TWI701607B - Dynamic face recognition system and method thereof - Google Patents

Abstract

A dynamic face recognition system and method is provided based on facial motion data. The system includes a database, an image capture device, and a processing unit. The database is configured to store a preset facial motion of a user, which including at least one facial feature point. The image capture device tracks the facial motion of the user and captures an approximate video segment, wherein the user makes an approximate facial motion and at least an approximate facial expression in a time frame. The processing unit is coupled to the database and the image capturing device. The processing unit establishes a reference data and an approximate data of the approximate facial motion in the same file establishment process. The processing unit further compares the data with the reference data to verify the approximate facial motion.

Description

Dynamic face recognition system and dynamic face recognition method

The present invention relates to a dynamic face recognition, in particular to a setting of a user's facial motion file used for dynamic face recognition.

Due to its wide range of applications, face recognition has become a conventional solution for identity recognition. The industry has proposed a variety of new methods to improve the facial recognition rate. The prior art facial recognition is purely based on static images. Some systems may claim that they are dynamic, that is, the user can recognize them while they are moving, but their operations are entirely based on real-time captured static images. In essence, these so-called dynamics are still in the static category.

According to an embodiment, a facial recognition system based on facial motion is described. The system includes a database, an image capture device, and a processing unit. The database is used to store a reference file based on a predetermined action made by a user and includes at least one facial feature point. The image capturing device tracks the facial movements of the user and shoots an approximate video segment, wherein the user performs an approximate facial motion and at least one facial expression in a time frame. The processing unit and the database and the image capture device Set phase connection. The processing unit is used to set up the reference file and an approximate file of the approximate facial motion. The processing unit further compares the approximate file with the reference file to verify the approximate facial motion.

According to an embodiment, a method for setting a facial motion file based on a facial motion performed by a user with at least one facial feature point is described. This method includes: setting a reference file containing a preset facial motion through a file setting process, shooting an approximate video clip in a time frame, and in the time frame, the user makes an approximate facial motion and at least A facial expression; the approximate facial motion file is set through the file setting process; the approximate facial motion file is compared with the reference file to verify the approximate facial motion file.

According to an embodiment, a method of dynamic face recognition is as follows. The method includes creating a file of a facial movement of a user for recognition. The file establishment process further includes: extracting at least one facial feature point of the user; tracking and shooting a video clip on a time frame, wherein the video clip contains a plurality of main facial features of the user's facial movements, and Before, after, or before and after the facial action, the user makes at least one facial expression with a plurality of subsidiary facial features; mark the facial feature points on each main and subsidiary facial features; and based on the main and subsidiary facial features Facial features to set up a profile of the facial movement.

The foregoing and other technical content, features, and effects of the present invention will be clearly presented in the following detailed description of the embodiments with reference to the drawings.

100: identification system

110: Image capture device

120: database

130: processing unit

300: Video

310: Preset dynamic facial images

320: preset blank face image

S801~S806, S901~S904: steps

Figure 1 is an example of extracting 68 facial feature points from a user’s face; Figure 2 illustrates a facial recognition system based on facial motions related to the present invention; Figure 3 illustrates a moment A video segment of the frame contains multiple dynamic facial features and facial expressionless features; Figure 4 illustrates 20 facial feature points in time frames t ₁ , t ₁₆ , t _a , t _b , t _N-15 and The change of t _N ; Figure 5 illustrates the relative positions of the facial feature point f ₄₉ in time frames t ₁ , t ₁₆ , t _a , t _b , t _N-15 and t _N ; Figure 6 illustrates the face A histogram of the facial feature point f ₄₉ ; Figure 7 illustrates a smile-like facial feature and other facial expressions; Figure 8 illustrates a flow chart of dynamic facial recognition based on the present invention; and Figure 9 illustrates a flow chart of setting up a facial motion file for facial recognition based on the present invention.

The directional terms mentioned in the following embodiments, for example: up, down, left, right, front or back, etc., refer to the directions of the attached drawings. Therefore, the directional terms used are used to illustrate but not to limit the present invention.

