CN105989694A

CN105989694A - Human body falling-down detection method based on three-axis acceleration sensor

Info

Publication number: CN105989694A
Application number: CN201510074961.1A
Authority: CN
Inventors: 孙子文; 孙晓雯
Original assignee: Jiangnan University
Current assignee: Jiangnan University
Priority date: 2015-02-05
Filing date: 2015-02-05
Publication date: 2016-10-05

Abstract

The invention discloses a human body falling-down detection method based on a three-axis acceleration sensor. The human body falling-down detection method comprises the steps of acquiring X-axis acceleration data, Y-axis acceleration data and Z-axis acceleration data by an acceleration sensor when a user moves, and processing the acquired data; judging whether a Y-axis acceleration peak value reaches a threshold TH1 or not; when the Y-axis acceleration peak value exceeds TH1, intercepting a time period within 2s after the Y-axis acceleration peak value appears, extracting features in the time period, and carrying out matching on difference time sequences and feature sequences of a falling-down behavior in sequence by using DTW (Dynamic Time Warping) in the time period so as to recognize whether a user falls down or not. The human body falling-down detection method is convenient to apply in falling-down detection, and can effectively differentiate falling-down and daily behaviors.

Description

A human fall detection method based on a three-axis acceleration sensor

技术领域：Technical field:

本发明涉及模式识别与传感器技术领域，具体涉及一种基于三轴加速度传感器的人体跌倒检测方法。The invention relates to the field of pattern recognition and sensor technology, in particular to a human fall detection method based on a three-axis acceleration sensor.

背景技术：Background technique:

当今的意外事故中，约有29％是由跌倒直接或间接导致。跌倒轻者会导致伤者擦伤、骨折，重者甚至导致死亡，尤其对老年人等弱势群体来讲，跌倒是其健康生活的巨大威胁。医学研究表明：只要人体在跌倒发生后得到及时救治，就可以避免重大疾病的发生。因此，对弱势群体进行跌倒检测，最大限度地提高他们的健康水平，节省医疗开销，则是一个十分重要的医疗问题和社会问题。About 29% of today's accidents are directly or indirectly caused by falls. Falls can cause bruises, fractures, and even death in severe cases. Especially for vulnerable groups such as the elderly, falls are a huge threat to their healthy life. Medical research shows that as long as the human body is treated in time after a fall, the occurrence of major diseases can be avoided. Therefore, it is a very important medical and social issue to perform fall detection on vulnerable groups to maximize their health and save medical expenses.

目前，检测跌倒的方法有两大类：基于环境遥感技术的跌倒检测和基于穿戴式传感器的跌倒检测。(1)基于环境遥感技术的跌倒检测系统通过在特定环境中放置位置固定的振动传感器或者摄像头获取人体运动信息进，从振动传感器获取周围一定范围内的地面振动信号特征，从一个或多个摄像头获取人体运动图像特征，进而通过数据挖掘、模式识别、数据融合、小波分析、人工智能等技术对提取的运动信息进行分析处理，判断人体是否发生跌倒现象。此类方法只在特定环境中有效，易受环境干扰，适应性差，设备价格往往较高，因此不易被用户接受。(2)基于穿戴式传感器的跌倒检测系统，通常是在日常生活用品中嵌入微型传感器(加速度传感器、陀螺仪、压力传感器等)放置在人体的头部、胸部、腰部、手臂、大腿、足底等位置采集人体的加速度值、角度值、压力值等信号特征，实时地监测人体活动，在人体的活动参数有较大改变时通过一定的算法判断是否发生了跌倒，一旦发生了跌倒，通过无线发送模块对这一情况进行定位以及报警。基于穿戴式传感器的跌倒检测过程不受环境限制，克服环境遥感技术只能限定在一定范围的问题，灵活性强，具有广阔的发展前景。At present, there are two categories of fall detection methods: fall detection based on environmental remote sensing technology and fall detection based on wearable sensors. (1) The fall detection system based on environmental remote sensing technology obtains human body movement information by placing a vibration sensor or camera with a fixed position in a specific environment, and obtains the ground vibration signal characteristics within a certain range around the vibration sensor, and from one or more cameras. Obtain the characteristics of human motion images, and then analyze and process the extracted motion information through data mining, pattern recognition, data fusion, wavelet analysis, artificial intelligence and other technologies to determine whether the human body has fallen. Such methods are only effective in a specific environment, are susceptible to environmental interference, have poor adaptability, and often have high equipment prices, so they are not easily accepted by users. (2) A fall detection system based on wearable sensors is usually embedded in daily necessities with miniature sensors (acceleration sensors, gyroscopes, pressure sensors, etc.) placed on the head, chest, waist, arms, thighs, soles of the human body Collect signals such as the acceleration value, angle value, and pressure value of the human body at other positions, and monitor human body activities in real time. The sending module locates and alarms this situation. The fall detection process based on wearable sensors is not limited by the environment, overcomes the problem that environmental remote sensing technology can only be limited to a certain range, has strong flexibility, and has broad development prospects.

