CN106971203A - Personal identification method based on characteristic on foot - Google Patents

Abstract

The invention discloses an identity recognition method based on walking feature data, comprising the steps of: segmenting the collected triaxial acceleration data, decomposing a continuous movement into time-continuous fixed-length data segments; The features are calculated in the time domain and frequency domain respectively; the features of each segment are classified with a pre-trained classifier; the identification results of a certain number of continuous data segments are summarized to obtain the identification result. It is possible to collect walking data through a portable smart phone, extract features for classification, and then use a trained classifier to classify the collected walking data, thereby identifying the identity of the current user. The cost of equipment is effectively saved, and at the same time, a good recognition effect is achieved.

Description

Identification method based on walking feature data

technical field

The invention belongs to the technical field of identification, and in particular relates to a method for identification based on walking data collected by a sensor of a smart phone.

Background technique

The application of various intelligent systems in daily life is becoming more and more common. Identity recognition is often required in intelligent systems to provide personalized services. Identification is a very difficult problem. In traditional methods, it is mainly identified through some things with personal identity characteristics, such as identification items such as certificates and keys, or identification knowledge such as user names and passwords. However, the disadvantages of traditional identification methods are quite obvious: identification items are easily lost or forged, and identification knowledge is easily forgotten or stolen. In addition, the biological characteristics of each person such as facial features, fingerprints, etc. can be used to achieve good results. However, this method requires too much technology and equipment.

Automatically identifying human physical activities, usually referred to as user activity recognition (HAR, Human Activity Recognition) technology, is an important research direction in the field of human-computer interaction and pervasive computing, and its purpose is to automatically obtain information about user activities And provide related services or applications, so that they can more actively and accurately assist users to complete their goals. Traditional user activity recognition technologies mainly use methods based on computer vision. This type of technology analyzes still images or videos through image processing methods, thereby extracting user activities and judging the categories of activities. Although this type of technology has been extensively researched, it still has obvious defects:

1. This type of technology relies on external devices, so the scope of use is limited to areas where image acquisition devices (such as cameras) have been deployed and can be observed by these devices.

2. Since the information conveyed by images is very rich, there is a risk of leakage of information other than user activities, so this type of technology also has serious privacy issues.

3. Since image processing and video processing technology require high network transmission bandwidth and computing power, it is difficult to achieve real-time processing under the existing technical conditions, which also limits the application of this type of technology in real-time systems.

In recent years, with the rapid development of technologies such as smart mobile devices (such as smartphones, wearable devices) and related sensors (such as motion sensors, electrodermal sensors), the focus of user activity recognition technology research is shifting from computer vision-based methods to Turn to other sensor-based recognition methods on smart mobile devices that users carry with them. The present invention thus comes.

Contents of the invention

In view of the above-mentioned technical problems, the object of the present invention is to provide an identification method based on walking feature data, which can collect walking data through a portable smart phone, extract features for classification, and then use the trained classification The sensor classifies the collected walking data to identify the identity of the current user. The cost of equipment is effectively saved, and at the same time, a good recognition effect is achieved.

Technical scheme of the present invention is:

An identification method based on walking feature data, comprising the following steps:

S01: Segment the collected three-axis acceleration data, and decompose a continuous movement into time-continuous fixed-length data segments;

S02: Calculate features in time domain and frequency domain for each data segment;

S03: Classify the features of each segment with a pre-trained classifier;

S04: Summarize the identification results of a certain number of continuous data fragments to obtain an identification result.

Preferably, in the step S01, a rectangular coordinate system is established with the x-axis in the vertical direction, the y-axis in the left-right direction, and the z-axis in the front-rear direction, and the three-axis acceleration data includes x-axis acceleration data, y-axis acceleration data and z-axis acceleration data.

Preferably, the features calculated in the time domain in the step S02 include maximum value, minimum value, mean value, amplitude, root mean square, standard deviation, zero-crossing rate and kurtosis, and the features calculated in the frequency domain include the maximum frequency , second largest frequency and spectral slope.

Preferably, the calculation formula of the root mean square is:

Among them, i represents the i-th segment, a _ik represents the k-th sample point in the segment, and N represents the total number of samples in the segment.

Preferably, the formula for calculating the zero-crossing rate is:

Among them, i represents the i-th segment, a _ik represents the k-th sample point in the segment, N represents the total number of samples in the segment, and I _R<0 is an index function:

Preferably, the calculation formula of the kurtosis is:

Among them, i represents the i-th segment, a _ik represents the k-th sample point in the segment, is the average number of all sample points, and N represents the total number of samples in the segment.

Preferably, the calculation formula of the spectrum slope is:

where i represents the i-th segment, a _i (k) is the corresponding amplitude of frequency f _i (k) as the k-th frequency component of the i-th segment, and N represents the total number of samples in the segment.