This application is related to Taiwan Patent Application No. 107115835 ("835 application") filed on May 10, 2018, and this application incorporates the content of the 835 application here. According to the disclosure of the 835 application, it is assumed that a user pre-records a key to a facial motion (for example, a smile) binding motion (for example, turning on a wiper of a vehicle). According to the 835 application, an image capture device in the car continuously tracks the user's facial motion to detect whether the user is smiling to perform the specific action. However, assuming that in a scenario, the user is talking with others while driving, the image capture device may capture the user's facial patterns which are exactly the same as the smile. In this case, due to the successful recognition, the wiper may start to operate without further confirmation. This sudden movement can distract the driver and cause danger.

Although traditional facial recognition systems have made many innovations to reduce misrecognition, they are not applicable when it comes to the field of dynamic facial recognition. The main reason is in the traditional static face In part recognition, all factors and parameters are constant, so it is relatively easy to debug. On the contrary, when dealing with dynamic face recognition, all facial feature changes need to be considered, which complicates the calculation and processing of dynamic face recognition.

The present invention is mainly a re-invention of the 835 application. The invention discloses a system and method for dynamic facial recognition by setting up a facial action profile (Profile). The invention can not only improve the accuracy of dynamic face recognition, but also eliminate the possibility of misjudgment of actions.

The operation of facial recognition begins with the collection of facial features of a user. There are many known methods for professionals to select facial features, landmarks and/or key points. In one embodiment, a facial recognition system recognizes a plurality of unique facial features/facial feature points of a primary user's face. As shown in Figure 1, in one embodiment, a user’s facial image (facial image) filters out 68 unique feature points (collectively referred to as "label data"). Training of conventional technologies such as neural networks can obtain a trained model. The annotation data and training model are stored in the database for later use. In the subsequent recognition, the system will try to extract 68 facial feature points of the user to be verified and input them into the training model. If the facial feature points of the user to be verified match the data stored in the database, the system verifies that the user is indeed the primary user. For the system, the annotation data of multiple users can be stored in the database for verification. In addition, the above example extracting 68 feature points is only an example. The system may extract more or less feature points, but it can also achieve the same purpose as the application. Therefore, the number 68 should not be used as a limitation to the present invention. .

It is also worth noting that all 68 facial feature points are not necessarily considered in the process of face recognition. In some cases, the system may only consider facial feature points that it considers relevant and ignore those that are not relevant. Although accuracy may be affected, this method can shorten the calculation processing time.

The process of verifying the user's identity based on dynamic facial movements has been discussed in detail in the 835 application. Based on the 835 application, the present invention further discloses a method and system for reducing false verification of dynamic facial movements.

The operation of the dynamic facial recognition system 100 of the present invention is started by creating facial motion profiles. The structure of the recognition system 100 is shown in FIG. 2. The recognition system 100 of the present invention includes a database 120 that stores at least one facial feature point of a user and the user's reference file (predetermined profile). Among them, the reference file is created by the user making a default facial motion. The recognition system 100 of the present invention further includes an image capturing device 110 for tracking the user's face changes and shooting a time frame of video, and a processing unit 130. The function of the processing unit 130 includes creating a facial motion file, comparing a resembling profile with a reference file, and finally determining whether the resembling profile is similar to the reference file. The recognition system 100 can also be equipped with various types of memory (not shown in the figure) to store the images and videos taken by the image capture device 110 and the data processed by the processing unit 130. Figure 3 shows one embodiment of the present invention. As shown in Figure 3, the image capturing device 110 captures a video of the user smiling and expressionless within a period of time T. In the following description, please refer to FIG. 2 and FIG. 3 together to understand the operation of the face recognition system 100.