加速度传感器相比于陀螺仪、压力传感器等其他穿戴式传感器，具有以下优越性：①陀螺仪只能测量各个轴的方向信息。而加速度传感器不仅能采集加速度的方向信息，还能采集加速度的大小信息。②压力传感器通常放在足底，缺乏灵活性，一般要与其他传感器协同工作对人体活动进行监测。而加速度传感器可以放置的人体躯干各个部位，识别率高，可以单独使用来完成跌倒检测，已是目前跌倒检测研究的主要趋势。Compared with other wearable sensors such as gyroscopes and pressure sensors, acceleration sensors have the following advantages: ①Gyroscopes can only measure the direction information of each axis. The acceleration sensor can not only collect the direction information of the acceleration, but also the magnitude information of the acceleration. ②The pressure sensor is usually placed on the sole of the foot and lacks flexibility. It generally needs to work with other sensors to monitor human activities. Acceleration sensors can be placed in various parts of the human body, have a high recognition rate, and can be used alone to complete fall detection, which is the main trend of current fall detection research.

发明内容：Invention content:

技术问题：本发明解决的技术问题在于克服了基于环境遥感技术跌倒检测在使用环境上的局限性、适应性差、易受干扰的弊端，克服了陀螺仪获取人体运动信息不全面的弊端，克服了压力传感器灵活性差的缺陷，提供了一种基于三轴加速度传感器的人体跌倒检测方法，检测方法简单易行，能较好的从人体的各种行为中检测出跌倒。Technical problem: The technical problem solved by the present invention is to overcome the limitations of the use environment, poor adaptability, and easy interference of fall detection based on environmental remote sensing technology, overcome the disadvantages of incomplete human body movement information obtained by gyroscopes, and overcome the Due to the defect of poor flexibility of the pressure sensor, a human fall detection method based on a triaxial acceleration sensor is provided. The detection method is simple and easy, and can better detect falls from various human behaviors.

技术方案：本发明提供了一种基于三轴加速度传感器的人体跌倒检测方法，采用以下技术方案解决上述技术问题：Technical solution: The present invention provides a human body fall detection method based on a three-axis acceleration sensor, and adopts the following technical solutions to solve the above technical problems:

a.模板生成：a. Template generation:

1.利用三轴加速度传感器采集人体在跌倒过程中的X、Y、Z三轴加速度数据序列。1. Use the three-axis acceleration sensor to collect the X, Y, Z three-axis acceleration data sequence of the human body during the fall process.

2.对上述采集到的X、Y、Z三轴加速度数据序列进行预处理，定义在第t个时刻点所得原始X、Y、Z三轴加速度数据分别为a_x(t)、a_y(t)、a_z(t)，用宽度m为3的滑动平均滤波器进行平滑去噪预处理，设a′_x(t)、a′_y(t)、a′_z(t)为预处理之后的三轴加速度，则方法如式(1)、(2)(3)所示：2. Perform preprocessing on the X, Y, and Z three-axis acceleration data series collected above, and define the original X, Y, and Z three-axis acceleration data obtained at the tth time point as a _x (t), a _y ( t), a _z (t), use a moving average filter with a width m of 3 for smoothing and denoising preprocessing, let a′ _x (t), a′ _y (t), and a′ _z (t) be the preprocessing After the three-axis acceleration, the method is shown in formulas (1), (2) and (3):

