TW201442019A - Building method and using method for infant crying implication determining model - Google Patents

Abstract

The present invention provides a building method and a using method for infant crying implication determining model. The building method is applicable when an infant cries for a specific reason. The building method includes the steps of capturing a plurality of audio characteristics and a plurality of facial image characteristics of the infant, repeating the above step, and using a sorter to respectively integrate the audio characteristics and image characteristics into an audio model and an image model. The audio model and the image model form a determining model. The using method is applicable when an infant is crying, and the using method includes the steps of capturing a plurality of audio characteristics and a plurality of facial image characteristics of the infant, and making a comparison with the audio model and the image model, and if similarities of the audio and the image are respectively higher than a first threshold value and a second threshold value, a reason of the infant crying is determined to be identical to the specific reason when building the determining model.

Description

Establishment method and use method of baby crying meaning interpretation model

The invention relates to sound and image recognition; in particular to a method and a method for establishing a baby crying meaning interpretation model.

Before the baby learns to speak, he expresses various physiological, psychological and emotional needs with crying and facial expressions, so to some extent, crying and facial expressions can be regarded as an innate In special languages, babies communicate with the outside world through this special language.

In general, the cause of babies crying is not to be uncomfortable (overheated or too cold), to take care (hungry or diaper wet) and emotional dependence (craving attention), etc., while the medical staff or parents are caring for In babies, most of them rely on experience to guess the reasons behind baby crying, lack of objective and reliable judgment basis, especially novice parents who lack parenting experience, but often feel helpless when the baby is crying.

In view of this, the object of the present invention is to provide a method for establishing a baby crying meaning interpretation model and a method for using the same, which can accurately determine the reason for the baby crying and provide a reference for the infant caregiver.

In order to achieve the above object, the method for establishing a baby crying meaning interpretation model provided by the present invention is performed when the baby is crying for a specific reason, and includes the following steps: a. recording a crying signal when the baby is crying. And extracting the plural audio features from the crying signal; b, taking the baby crying a plurality of image features of the facial image; c, repeating steps a to b, and using a classifier to integrate the audio features extracted from steps a to b each and the image features respectively An audio model and an image model are formed, and the audio model and the image model together form the interpretation model.

The invention provides a baby crying intentional interpretation method using the foregoing interpretation model, and performs the following steps when a baby is crying: a. recording a crying signal when the baby is crying, and extracting from the crying signal a plurality of audio features; b, capturing a plurality of image features of the face image of the baby crying; c, comparing the audio features captured in step a with the audio model, and taking the step b The image features are compared with the image model, and if the audio features are compared with the audio model, the similarity is higher than a first threshold, and the image features are compared with the image model to obtain a high degree of similarity. At a second threshold, the reason for the baby to be crying at this time is the same as the specific reason for the establishment of the interpretation model.

Thereby, the method and the method for establishing the baby crying meaning interpretation model provided by the invention can reliably determine the reason for the baby crying, so that the medical staff or the parents can properly care for the baby.

1 is a flow chart of a method for establishing a baby crying meaning interpretation model of the present invention; FIG. 2 is a schematic diagram showing a feature point on a baby face image by using an active appearance model; FIG. 3 is a schematic diagram of a baby crying meaning interpretation model of the present invention. A flow chart of the method of use.

In order to explain the present invention more clearly, the preferred embodiment will be described in detail with reference to the accompanying drawings. Referring to FIG. 1, a preferred embodiment of the method for establishing a baby crying meaning interpretation model of the present invention utilizes a baby because When crying for a specific reason, the plural audio features of the baby crying and the complex image features of the facial expression are respectively taken for later integration into the interpretation model; regarding the audio features and the image features The details are as follows: the audio features are extracted from a crying signal, and the crying signal is composed of a plurality of frames, and the result of capturing the audio features is more accurate. Fast Fourier Transform is performed on the crying signal, and Spectral Analysis is performed for each of the frames, and then complex audio features including intensity (Intensity) and tone (Timbre) are extracted. The audio features belonging to the timbre include a Centroid, a Bandwidth, a Roll-off, an Octave-Based, and a Zero-Crossing. Rate) and so on. It should be noted that the types of audio features enumerated above are not the limitations of the present invention. The audio features captured in other embodiments may of course be of a type not found here.

As for the image features, the present embodiment uses Active Appearance Models technology, but is not limited thereto. Before using the active appearance model technology, it is necessary to mark a plurality of feature points on the other facial image as training (Training); after the training is completed, the active appearance model technology can be used on the face of the baby. The corresponding feature points are extracted from the image, but the training involved and the methods for extracting the feature points are known in the art, and will not be described here. FIG. 2 is an example of extracting the feature points. In the preferred embodiment, the feature points extracted by the active appearance model technology are respectively located in the eyelids, noses, and lips of the facial image. Can be selected among the feature points Taking certain feature points, taking the distance between the selected feature points and the size of the surrounding geometric features as the image features, but the selection of the feature points is not static, Different embodiments may have different ways of selecting.