In one embodiment, the system 100 first creates a reference file for a predetermined facial movement made by a main user. This benchmark file serves as a reference for dynamic facial recognition later. As mentioned in the 835 application, the default facial action can be bound to a function, and the binding function will be activated when the user performs the preset facial action. The binding function may be related to the car function, such as turning on the wiper, etc., which previously required manual control by the driver. The steps to create a benchmark file are as follows. First, the image capturing device 100 captures the facial movements of the main user and records it into a video 300, the total time length of the video is T. The video 300 includes at least a preset dynamic face image 310 (for example, a face image of the main user smiling) and a preset blank face image 320 (for example: a face of the main user without expression image). The total length of video time T can be divided into multiple unit times Tx={t ₁ , t ₂ ,..., t _N }, and each unit time point Tx corresponds to a facial image (facial pattern). It should be noted that the number of facial images is not necessarily equal to the number of frames or positively correlated with the video sampling speed. According to an embodiment, a face image can be a combination of multiple frames (for example, 10 frames), and a unit time is equivalent to the total time length of the multiple frame combination (for example, the total time length of 10 frames).

It is assumed that the video 300 contains a total of N facial images Mx={M ₁ , M ₂ ,..., M _N }, part of which is a dynamic facial image 310 and a part is a blank face image 320. In addition, each facial image expresses a facial snapshot at an instant in the relevant unit time point. For example, a first facial image M ₁ is a snapshot of the user at a first unit time point t ₁ , and a second facial image M ₂ is a snapshot of the user at a second time point t ₂ . In one embodiment, it is assumed that a preset facial action is a smile. As shown in Figure 3, from the perspective of the video 300, the main user at the 16th unit time point t ₁₆ (that is, corresponding to the 16th The facial image M ₁₆ ) starts to smile; and the blank facial image appears in each of the 15 unit time points before and after the video 300 (that is, corresponding to the first facial image M ₁ to the 15th facial image like M _15, N-14 and a second face images M _N-14 to N-th face image M _N). In this embodiment, it is assumed that blank facial expressions 320 appear on both sides of the dynamic facial image 310, but the invention is not limited to this. The primary user can decide to make a blank gesture before or after smiling. In other words, the blank facial image 320 can appear before, after, or before and after the dynamic facial image 310; all the above methods can achieve the purpose of the present invention.

When the recording of the video 300 is completed, 68 facial feature points (ie "labeled data") can be labeled on each dynamic and blank facial image Mx={M ₁ ,M ₂ ,...,M _N } Above, and the variation of each facial feature point over time T can also be obtained. It is worth mentioning that in the embodiment, a smile is used as the preset facial action, so the system 100 can only extract facial feature points around the main user's mouth instead of extracting all the 68 facial feature points. For example, the system 100 of the present invention can extract only 20 facial feature points around the mouth as a calculation reference in this situation. As shown in Figure 1, suppose the 20 relevant facial feature points are f ₄₉ , f ₅₀ ,..., and f ₆₈ , expressed as Fx={f ₄₉ ,f ₅₀ ,...,f ₆₈ } .

Figure 4 demonstrates that the system 100 refers to the changes in the facial image at the relevant time points of the 20 facial feature points, especially the 20 facial features at the unit time points t ₁ , t ₁₆ , t _a , t _b, t _{N-15 and, M 16, M a, M} b, M N-15, M _{N 1} and change the facial image corresponding to the M t _N. As mentioned earlier, the smile starts at the 16th unit time point and continues to the (N-15)th unit time point (ie t ₁₆ -t _N-15 ), where blank facial expressions appear in the first 15 and Finally, 15 units of time (t ₁ -t ₁₅ and t _N-14 -t _N), wherein M ₁ and M _N is a blank facial image, and _{M 16, M a, M b} , M N-15 is Dynamic facial images.

