CN102551731A - Tumbling movement detecting method based on data curve comparison - Google Patents
Tumbling movement detecting method based on data curve comparison Download PDFInfo
- Publication number
- CN102551731A CN102551731A CN2011104365326A CN201110436532A CN102551731A CN 102551731 A CN102551731 A CN 102551731A CN 2011104365326 A CN2011104365326 A CN 2011104365326A CN 201110436532 A CN201110436532 A CN 201110436532A CN 102551731 A CN102551731 A CN 102551731A
- Authority
- CN
- China
- Prior art keywords
- curve
- characteristic
- matrix
- falling
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
A tumbling movement detecting method based on data curve comparison belongs to the creative application of pervasive computing human-machine interaction technology and introduces peripheral magnetic field change to reflect the movement situation of a body part in human body movements, grasps complex behavior characteristic information in the human body movements, achieves the detection of the tumbling movements on obtained sensing data curves by means of a curve comparison technical scheme of independent design based on curve shape context investigation, curve characteristic matrix generation and difference degree computing, and meets the real time high efficiency requirements of the movement detection in the pervasive computing application human-machine interaction.
Description
Technical field
The invention belongs to the general fit calculation applied technical field, relate to a kind of physical activity detection method in the human-computer interaction technology, this method effectively detects people's the activity of falling through to following the comparison process of the perception data curve that physical activity produces.
Background technology
Human body is fallen activity detection as a typical application of general fit calculation, also is important key technology in the man-machine interaction of facing movament.It has the basis and important meaning for avoiding falling to people's serious health hazard that especially old people brought, progressively in depth studying and accurately detect various physical activity action messages in the man-machine interaction.The activity detection of falling technical concerns will reasonably change quantitative description into to active qualitative description, judge with the identification that realizes detection system.The existing activity detection research work of falling mainly adopts three types of detection methods: based on data base's classification of motion identification detect, based on the detection of acceleration with based on the detection of Flame Image Process.
Follow the active behavioral data of various different human body to store among the data base and analyze based on what data base's classification of motion identification detection method will collect; To extract corresponding to the active different characteristic of difference; Thereby according to these characteristics user's behavioral activity is carried out Classification and Identification, judge whether the activity of falling.The activity detection of falling that is established as in physical activity perception data storehouse provides powerful data support; But its problem is to gather enough data, sets up data base's the very complicated and trouble of work itself for user's individuality, has influenced this working feasibility to a great extent.
The activity detection of falling based on acceleration is one type of more widely used method, and main dependence investigation acceleration information is provided with threshold value and detects.It gathers acceleration information through accelerometer sensor; Set up detection criterion according to the different disposal method of data and the different set mode of threshold value: perhaps according to the absolute value of acceleration peak value; Perhaps according to acceleration sum vector and, acceleration dynamic vector and, the dynamic indicators such as difference of normal acceleration and acceleration maximin, judge.The problem of this type of detection method is that the acceleration information variation all can produce in many physical activities, the activity detection of falling in view of the above is easy to generate bigger rate of false alarm, and the specificity of detection is relatively poor.
Use the method catch the physical activity image and to detect the action of falling of " visual " based on image processing techniques based on the activity detection of falling of Flame Image Process.But the acceptability of these class methods is not good with the property born; And detect owing to use image technique to carry out body movement; But surveyed area is limited in the very limited monitoring range fully, but the sky high cost of setting monitoring range has limited the expansion of surveyed area again.
Summary of the invention
The objective of the invention is to; In order to detect the activity of falling more effectively; Avoid the health hazard of falling the people being caused, the novel activity detection method of falling is provided, and then solve crucial physical activity detection technique problem in the man-machine interaction; Detect, distinguish the activity of falling and other conventional physical activities accurately and efficiently, improve " sensitivity " and " specificity " of testing result.
In order to realize the foregoing invention purpose, the present invention adopts following technical scheme.Should be based on the activity detection method of falling of data and curves comparison; Utilize the multiple sensors collection follow physical activity, comprise movable accurately real-time perception data minutia, that can be used for follow-up data curve processing and comparison; Form signal curve and investigate the shape facility of curve according to perception data, the active detection of relatively falling of the quantification of utilization curve.Different with the fall detection method that relies on acceleration signature; This method not only utilizes accelerometer sensor that the acceleration information of following physical activity is carried out record; Also novelty ground uses magnetic field induction sensor acquisition data; Place magnetic field induction pick off and micro magnetic adnexa in health effective site; Utilize magnetic field induction sensor senses surrounding magnetic field to change, draw the magnetic field strength date of reflection physical activity characteristic, embodied position and the variable in distance situation that is aprowl taken place between the body part of two kinds of equipment (pick off and magnetic attachment) of placing.
