within the surrounding time interval (i.e., verifying the existence of the instantaneous posture) will be verified. Only when both verifications pass, the data will be output to the classifier for recognition. Since the data segments of noise and moving gestures can be skipped during gesture detection, the recognition rate is improved. At the same time, the window length of the mean shift is smaller than that of the sliding time window, the time delay is smaller, and the window moves dynamically, thus improving the recognition efficiency.

[0121] The above only expresses the implementation modes of this application, and its description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application.

Claims

1. A method for recognizing dynamic gestures of pilots, characterized in that, The method includes: Receiving the angular data of the dynamic gesture to be recognized sent by a data glove. The pilot wears the data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect the angular data of finger joints at a preset frequency; Detecting the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture flow surface model according to the angular data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture flow surface model, determining the starting time point of the instantaneous gesture posture; Determining the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture; The step of detecting the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture flow surface model according to the angular data of the dynamic gesture to be recognized, and when the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture flow surface model, determining the starting time point of the instantaneous gesture posture includes: For each gesture flow surface model: Using the mean shift algorithm to estimate the position of the low-frequency band interval in the gesture flow surface model, and determining the estimated position as the convergence position of the mean shift point. The range of the low-frequency band is 0-100 Hz; Using the sliding time window algorithm, searching forward along the positive x-axis direction from the convergence position of the mean shift point to find the starting point of the low-frequency band interval; Taking the starting point of the low-frequency band interval as the starting time point of the instantaneous gesture posture; The step of determining the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture includes: Determining a time interval where there is a low-frequency band interval, and the starting point of the time interval is the starting time point of the instantaneous gesture posture; Determining the angular data numbers within this time interval; Inputting the angular data numbers within this time interval into a classifier, and determining the output of the classifier as the gesture type of the dynamic gesture to be recognized.

2. The method according to claim 1, characterized in that, The method further includes: Receiving the angular data of each specified dynamic gesture sent by the data glove; Establishing a gesture flow surface model corresponding to each specified dynamic gesture based on the angular data of each specified dynamic gesture.

3. The method according to claim 2, characterized in that, The step of establishing a gesture flow surface model corresponding to each specified dynamic gesture based on the angular data of each specified dynamic gesture includes: Establishing a three-dimensional coordinate system, setting the x-axis in the three-dimensional coordinate system as time, the y-axis as the angular data number, and the z-axis as the magnitude of the angular data. The x-axis indicates the dimension direction, and the y-axis indicates the longitude direction; Taking the established three-dimensional coordinate system as the gesture flow surface model of the specified dynamic gesture.

4. The method according to claim 2, characterized in that, After receiving the angular data of each specified dynamic gesture sent by the data glove, the method further includes: Performing normalization processing on the angular data of each specified dynamic gesture.

5. The method according to claim 1, characterized in that, The method further includes: Training a classifier according to the angular data numbers of each specified dynamic gesture. The input of the classifier is the angular data number of the dynamic gesture, and the output is the gesture type of the dynamic gesture.

6. The method according to claim 1 or 5, characterized in that, The classifier records the previously recognized dynamic gesture. The method further includes: When the currently recognized dynamic gesture is the same as the previously recognized dynamic gesture recorded by the classifier, determining that no new gesture dynamics are detected and not performing the output operation; When the currently recognized dynamic gesture is different from the previously recognized dynamic gesture recorded by the classifier, determine that the currently recognized dynamic gesture is a new dynamic gesture and perform the output operation.

7. A device for recognizing dynamic gestures of pilots, characterized in that, For the pilot dynamic gesture recognition method described in claim 1, the device includes: A receiving module, configured to receive the angle data of the dynamic gesture to be recognized sent by the data glove. The pilot wears the data glove on the hand, and the data glove is provided with multiple sensors, and each sensor is used to collect the angle data of the finger joints at a preset frequency; A detection module, configured to detect the instantaneous gesture posture of the dynamic gesture to be recognized in each pre-established gesture stream surface model according to the angle data of the dynamic gesture to be recognized. When the instantaneous gesture posture of the dynamic gesture to be recognized is detected in a gesture stream surface model, determine the starting time point of the instantaneous gesture posture; A determination module, configured to determine the gesture type of the dynamic gesture to be recognized according to the starting time point of the instantaneous gesture posture.

Images