The face recognition system 100 of the present invention calculates a value-Res_fx(t) for each feature point fx (fx is f ₄₉ , f ₅₀ ,..., or f ₆₈ ) from each unit time point to express in the unit The quantized value of the feature point at the time point. The value can be the angle of the feature point in a unit time, the distance, or the position change accumulated by any angle and distance. The calculation of the magnitude Res_fx(t) can be calculated by mathematical conversion (conversion). For example, the magnitude can be calculated by converting the coordinates, angle, distance, position, or relative position of a feature point. In one embodiment, the magnitude Res_fx(t) is defined as the relative position of a characteristic point based on a predetermined anchor point (such as the tip of the nose) in a unit time. In addition, the magnitude can also be the cumulative integral value of all positions of a characteristic point before a certain unit time point. For example, Res_fx(t ₃ ) can be obtained by integrating the position trajectory of a feature point from t ₁ to t ₃ . Figure 5 illustrates the relative position changes of the feature point f ₄₉ at unit time points t ₁ , t ₁₆ , t _a , t _b , t _N-15 and t _N and a preset anchor point (for example, the tip of the nose). In another embodiment, the calculation of the magnitude takes into account the width of the user's face to eliminate the influence of proportion. Using the above method, the magnitude Res_f ₄₉ (t) (t is t ₁ ,..., or t _N ) corresponding to any eigenvalue at any unit time point can be obtained. In addition to the above methods, people in the prior art can also use other methods to obtain the quantified value of the characteristic value per unit time point. The present invention has no limitation on how to obtain these quantized values. The above examples are only used to demonstrate the relevant operations of the present invention.

After obtaining the magnitude of a feature point fx at each unit time point, these magnitudes can be marked as a histogram that changes with time T. Based on the value, the system 100 can know the change of any characteristic point in time T. This change can be represented by a histogram. Figure 6 is an example of how a feature point f ₄₉ changes at time T. The X axis represents the unit time point, and the Y axis represents the value at each unit time point. As shown in the figure, assuming that the values are at t ₁₆ , t ₁₇ and t ₁₈ , the time is V ₁₆ , V ₁₇ and V _{18 respectively} . On the whole, all the values of the feature point f ₄₉ at time T can be represented by a set of residue functions (residue function) Res_f ₄₉ (t).

,其中fx為f₄₉,f₅₀,...及f₆₈ Then the processing unit 1302 of the system 100 performs an integral operation on the remainder function. The result of the integration is defined as the residual value of the feature point f ₄₉ , which is represented by RES(f ₄₉ ). It should be noted that the calculation of the present invention not only considers the time period when a smile occurs (such as t ₁₆ to t _N-15 ), but also considers the time period when a blank expression occurs (that is, t ₁ to t ₁₅ and t _{N- 14} to t _N ). In summary, the range of integral calculation is the total time T of the entire video (that is, from t ₁ to t _N ). The residual value of any characteristic point of the present invention can be expressed by the following integral formula:

, Where fx is f ₄₉ , f ₅₀ ,... and f ₆₈

In the present invention, the blank facial image 320 can be regarded as the guard-band of the dynamic facial image 310. Actually, the blank facial image 320 is used as a parity check for preset facial movements ( parity check). Although Figure 6 shows the magnitude of these blank facial images as null, even the blank expression can be quantified by any of the above methods to obtain the magnitude of each unit time point. In addition, the present invention may additionally define a threshold (threshold), and when the value is lower than the threshold, it means that the corresponding blank face image is indicated.

The residual values of all 20 considered feature points can be calculated by the above integral formula, they can be expressed as RES(f49), RES(f50),...RES(f68); and can be collectively expressed as RES(X )={RES(f49),RES(f50),...RES(f68)}, defined as the change vector of the preset facial action (ie smile) (variation vector). This change vector is the reference file for the main user to make the default facial action-smile. Thereafter, the system 100 refers to the reference file to verify whether the user intends to perform a binding action.

As described above, the blank facial image 320 appears before and after the dynamic facial image 310. In the present invention, the creation of a reference file for a preset facial action (such as a smile) not only considers the changes in the feature points of the preset facial action itself, but also considers the absence of expression before and after the user makes the preset facial action The feature points of the lower face remain unchanged. It is the key of the present invention to consider blank expression images while verifying facial movements. As mentioned above, even when the user makes a facial motion that is very similar to the default facial motion, because the similar facial motion will not be blank expressions before and after the similar facial motion, the file created for the approximate facial motion is in In any case, it will not match the reference file, so the system 100 will not misjudge and perform the binding action by mistake.