Should comprise a cover activity detection framework based on the activity detection method of falling of data and curves comparison; As shown in Figure 1; Be divided into relatively three parts of the generation of physical activity characteristic perception data and curves, the investigation of perception data curve shape context and perception data curve similarity; Some steps of the corresponding respectively activity detection method of falling, specific as follows:
Step 1: the real time data according to multiple sensors is gathered like acceleration information, bearing data, magnetic field strength date, generates the perception data curve.
Step 2: generate hereinafter rectangular histogram in shape according to the perception data curve.Particularly, on the perception data curve, choose a characteristic point
S, be that the center generates a log-polar system of containing every other point on the curve with the characteristic point of choosing,
RBe the ultimate range of coordinate system central point to other points, with coordinate system according to log
RValue is divided into equably
KdIndividual distance range, note is done
D1,
D2 ...,
Dk, and be divided into polar coordinate central angle 2 π
KaIndividual angular range, note is done
A1,
A2 ...,
AkLog-polar system is promptly with characteristic point
SFor the center is divided into
Kd*
KaThe individual histogram that contains entire curve, as shown in Figure 2.Remove characteristic point
SAny other points in addition all will (ρ θ) be in the some distances and angular range of coordinate system, promptly drops on characteristic point according to its polar coordinate value
SIn certain histogram for the center, the number of adding up point in each histogram just can obtain a little
SHereinafter in shape on this curve.
Step 3: the hereinafter rectangular histogram converts matrix notation in shape, i.e. the formation curve eigenmatrix.Feature points
SThe shape context-aware matrix
Ms, corresponding with histogram quantity, matrix
MsHave
Kd*
KaIndividual element, wherein element (
i,
j) value drop on coordinate system middle distance scope exactly and do
Di, angular range does
AjThe interval in the quantity put.Equally, to other characteristic points
XThe rectangular histogram of hereinafter in shape also can convert its shape context-aware matrix into
MXBecause the curved portion characteristic that it is the center that the hereinafter in shape of characteristic point has embodied with this point, so the matrix group of hereinafter in shape of a plurality of characteristic points is lumped together the characteristic that then can embody the whole piece curve.The shape context-aware matrix of all characteristic points splices and combines, and generates one
Nkd*
KaMatrix
M',
M' be the eigenmatrix of perception data curve, wherein
nBe the quantity of selected characteristic point.The curvilinear characteristic matrix can represent intuitively that as shown in Figure 3, the element value of dark more color showing correspondence is big more in the matrix square with the form of gray-scale map.The graphical representation of similar active curvilinear characteristic matrix is similar, and corresponding different active images then have than big-difference.
Step 4: after obtaining the curvilinear characteristic matrix, can be with the comparison that relatively is converted into its eigenmatrix of curve.The similarity of curvilinear characteristic matrix relatively can be accomplished quantitative Analysis through the Hausdorff distance measurement method to its graphical representation.The size of comparative result show each eigenmatrix institute corresponding data curve the degree of approximation situation---numerical value is more little, degree of approximation is high more.This compares numerical value also is that quantification is compared the perception data curve, judged the active output result that falls; For one group of corresponding unknown active perception data; It is organized the known corresponding active sample perception data of falling with another and compares; Comparative result numerical value as if between two perception data curves is enough little, then judges the unknown movable activity of falling that is.
Beneficial effect of the present invention is; This activity detection method of falling has combined the body movement The Characteristics; Introduce surrounding magnetic field novelly and change the movements of parts of the body situation in the physical activity that embodies; Hold the complex behavior characteristic information in the physical activity, and to the perception data curve negotiating autonomous Design that obtains based on the curve shape context is investigated, the curvilinear characteristic matrix generates and diversity factor is calculated curve ratio than technical scheme, the active real-time high-efficiency of realizing falling detects; Overcome and used simply based on the acceleration rate threshold condition when activity of falling is judged; Can't embody the activity of falling and the active characteristic difference of normal human comprehensively, detect the difficulty that effect " sensitivity " and " specificity " requirement is difficult to be in harmonious proportion, satisfy the demand of activity detection in the general fit calculation practical application of falling.