Preferably, the collected triaxial acceleration data is preprocessed before the data is segmented to remove the DC component in the data.

Preferably, in the step S04, if more than half of the data segments are identified as non-users, it is judged as no, and an alarm is issued; otherwise, the judgment is yes.

Preferably, before the step S01, the step S01 is performed by collecting three-axis acceleration data through the built-in three-axis acceleration sensor of the smart phone and judging that the user is walking.

Compared with prior art, the advantage of the present invention is:

1. Practicability: This method uses the built-in three-axis acceleration sensor of a commercial off-the-shelf smart phone to collect data, without the need for additional special equipment, which effectively saves the cost of the equipment.

2. Reliability: This method uses Human Activity Recognition (HAR) to train a classifier, and then uses the trained classifier to classify the data of the current user, thereby identifying the identity of the current user. As long as the training set is sufficient, the recognition error is very small. Finally, the correct rate can be further improved through the identity discrimination and decision-making method in this method.

3. Convenience: the method used to identify the user's identity is based on the user's walking habits. Walking habits are different from the traditional basis for identification such as keys and ID cards, and there is no loss or forgetting.

4. Flexibility: This method has a large scope of application and can be used reliably in bad weather such as cloudy and rainy days.

Description of drawings

The present invention will be further described below in conjunction with accompanying drawing and embodiment:

Fig. 1 is method flowchart of the present invention;

Figure 2 shows a smartphone with an acceleration sensor and its coordinate system

Fig. 3 is the schematic diagram of the data collected by triaxial acceleration sensor of the present invention;

Fig. 4 is the correct rate of 5 kinds of classifiers of the present invention;

Fig. 5 is the time spent of five classifiers of the present invention.

detailed description

The above solution will be further described below in conjunction with specific embodiments. It should be understood that these examples are used to illustrate the present invention and not to limit the scope of the present invention. The implementation conditions used in the examples can be further adjusted according to the conditions of specific manufacturers, and the implementation conditions not indicated are usually the conditions in routine experiments.

Example:

As shown in FIG. 1 , this embodiment performs identity recognition based on walking data. The method mainly consists of four parts: data acquisition, data feature extraction, multi-classifier training and daily activity recognition. Specifically include the following steps:

S01: Segment the collected three-axis acceleration data, and decompose a continuous movement into time-continuous fixed-length data segments;

S02: Calculate features in time domain and frequency domain for each data segment;

S03: Classify the features of each segment with a pre-trained classifier;

S04: Summarize the identification results of a certain number of continuous data fragments to obtain an identification result.

In this embodiment, the three-axis acceleration data is collected by the built-in three-axis acceleration sensor of the smart phone, and of course other smart terminals with built-in three-axis acceleration sensors can also be used for collection. The specific implementation steps are as follows:

(1) The collection of data is the first step of the present invention. In order to collect raw acceleration data for analysis and experiments, we developed an application on the Android platform. The app generates approximately 15 records of triaxial accelerometer data per second and can also continue to run in the background. Then 16 volunteers of different heights and weights were recruited to walk and collect data. As shown in Figure 2, a Cartesian coordinate system is established with the x-axis in the vertical direction, the y-axis in the left-right direction, and the z-axis in the front-back direction. The three-axis acceleration data includes x-axis acceleration data, y-axis acceleration data and z-axis acceleration data. The collected The data is shown in Figure 3. Volunteers walked forward smoothly, allowed 90-degree turns, but did not allow sudden stops and 180-degree turns. One of the 16 people is selected as a positive sample, that is, the owner of the mobile phone, and the remaining 15 people are negative samples.

(2) The collected data is preprocessed to remove the DC component. We then segment it using time-based sliding windows with 40% segments overlapping between adjacent windows. In this way, a section of preprocessed raw data is divided into several sections with equal window sizes, and a data set is established. The data set is divided into two parts, the training set and the test set. The training set contains 1 positive sample, 627 entries, and 11 negative samples, 850 entries. The test set contains 1 positive sample, 327 entries, and 4 negative samples, 319 entries. Among them, the data of the positive sample does not have samples in the same collection process, and the data of the negative sample does not have the data of the same person.

(3) Perform feature extraction for each sample in step (2), and in the time domain, calculate the following features: maximum value, minimum value, mean value, amplitude, root mean square, standard deviation, zero-crossing rate and kurtosis. It includes three components of the x, y, and z axes. In addition, since the x, y, and z axes may have a relationship at the same time, the introduction as the fourth component. Each component has the above 7 features except ZCR ( is always greater than 0, so ZCR is always 0, meaningless). So there are a total of 31 features in the time domain.

(4) For the feature extraction in step (2), in the frequency domain, the following features are calculated: maximum frequency, second maximum frequency, and spectral slope (Spectral Slope). Including 3 components of x, y, z axis, without So there are a total of 9 features in the frequency domain.