Figure 7 is an example of using the system and method of the present invention to verify the user's facial movements. Continuing the example mentioned above, suppose the default facial action is a smile. Even if the user may have made a similar smiling action during the conversation, since the facial images that appear before and after the similarly smiling facial image are not blank facial images, under the overall judgment, the change vector is The extracted facial action that looks like a smile is not the same as a pure smile. Therefore, by adding at least one guard interval during the file creation process, the face recognition system of the present invention can further enhance the recognition accuracy and eliminate misrecognition.

In the above embodiment, the calculation of the magnitude Res_fx(t) is based on the change trajectory of a single facial feature point. However, as understood by those of ordinary skill in the art, this calculation is only an example. The present invention can also consider two or more facial feature points to calculate the magnitude Res_fx(t). For example, a value of Res_fx(t) can be the angle trajectory of any three facial feature points, or the distance between any two facial feature points, and so on.

In addition, the above embodiment uses a blank expression as a guard interval to verify whether the preset facial motion is actually performed, but those skilled in the art should know that the present invention is not limited to this. In other embodiments, As long as the magnitude of a facial expression is lower than a preset threshold, it can be used as a guard interval to distinguish the preset facial motion from other facial motions.

Once the reference file is created, the facial recognition system 100 of the present invention can recognize facial movements based on the reference file. The relevant steps can be seen in Figure 8 and refer to Figure 2 at the same time, and are described in detail in the following pages.

First, assume that the identity of a user has been determined to match. First, in step 801, the image capturing device 110 can continuously track and record the facial movements of the user. Next, in step 802, the processing unit 130 determines whether the user's continuous facial motion is similar to a previously recorded preset facial motion (for example, a smile). If the two facial movements are similar, the image capturing device 110 records the approximate facial movements into a comparison video of the time length T (step 803). The comparison video contains a plurality of critical facial images (critical facial patterns), and the combination of these key facial images is an approximate facial action; in addition, the comparison video also contains a plurality of secondary facial images (secondary facial patterns). patters), and these secondary facial images may appear in front of, behind, or in front of and behind contrasting facial movements. It should be noted that these secondary facial images can be any facial expressions or actions other than contrasting facial actions. In this embodiment, the time length T of the comparison video is also divided into a plurality of unit times.

Next, in step 804, the system 100 of the present invention adopts the same method as setting up a reference file to create a comparison file for approximate facial actions based on the comparison video. Continuing the example of smiling above, in simple terms, the processing unit 130 labels the same 20 facial feature points in each key facial image and secondary facial image. Next, calculate the magnitude of each facial feature point per unit time. Then calculate the histogram and co-function of each facial feature point. The processing unit 130 then integrates the remainder function of each facial feature point to obtain a remainder value. The set of all residual values is a file that approximates facial motions (also called vector variation). It is worth noting that the aforementioned integral calculation in the present invention considers both the comparison of facial actions and other facial expressions before and after it. As mentioned above, using the same method, The comparison file of facial movements can be created.

Then in step 805, after the comparison file is created, the processing unit 130 compares it with the reference file stored in the database 120. If the two files match, the system 100 verifies that the approximate facial motion is the default facial motion (step 806). In other words, the user actually made the same facial motion as the preset facial motion.

Due to the existence of the guard interval (ie: blank expression), the system 100 of the present invention will not cause the dynamic facial image 310 (ie, the preset facial motion: smile) to be compared with the key facial image (ie, comparing the facial motion: similar). Smile) is exactly the same and misjudgment occurs. The system 100 of the present invention considers facial expressions before, after, or before and after the facial motions are compared to determine whether the user actually made a smiling motion (ie, a preset facial motion).

If the two files match, the facial motion verification step of the present invention ends. However, as mentioned in the 835 application, the system 100 of the present invention can further perform a dynamic time warping to more accurately verify the approximate facial motion.

Figure 9 briefly describes the steps to create a profile for a facial action. This step can be applied to the creation of the aforementioned approximate facial motion files and reference files.