${a a}_{x x}^{' '} ((t t)) = = \frac{11}{m m} {Σ Σ}_{k k = = 00}^{m m - - 11} {a a}_{x x} ((t t + + k k)) - - - - - - ((11))$

${a a}_{y the y}^{' '} ((t t)) = = \frac{11}{m m} {Σ Σ}_{k k = = 00}^{m m - - 11} {a a}_{y the y} ((t t + + k k)) - - - - - - ((22))$

${a a}_{z z}^{' '} ((t t)) = = \frac{11}{m m} {Σ Σ}_{k k = = 00}^{m m - - 11} {a a}_{z z} ((t t + + k k)) - - - - - - ((33))$

3.Y轴加速度峰值计算公式如(4)所示：3. The formula for calculating the peak value of the Y-axis acceleration is shown in (4):

Y_max＝max a_y′(t) (4)Y _max = max a _y '(t) (4)

人体在跌倒过程中，竖直方向上的加速度变化最为明显，且与日常生活活动的变化不同，因此首先对Y轴加速度进行分析。判断Y轴加速度峰值Y_max是否超过TH1，若Y_max＞TH1，则执行下一步，反之重新采集数据。During the fall process of the human body, the acceleration change in the vertical direction is the most obvious, and it is different from the change in daily life activities, so the Y-axis acceleration is analyzed first. Determine whether the Y-axis acceleration peak value Y _max exceeds TH1, if Y _max > TH1, execute the next step, otherwise collect data again.

4.从Y轴加速度峰值Y_max出现开始到Y_max出现后2s作为特征提取的时间段，该时间段内有100个数据点的特征序列。为了正确将日常行为与别跌倒行为区分，本发明对预处理后数据提取了两个特征来进行识别，分别是合加速度与倾角。4. From the appearance of the Y-axis acceleration peak value Y _max to 2 seconds after the appearance of Y _max as the feature extraction time period, there are 100 feature sequences of data points in this time period. In order to correctly distinguish daily behaviors from non-falling behaviors, the present invention extracts two features from the preprocessed data for identification, namely combined acceleration and inclination angle.

①合加速度① Combined acceleration

合加速度表明人体运动的剧烈程度，其值越大，运动越剧烈。人体在跌倒后2s之内身体动作幅度不会太大，合加速度较小，会在1g上下波动。在第t个序列点的合加速度计算公式如(5)所示：The resultant acceleration indicates the intensity of the human body movement, the greater the value, the more intense the movement. Within 2 seconds after the human body falls, the range of body movements will not be too large, and the resultant acceleration will be small, fluctuating around 1g. The calculation formula of the resultant acceleration at the tth sequence point is shown in (5):

$A A ((t t)) = = \sqrt{{a a}_{x x}^{' ' 22} ((t t)) + + {a a}_{y the y}^{' ' 22} ((t t)) + + {a a}_{z z}^{' ' 22} ((t t))} - - - - - - ((55))$

②倾角②Inclination

倾角表明人体的倾斜程度，其值越大，倾斜程度越大。当人体处于直立状态时倾角为0°，平躺状态时倾角为90°。人体在跌倒后2s之内无法自己站立起来，因此跌倒后2s之内处于平躺状态，倾角在90°附近波动。第t个序列点的倾角计算公式如(6)所示：The inclination angle indicates the degree of inclination of the human body, and the larger the value, the greater the degree of inclination. When the human body is in an upright state, the inclination angle is 0°, and when the human body is lying flat, the inclination angle is 90°. The human body cannot stand up by itself within 2 seconds after falling, so it is in a flat state within 2 seconds after falling, and the inclination fluctuates around 90°. The formula for calculating the inclination angle of the tth sequence point is shown in (6):

5.最终得到四次跌倒行为的特征序列作为样本模板，设第i次跌倒行为的特征序列为F_i，其计算公式如(7)所示：5. Finally, the feature sequence of the four falling behaviors is obtained as a sample template, and the feature sequence of the i-th falling behavior is set as F _i , and its calculation formula is shown in (7):

$\{\begin{matrix} {FA FA}_{i i} = = (({A A}_{i i} ((11)),, {A A}_{i i} ((22)),, . . . . . .,, {A A}_{i i} ((t t)),, {A A}_{i i} ((n no)))) \\ {FQ FQ}_{i i} = = (({φ φ}_{i i} ((11)),, {φ φ}_{i i} ((22)),, . . . . . .,, {φ φ}_{i i} ((t t)),, {φ φ}_{i i} ((n no)))) \end{matrix} - - - - - - ((77))$