(5) Use unified training features to train classifiers. In this example, various classifiers such as SVM (Support Vector Machine), Random Forest, Naive Bayes, Logistic and MLP (Multi-layer perceptronneural networks) are trained. Then use the trained classifier to classify each data segment in the test set to identify the identity of each segment. The verification results are shown in Figure 4, where the accuracy rates of SVM, Random Forest, Logistic and MLP are all 98%. nearby. Among them, the recognition ability of MLP is the best, reaching 98.1132%, followed by 97.7987% of SVM, and the accuracy rate of Random Forest and Logistic is 97.6415%. However, although MLP has the best recognition accuracy, it takes much more time to train the model than the others, as shown in Figure 5.

(6) Identify 20 consecutive time-based segments, marked as {s1, s2, ..., s20}, if more than half of the segments are identified as non-users, it is judged as no and an alarm is issued. Otherwise, the decision is yes. Since the recognition accuracy rate of the segment in the step (5) has reached more than 95%, and then through the identification method in this step, the recognition rate is almost close to 100%.

To sum up, if a user carries a mobile phone installed with the software of the present invention, when the mobile phone is stolen, the mobile phone will recognize the possibility of theft and send an alarm. So as to achieve the purpose of protecting the user's property.

The above examples are only to illustrate the technical conception and characteristics of the present invention, and its purpose is to allow people familiar with this technology to understand the content of the present invention and implement it accordingly, and cannot limit the protection scope of the present invention. All equivalent changes or modifications made according to the spirit of the present invention shall fall within the protection scope of the present invention.

Claims (10)

1. A kind of identification method based on walking feature data, it is characterized in that, may further comprise the steps: S01: Segment the collected three-axis acceleration data, and decompose a continuous movement into time-continuous fixed-length data segments; S02: Calculate features in time domain and frequency domain for each data segment; S03: Classify the features of each segment with a pre-trained classifier; S04: Summarize the identification results of a certain number of continuous data fragments to obtain an identification result. 2. the identification method based on walking feature data according to claim 1, is characterized in that, in described step S01, with vertical direction x-axis, left and right direction is y-axis, front and rear direction is z-axis and establishes rectangular coordinate system, three The axis acceleration data includes x-axis acceleration data, y-axis acceleration data and z-axis acceleration data. 3. the identification method based on walking characteristic data according to claim 1, is characterized in that, the feature calculated in time domain in described step S02 comprises maximum value, minimum value, mean value, amplitude, root mean square, standard Difference, zero-crossing rate, and kurtosis, features computed in the frequency domain include maximum frequency, second maximum frequency, and spectral slope. 4. the identification method based on walking feature data according to claim 3, is characterized in that, the computing formula of described root mean square is: ${\mathrm{<span\; class="notranslate">\; RMS</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{\frac{\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; N</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$ Among them, i represents the i-th segment, a _ik represents the k-th sample point in the segment, and N represents the total number of samples in the segment. 5. the identification method based on walking feature data according to claim 3, is characterized in that, the computing formula of described zero-crossing rate is: ${\mathrm{<span\; class="notranslate">\; zcr</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$ Among them, i represents the i-th segment, a _ik represents the k-th sample point in the segment, N represents the total number of samples in the segment, and I _R<0 is an index function: 6. the identification method based on walking characteristic data according to claim 3, is characterized in that, the computing formula of described kurtosis is: ${\mathrm{<span\; class="notranslate">\; Kurtosis</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 4</span>}}}{{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$ Among them, i represents the i-th segment, a _ik represents the k-th sample point in the segment, is the average number of all sample points, and N represents the total number of samples in the segment. 7. the identification method based on walking feature data according to claim 3, is characterized in that, the computing formula of described frequency spectrum slope is: ${\mathrm{<span\; class="notranslate">\; Spectral\; Slope</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>}\frac{\mathrm{<span\; class="notranslate">\; N</span>}\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; N</span>}\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$ where i represents the i-th segment, a _i (k) is the corresponding amplitude of frequency f _i (k) as the k-th frequency component of the i-th segment, and N represents the total number of samples in the segment. 8. The identification method based on walking characteristic data according to claim 1, characterized in that, before the data segmentation, the triaxial acceleration data collected is preprocessed to remove the DC component in the data. 9. The identification method based on walking characteristic data according to claim 1, characterized in that, in the step S04, if more than half of the data fragments are identified as non-users, it is judged as no, and an alarm is issued; Otherwise, the decision is yes. 10. The identity recognition method based on walking characteristic data according to claim 1, characterized in that, before the step S01, it also includes collecting three-axis acceleration data through a built-in three-axis acceleration sensor in the smart phone, and judging that when the user walks, Go to step S01.