First, in step 901, the system 100 extracts at least one facial feature point of the user. The facial feature points can be stored in the database 120. Next, in step 902, the image capturing device 110 records a video, where the duration of the video is T, and the user makes a key facial motion and at least one facial motion in the video. In addition, the video contains multiple primary facial patterns and collateral facial patterns; the primary facial image is a combination of key facial actions, and the secondary facial image is A collection of secondary facial movements. In addition, facial images can be obtained by sampling the video per unit time. As mentioned above, secondary facial movements may appear before, after, or before and after key facial movements. The processing unit 130 of the present invention is equipped to distinguish between key facial movements and secondary facial movements. Ability to move. For example, the processing unit 130 can distinguish a facial image as a main facial image or a subsidiary facial image by a predetermined threshold. Then in step 903, the processing unit 130 respectively annotates facial feature points in each of the main facial image and the subsidiary facial image. Finally, in step 904, the processing unit 130 creates a facial motion file for these labeled facial feature points.

In one embodiment, if the method in Figure 9 is applied to create a reference file of preset facial motions, the main facial image is the dynamic face of the user that makes the default facial motions (for example, smiling) Image; while the attached face image is a blank face image when the user has no expression on his face. In addition, the system 100 of the present invention can store multiple reference files to correspond to multiple binding functions. The relevant explanation was disclosed in the 835 application, so I won’t repeat it here.

In another embodiment, if the method in Figure 9 is suitable for creating a file for comparing facial motions, the main facial image is the facial image of the user who made the facial motions (for example, a similar smile). ; The subsidiary facial image is the facial motion that appears before, after, or before and after the contrast facial motion.

In addition to the above, the creation of a file can also include the following additional steps (not shown in Figure 9). First, the processing unit 130 calculates a plurality of magnitudes of each facial feature point based on the main facial image and the subsidiary facial image. The value represents the change of the facial feature point relative to a preset anchor point at time T. The change can be a change in position, distance, angle, or any combination of the above. After obtaining the value represented by the co-function, the processing unit 130 integrates the co-function in the time range T to obtain the residual value of a certain facial feature point. In this case, since only one facial feature point is considered, the residual value (that is, the change vector) is the file of the facial motion. If multiple facial feature points are considered, the file is a collection of residual values of all facial feature points in real time.

In all the above embodiments, it is assumed that the user identity has been verified. The verification method can be any conventional method or the method disclosed in the 835 application. In addition, even if the user’s identity is not verified in advance, because the present invention uses the user’s facial feature points to verify facial movements, This verification can also complete the user identity verification at the same time. In other words, in one of the embodiments of the present invention, the user authentication step can be omitted.

Although the above embodiments illustrate the working principle of the present invention through a single facial motion, in other embodiments of the present invention, multiple facial motions can also be used. For example, suppose that the system presets two consecutive facial actions (for example, smiling and blinking) as the preset facial actions for a binding operation, and the guard interval of the smile (first facial action) may be smile The expressionless face action before the occurrence and/or the first few blinks (second facial action) face images that occurred after the smile. In addition, the guard interval for blinking (second facial motion) may be the last few facial images of a front smile (first facial motion) and/or a facial expression motion made after blinking. Or alternatively, the system and method of the present invention can treat smiling and blinking as the same facial action and use the same method for recognition.

The foregoing descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made in accordance with the scope of the patent application of the present invention shall fall within the scope of the present invention.

100: identification system

110: Image capture device

120: database

130: processing unit

Claims (18)