其中1≤i≤4，1≤t≤n，n＝100，i，t∈Z。用DTW算法求取i、j两个模板对应于合加速度与倾角特征序列之间的最小累积距离分别为D(FA_i，FA_i)、D(FQ_i，FQ_j)，则i、j两个模板之间的距离如式(8)所示：Wherein 1≤i≤4, 1≤t≤n, n=100, i, t∈Z. Use the DTW algorithm to obtain the minimum cumulative distance between the two templates i and j corresponding to the resultant acceleration and the dip angle feature sequence as D(FA _i , FA _i ), D(FQ _i , FQ _j ), then the two templates i and j The distance between templates is shown in formula (8):

D(i，j)＝D(FA_i，FA_j)+D(FQ_i，FQ_j) (8)D(i, j)=D(FA _i , FA _j )+D(FQ _i , FQ _j ) (8)

用式(8)求取其他两个模板组合的距离，共有6种组合，根据公式(9)求取两个跌倒行为之间的平均距离：Use formula (8) to find the distance between the other two template combinations. There are 6 combinations in total. Calculate the average distance between two fall behaviors according to formula (9):

$DM DM = = \frac{((D D. ((1,2 1,2)) + + D D. ((1,3 1,3)) + + D D. ((1,4 1,4)) + + D D. ((2,3 2,3)) + + D D. ((2,4 2,4)) + + D D. ((3,4 3,4))))}{66} - - - - - - ((99))$

最终将DM以及F_i存入模板库。Finally, DM and F _i are stored in the template library.

b.跌倒检测：b. Fall detection:

1.用户完成日常生活中可能出现的行为，用三轴加速度传感器采集人体在跌倒过程中的X、Y、Z三轴加速度数据a_x(t)、a_y(t)、a_z(t)；1. The user completes the behaviors that may occur in daily life, and uses the three-axis acceleration sensor to collect the X, Y, and Z three-axis acceleration data a _x (t), a _y (t), and a _z (t) of the human body during the fall process ;

2.通过式(1)(2)(3)对原始加速度数据进行预处理，得到预处理后的加速度数据a′_x(t)、a′_y(t)、a′_z(t)；2. Preprocess the original acceleration data by formula (1) (2) (3) to obtain preprocessed acceleration data a' _x (t), a' _y (t), a' _z (t);

3.根据式(4)求取运动过程中Y轴加速度峰值Y_max，并判断式Y_max是否超过阈值TH2，根据式(10)获得判断结果：3. Calculate the Y-axis acceleration peak value Y _max during the movement according to formula (4), and judge whether the formula Y _max exceeds the threshold value TH2, and obtain the judgment result according to formula (10):

${f f}_{11} = = \{\begin{matrix} 11,, {Y Y}_{max max} &GreaterEqual; &Greater Equal; TH TH 11 \\ 00,, {Y Y}_{max max} < < TH TH 11 \end{matrix} - - - - - - ((1010))$

若判断结果f₁为1，则进入下一步判断，反之，重新采集数据。人体跌倒过程中接触地面时Y轴加速度会急剧增加达到最大，通过对Y轴加速度峰值Y_max大小的检测，可以知道人体在竖直方向上的运动是否较为剧烈。If the judgment result f ₁ is 1, enter the next step of judgment, otherwise, collect data again. When the human body falls and touches the ground, the Y-axis acceleration will increase sharply and reach the maximum. Through the detection of the Y-axis acceleration peak value Y _max , it can be known whether the human body is moving more violently in the vertical direction.