The method for establishing a baby crying intentional interpretation model of the present invention repeats the steps of capturing the audio features and the image features a plurality of times when the baby is crying for the specific reason; the embodiment is used to support Vector Regression (Support Vector Regression) is an example of a classifier that integrates the aforementioned audio features into an audio model, and also integrates the image features into an image model. The audio model and the image model together form the interpretation model for this particular reason.

Referring to FIG. 3, the baby crying intentional interpretation method of the present invention using the aforementioned interpretation model is used when a baby is crying. First, the complex audio features of the baby crying and the complex image features of the facial image are captured. As mentioned above, the audio features are extracted from a crying signal, and the crying signal is also composed of a plurality of sound frames; for the same reason as before, in order to make the audio features extracted by the hammer more accurate, The crying signal is processed by fast Fourier transform, and each of the sound frames is also subjected to spectrum analysis. The types of the audio features extracted are the same as those described in the method for establishing the interpretation model of the present invention, and are not described herein again.

Similarly, the image features are obtained by using the active appearance model technology for the facial images, and the image features must be extracted in the same manner as described in the method for establishing the interpretation model, so as to facilitate subsequent comparison steps. Go on.

After the audio features and the image features are removed, the present embodiment uses a multi-class support vector machine (Multi-Class Support Vector Machine) as an example to compare the audio features and the audio model. Comparing the image features with the image model; if the audio features are compared with the audio model, the similarity is higher than a first threshold, and the If the similarity between the image feature and the image model is higher than a second threshold, it is determined that the reason for the baby crying at this time is the same as the specific reason when the interpretation model is established; otherwise, the baby is determined to cry. The reason for the trouble is different from the specific reason when the interpretation model was established.

In practice, a plurality of interpretation models can be established for a plurality of specific reasons. When the baby is crying, the captured audio features and the image features are performed one by one with the audio model of the interpretation models and the image model. The reason for comparing the baby's crying is closest to the specific reason for which of the interpretation models was established.

In summary, the method and method for establishing a baby crying meaning interpretation model of the present invention can provide a reliable and accurate judgment criterion, so that the infant caregiver can easily infer the cause of the baby crying and can appropriately respond. Dispose of.

The above is only a preferred embodiment of the present invention, and equivalent variations of the method of the present invention and the scope of the patent application are intended to be included in the scope of the present invention.

Claims (10)

A method for establishing a baby crying meaning interpretation model is to perform the following steps when the baby is crying for a specific reason: a, recording a crying signal when the baby is crying, and taking a plural from the crying signal Audio feature; b, capturing a plurality of image features of the face image of the baby when crying; c, repeating steps a through b, and using a classifier, each step a to step b is taken The audio features and the image features are respectively integrated into an audio model and an image model, and the audio model and the image model together form the interpretation model. The method for establishing a baby crying meaning interpretation model according to claim 1, wherein the audio features described in step a include intensity and tone (Timbre); wherein the audio features belonging to the tone include a waveform center of gravity (Centroid), Bandwidth, Roll-off, Octave-Based, and Zero-Crossing Rate. The method for establishing a baby crying meaning interpretation model according to claim 2, wherein the crying signal described in step a is composed of a plurality of frames, and the crying signal is fast Fourier transform (Fast Fourier) Transform), each of the frames is subjected to Spectral Analysis processing. The method for establishing a baby crying meaning interpretation model according to claim 1, wherein the image features described in step b are captured using Active Appearance Models. The method for establishing a baby crying meaning interpretation model according to claim 1, wherein the classifier described in step c is a support vector regression (Support Vector Regression). A baby crying intentional interpretation method using the interpretation model as claimed in claim 1, the reading method performing the following steps when a baby is crying: a, recording a crying signal when the baby is crying, and crying from the crying sound Obtaining a plurality of audio features in the signal; b, capturing a plurality of image features of the facial image of the baby crying; c, comparing the audio features captured in step a with the audio model, and step b Comparing the image features with the image model, if the audio features are compared with the audio model, the similarity is higher than a first threshold, and the image features are compared with the image model. If the similarity is higher than a second threshold, the reason for the baby to be crying at this time is the same as the specific reason for the establishment of the interpretation model. The infant crying meaning interpretation method according to claim 6, wherein the audio features described in step a include intensity and timbre; wherein the audio features belonging to the timbre include a waveform center of gravity, a range width, a degree of attenuation, Octave and zero crossing rate. The infant crying meaning interpretation method according to claim 7, wherein the crying signal described in the step a is composed of a plurality of sound boxes, and the crying signal is processed by fast Fourier transform, and each of the sound boxes is accepted. Spectrum analysis processing. The baby crying meaning interpretation method according to claim 6, wherein the image features described in step b are captured using an active appearance model technique. The baby crying meaning interpretation method according to claim 6, wherein the comparison method described in the step c is performed by using a multi-class support vector machine (Multi-Class Support Vector Machine).