A dynamic facial recognition device, including; a database for storing a reference file of a predetermined facial motion made by a user, wherein the predetermined facial motion is recorded as a first video segment, the first video segment A video segment includes a plurality of facial images, and each facial image includes at least one facial feature point of the user and a preset anchor point on the user's face; an image capturing device, which Used to track the user's facial movement and record it into a second video segment, where the second video segment includes a comparative facial movement made by the user in a time; and a processing unit connected to the database And the image capturing device, the processing unit creates the reference file for the preset facial motion through a facial motion file creation step, and creates a comparison file for the contrast facial motion, wherein the processing unit creates the comparison file It is compared with a reference file stored in the database to determine whether the contrast facial motion is the default facial motion. The dynamic facial recognition device according to claim 1, wherein the step of creating a facial motion file includes: the processing unit classifies a video segment captured by the image capturing device within a time into a plurality of main facial images And a plurality of subsidiary facial images, where the main facial images are composed of a key facial action, and the subsidiary facial images are composed of a primary facial action. The dynamic face recognition device according to claim 2, wherein the secondary facial action occurs before, after, or before and after the key facial action. The dynamic face recognition device according to claim 2, wherein the key facial action is the preset The facial motion or the contrast facial motion, and the secondary facial motion is an expressionless motion or other facial motions. The dynamic facial recognition device according to claim 2, wherein the processing unit finds the facial feature points in the main facial images and the subsidiary facial images. The dynamic face recognition device according to claim 5, wherein the processing unit calculates a plurality of values for the facial feature points, and the processing unit further performs mathematical integration of the values for the time to Get a residual value. The dynamic face recognition device according to claim 6, wherein the value is the relative position of the facial feature point with respect to the preset anchor point at a point in time. The dynamic face recognition device according to claim 1, wherein the preset facial action is bound to a function, and if the processing unit determines that the comparison file matches the reference file, the function is executed. A dynamic face recognition method, including: creating a reference file based on a predetermined facial action of a user; capturing a video segment in time, wherein the video segment includes at least one approximation made by the user Facial action; create a comparison file for the video clip; and compare the comparison file with the reference file to determine whether the approximate facial motion matches the preset facial motion, The preset facial action is bound to a function, and the function is executed when the comparison file matches the reference file. The dynamic face recognition method according to claim 9, wherein the establishment of the reference file is based on a plurality of main facial images and a plurality of subsidiary facial images, and the plurality of main facial images are used The person makes a continuous action image of the preset facial action, and the subsidiary facial images are expressionless images that appear before, after, or before and after the preset facial action. The dynamic face recognition method according to claim 9, wherein the establishment of the comparison file includes the following steps: classifying the video segment into a plurality of key facial images and a plurality of secondary facial images, and wherein The plurality of key facial images are the image combination of the user making the approximate facial motion, and the secondary facial images are the user before, after, or before and after the approximate facial motion Make other facial action combinations; mark at least one facial feature point in each of the key facial images and the secondary facial images; according to each of the key facial images and each Each of the secondary facial images calculates a magnitude for the facial feature point; and mathematically integrates the magnitudes over the time to obtain a residual value of the facial feature point. The dynamic face recognition method according to claim 11, wherein the value is the position of the facial feature point relative to a preset anchor point at a time. A computer program used to perform facial motion recognition functions, the computer program is stored in a A non-transitory computer can read the storage medium, and the computer program contains the following code that can execute the following commands: create a reference file for a user's preset facial motion through a face file creation step; obtain An approximate facial action made by the user; a comparison file is created for the contrast facial motion through the facial file creation step; and the comparison file is compared with the reference file to determine whether the contrast facial motion meets the prediction Set facial movements. The computer program according to claim 13, wherein the face file creation step includes the following instruction steps: recording a video clip within a time, wherein the video clip includes a key facial movement and a primary facial movement; The video clip is sampled per unit time to obtain a plurality of key facial images and a plurality of secondary facial images; each of the key facial images and the secondary facial images is marked with the user’s At least one facial feature point; calculating a value for the facial feature point within the unit time; and mathematically integrating the values within the time to obtain a residual value. The computer program according to claim 14, wherein the key facial motion is the default facial motion, and the secondary facial motion is a side that occurs before, after, or before and after the default facial motion Expressionless action; and the residual value is the reference file. The computer program according to claim 14, wherein the key facial motion is the approximate facial motion The secondary facial motion is an expressionless motion or any other facial motion that occurs before, after, or before and after the approximate facial motion; and the residual value is the comparison file. The computer program according to claim 14, wherein the measurement value is the position of the facial feature point relative to a predetermined anchor point in each unit time. The computer program according to claim 13, wherein the default facial action is bound to a function, and if the comparison file matches the reference file, the function is executed.

Images