4.根据式(5)(6)提取Y轴加速度峰值Y_max出现开始到Y_max出现后2s时间段的特征序列F_p，即合加速度特征序列以及倾角特征序列，如式(11)所示：4. Extract the characteristic sequence F _p of the Y-axis acceleration peak value Y _max to 2s after the appearance of Y _max according to the formula (5) (6), that is, the characteristic sequence of the combined acceleration and the characteristic sequence of the inclination angle, as shown in the formula (11) :

$\{\begin{matrix} {FA FA}_{p p} = = (({A A}_{p p} ((11)),, {A A}_{p p} ((22)),, . . . . . .,, {A A}_{p p} ((t t)),, {A A}_{p p} ((n no)))) \\ {FQ FQ}_{p p} = = (({φ φ}_{p p} ((11)),, {φ φ}_{p p} ((22)),, . . . . . .,, {φ φ}_{p p} ((t t)),, {φ φ}_{p p} ((n no)))) \end{matrix} - - - - - - ((1111))$

5.将该行为的特征序列F_p与4组模板的特征序列进行匹配，通过式(8)计算该行为与第i个跌倒行为模板之间的距离D(p，i)，根据如下式(12)计算该行为与跌倒行为之间的平均距离：5. Match the feature sequence F _p of this behavior with the feature sequence of 4 groups of templates, and calculate the distance D(p, i) between this behavior and the ith fall behavior template by formula (8), according to the following formula ( 12) Calculate the average distance between this action and the fall action:

$DC DC = = \frac{((D D. ((p p,, 11)) + + D D. ((p p,, 22)) + + D D. ((p p,, 33)) + + D D. ((p p,, 44))}{44} - - - - - - ((1212))$

6.跌倒行为识别的状态标志f₂如下式(13)所示：6. The state flag f ₂ of falling behavior recognition is shown in the following formula (13):

${f f}_{22} = = \{\begin{matrix} 11,, DC DC / / DM DM \leq \leq TH TH 22 \\ 00,, DC DC / / DM DM > > TH TH 22 \end{matrix} - - - - - - ((1313))$

其中DM/DC表明该组行为与跌倒的相似程度，其值越小，说明该组行为越接近于跌倒行为，对DM/DC设定阈值，通过判断是否达到阈值可以识别跌倒行为。若状态标志f₂为1，则该行为是跌倒行为，反之，该行为是非跌倒行为。Among them, DM/DC indicates the similarity between the behavior of this group and falling. The smaller the value, the closer the behavior of this group is to falling behavior. Set a threshold for DM/DC, and the falling behavior can be identified by judging whether the threshold is reached. If the state flag f ₂ is 1, the behavior is a falling behavior, otherwise, the behavior is a non-falling behavior.

本发明与现有技术相比，具有以下优点：Compared with the prior art, the present invention has the following advantages:

1.克服了振动传感器、摄像头的设备成本高、使用范围小、受环境干扰大等弊端，采用加速度传感器可以随时随地获取人体运动信息，不会对用户的日常生活造成影响，且成本低，更容易被用户接受。1. Overcoming the disadvantages of vibration sensors and cameras, such as high equipment cost, small use range, and large environmental interference, the use of acceleration sensors can obtain human body movement information anytime and anywhere, without affecting the daily life of users, and the cost is low. easily accepted by users.

2.对Y轴加速度峰值设定阈值，通过判断某行为运动过程的Y轴加速度峰值是否超过阈值，对行为进行了预判断，排除掉与跌倒差距较大运动。2. Set a threshold for the peak value of the Y-axis acceleration. By judging whether the peak value of the Y-axis acceleration of a certain behavior exceeds the threshold, the behavior is pre-judged, and the movement that is far from falling is excluded.

3.从人体运动幅度以及人体运动姿态两个方面提取了对人体运动区分度较高的特征，使用DTW方法对人体运动状态进行判断，能达到较高的识别率，本发明可以有效识别跌倒行为。3. From two aspects of human body motion amplitude and human body motion posture, features with high discrimination of human body motion are extracted, and the DTW method is used to judge the state of human body motion, which can achieve a high recognition rate. The present invention can effectively identify fall behavior .

附图说明：Description of drawings:

图1为模板生成流程图Figure 1 is a flow chart of template generation

图2为跌倒检测流程图Figure 2 is a flow chart of fall detection

图3为人体动作模型三轴加速度方向图Figure 3 is the three-axis acceleration direction diagram of the human action model

图4为提取特征时间段选取图Figure 4 is a selection diagram of the time period for extracting features

具体实施方式：detailed description:

人体的日常生活运动主要包括跳跃、跑步、步行、坐下、躺下、弯腰等。本发明是根据人体日常生活运动与跌倒行为的运动特征不同，基于三轴加速度传感器对跌倒行为进行检测。The daily activities of the human body mainly include jumping, running, walking, sitting, lying down, bending over, etc. The invention detects the falling behavior based on the three-axis acceleration sensor according to the difference in motion characteristics between the daily life movement of the human body and the falling behavior.

为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图对本发明一种基于三轴加速度传感器的人体跌倒检测方法进一步详细说明。应当理解，本发明的实施方式不限于此。In order to make the purpose, technical solution and advantages of the present invention clearer, a human fall detection method based on a triaxial acceleration sensor of the present invention will be further described in detail below in conjunction with the accompanying drawings. It should be understood that the embodiments of the present invention are not limited thereto.

如图1所示，为本发明提供的用于建立人体跌倒检测的匹配模板流程图，具体步骤包括：As shown in Figure 1, the matching template flow chart for establishing human fall detection provided by the present invention, the specific steps include:

步骤101：用户完成跌倒行为，由加速度传感器采集跌倒过程的三轴加速度数据，以水平向左为X轴，垂直向上为Y轴，水平向前为Z轴，X、Y、Z三轴互相垂直，三轴方向见图3。Step 101: The user completes the fall behavior, and the acceleration sensor collects the three-axis acceleration data during the fall process. The horizontal left is the X axis, the vertical upward is the Y axis, and the horizontal forward is the Z axis. The X, Y, and Z axes are perpendicular to each other , see Figure 3 for the three-axis direction.

步骤102：对获得的原始三轴加速度数据进行预处理，即采用滑动平均滤波器进行平滑去噪处理，处理过程见技术方案a.模板生成2。Step 102: Preprocessing the obtained original three-axis acceleration data, that is, using a moving average filter to perform smoothing and denoising processing. For the processing process, see technical solution a. Template generation 2.

步骤103：提取Y轴加速度峰值Y_max，判断Y_max是否大于TH1，若是，则进行步骤104，反之进行步骤101。Step 103 : Extract the Y-axis acceleration peak value Y _max , judge whether Y _max is greater than TH1 , if yes, go to step 104 , otherwise go to step 101 .

步骤104：截取Y轴加速度峰值出现时刻开始到Y轴加速度峰值出现2s这一时间段T，如图4所示。Step 104: Intercept the time period T from the time when the peak value of the Y-axis acceleration appears to 2 seconds after the peak value of the Y-axis acceleration appears, as shown in FIG. 4 .

步骤105：根据技术方案中a.模板生成4提取时间段T内的合加速度特征序列与倾角特征序列。Step 105: According to a. template generation 4 in the technical solution, extract the combined acceleration feature sequence and inclination angle feature sequence within the time period T.

步骤106：判断满足上述步骤的跌倒行为次数是否为4，若是，则进行步骤107，反之进行步骤101。Step 106: Determine whether the number of falls satisfying the above steps is 4, if so, go to step 107, otherwise go to step 101.

步骤107：根据技术方案中a.模板生成5建立模板，该模板包括4次跌倒行为的特征序列以及两次跌倒行为模板间的平均距离DM。Step 107: Create a template according to a. template generation 5 in the technical solution, the template includes the feature sequence of 4 falls and the average distance DM between the two fall behavior templates.

如图2所示，为本发明提供的用于对用户运动进行跌倒检测流程图，具体步骤包括：As shown in FIG. 2 , it is a flow chart for detecting a fall of a user movement provided by the present invention, and the specific steps include:

步骤201：用户完成日常中可能出现的活动行为，并采集运动过程的三轴加速度数据。Step 201: The user completes activities that may occur in daily life, and collects three-axis acceleration data during the exercise process.

步骤202：对三轴加速度数据进行预处理。Step 202: Perform preprocessing on the three-axis acceleration data.

步骡203：提取该行为Y轴加速度峰值Y_max，根据技术方案中b.跌倒检测3进行判断，若状态标志位f₁为1，表明人体在竖直方向上的运动较为剧烈，执行步骤204，反之执行步骤201。Step 203: Extract the peak value Y _max of the Y-axis acceleration of the behavior, judge according to b. fall detection 3 in the technical solution, if the status flag f ₁ is 1, it indicates that the human body is moving more violently in the vertical direction, and execute step 204 , otherwise step 201 is executed.

步骤204：从Y轴加速度峰值出现时刻起T开始计时2s，获取T时间段内的预处理后的三轴加速度数据。Step 204: Start timing T for 2s from the moment when the Y-axis acceleration peak value appears, and acquire the preprocessed three-axis acceleration data within the T time period.

步骤205：提取T时间段内三轴加速度数据的合加速度特征序列与倾角特征序列。Step 205: Extracting the combined acceleration characteristic sequence and inclination angle characteristic sequence of the three-axis acceleration data within the T time period.

步骤206：根据技术方案中b.跌倒检测5将该行为的特征序列与模板库中的4组跌倒行为依次用DTW方法进行匹配，求得该行为与跌倒行为模板的平均距离DC。根据根据技术方案中b.跌倒检测6获取状态标志位f₂，若状态标志位f₂为1，则匹配成功，反之，匹配失败。Step 206: According to b. fall detection 5 in the technical solution, match the characteristic sequence of the behavior with the 4 groups of fall behaviors in the template library in turn using the DTW method to obtain the average distance DC between the behavior and the fall behavior template. According to b. fall detection 6 in the technical solution, the state flag f ₂ is obtained. If the state flag f ₂ is 1, the matching is successful; otherwise, the matching fails.

步骤207：匹配失败，该行为是非跌倒行为。Step 207: The matching fails, and this behavior is a non-falling behavior.

步骤208：匹配成功，该行为是跌倒行为。Step 208: the matching is successful, and this behavior is a falling behavior.

步骤209：跌倒行为，进行报警。Step 209: Fall behavior, alarm.

Claims

1. a tumble detection method for human body based on 3-axis acceleration sensor, it is characterised in that the method comprises template generation and fall detection two Part, specifically comprises the following steps that

A. template generation:

Step (1): user completes the behavior of falling, by X, Y, Z 3-axis acceleration data during acceleration transducer acquisition human motion；

Step (2): the original 3-axis acceleration data collected are carried out pretreatment；

Step (3): extract through pretreated Y-axis acceleration peak value Y_max, it is judged that Y_maxWhether exceed threshold value TH1, if Y_max＞ TH1, Then perform step (4)；Otherwise, perform step (1)；

Step (4): extract the feature of special time period, and generate matching template；

B. fall detection:

Step (5): user completes the behavior being likely to occur in daily life, obtains 3-axis acceleration data；

Step (6): the original 3-axis acceleration data obtained are carried out pretreatment；

Step (7): extract through pretreated Y-axis acceleration peak value Y_max, it is judged that Y_maxWhether exceed threshold value TH1, if Y_max＞ TH1, Then perform step (8), otherwise, perform step (5)；

Step (8): extract the feature of special time period；

Step (9): successively different time sequence in special time period is mated with the template of generation with DTW, it is judged that whether user falls.

A kind of tumble detection method for human body based on 3-axis acceleration sensor the most according to claim 1, it is characterised in that step (2) It is that smoothing denoising processes that original 3-axis acceleration to gather described with step (6) carries out pretreatment.

A kind of tumble detection method for human body based on 3-axis acceleration sensor the most according to claim 2, it is characterised in that described smoothing is gone Make an uproar and be processed as moving average filter.

A kind of tumble detection method for human body based on 3-axis acceleration sensor the most according to claim 1, it is characterised in that step (4) Human motion range parameter, human body attitude parameter is included with the feature of step (8) described extraction.

A kind of tumble detection method for human body based on 3-axis acceleration sensor the most according to claim 1, it is characterised in that generate coupling mould When plate and fall detection, gained 3-axis acceleration data after pretreatment extract characteristic sequence.

A kind of tumble detection method for human body based on 3-axis acceleration sensor the most according to claim 1, it is characterised in that step (4), Step (8) and step (9) described special time period be after Y-axis acceleration peak value occurs starting to occur to Y-axis acceleration peak value in 2s this time Between section.

A kind of tumble detection method for human body based on 3-axis acceleration sensor the most according to claim 1, it is characterised in that step (4), Matching template described in step (9) includes that 4 groups of human bodies are fallen the characteristic sequence of behavior, and fall behavior and the average distance of falling between behavior.