Description of drawings
The activity detection method of falling that Fig. 1 is based on the data and curves comparison detects the framework sketch map.
Fig. 2 is a curve shape context log-polar histogram sketch map.
Fig. 3 is the gray-scale map graphical representation of curvilinear characteristic matrix.
The specific embodiment
In this falls the activity detection method; The perception data of magnetic field induction pick off has been represented the magnetic field intensity around it; Use the relative position between this data value deducibility pick off and the magnetic attachment; Through the position of pick off and magnetic attachment rationally is set, can catch the common characteristic of body part motion in the activity of falling effectively, fall to realize detecting.Concrete, can magnetic attachment be placed on the lower limb, a little more than knee, simultaneously pick off is placed near the position trousers pocket on the opposite side lower limb.The magnetic field intensity perception data that pick off is gathered has characterized the motion conditions between two lower limbs in the physical activity.
After obtaining comprising the perception data of physical activity characteristic and generating data and curves; Promptly investigate the hereinafter in shape of perception data curve based on the activity detection method of falling of data and curves comparison; Process comprises choosing of hereinafter characteristic point in shape---confirm the quantity and the position of characteristic point, and the interval division of log-polar system.Because the people is in the process of falling, the shank posture often has significant variation, therefore can cause magnetic field intensity and perception data clocklike to change.Data and curves will be considered these Changing Patterns during selected characteristic point during hereinafter is investigated in shape.Originally the activity detection of falling method is selected 2 " Bottom Of Descents " in a pair of " descending continuously " as the characteristic point of data and curves as.
" descend continuously " is meant: if the perception data curve shows that reading is at [t1, t2] in the time period behind once dull the decline, and then at [t2; T3] the once dull rising of interior appearance of time period, then again in [t3, t4] dull once more decline in the time period; Drop to successive twice that then claims to occur in time period [t1, t2] and [t3, t4]; Promptly a pair of " descending continuously ", the minimum point of 2 declines is called as " Bottom Of Descent " respectively in descending continuously.
With the data and curves characteristic point is that central point generates log-polar system, in this coordinate system, by the logarithm value log of other nodes at a distance of central point ultimate range
REvenly divide 5 distance ranges, note is done
D1,
D2 ...,
D5, simultaneously central angle 2 π are divided into 12 angular ranges, note is done
A1,
A2 ...,
A12.
In carrying out based on the activity detection method of falling of data and curves comparison, at first calculate in the perception data curve in every pair " descending continuously " accumulative total of twice decline with, check whether it surpasses threshold value
Thtm,, then carry out subsequent operation immediately in back 1.6 seconds at " Bottom Of Descent " of the 2nd decline if check result be true; Subsequent operation does; Be characteristic point with 2 Bottom Of Descents in current " descending continuously " respectively; The hereinafter in shape of calculated curve, and combination formation curve eigenmatrix, concrete; Be that log-polar is a central point with 2 characteristic points respectively; On the data and curves arbitrarily other points all fall into a particular distance scope and angular range according to its polar coordinate value (by representing) apart from ρ and angle θ, just in the interval that log-polar is, thereby obtain one 5 * 12 shape context-aware matrix
MX, in the matrix (
i,
j) representative of locational value falls into distance range
DiAnd angular range
AjThe quantity of point.The matrix of hereinafter in shape combination with 2 characteristic points generate respectively obtains one 10 * 12 curvilinear characteristic matrix
MThen with this curvilinear characteristic matrix
MRepresent with gray-scale map, and calculate itself and corresponding Hausdorff distance of falling between the active known features matrix, check that whether it is less than threshold value
Thhd, if the result is true, then the corresponding activity of the current perception data curve of labelling is " falling ", otherwise is labeled as " non-falling ".
Claims (3)
1. activity detection method of falling based on the data and curves comparison; It is characterized in that the perception data of following physical activity of sensor acquisition is handled; Adopt hereinafter investigation perception data curve in shape; The formation curve eigenmatrix representes with image that also the diversity factor of utilizing the Hausdorff distance to carry out the eigenmatrix image is relatively calculated, through detecting with known relatively realization of the falling active perception data curvilinear characteristic matrix image active real-time high-efficiency of falling.
2. a kind of activity detection method of falling according to claim 1 based on the data and curves comparison; It is characterized in that; Utilize the variable in distance between surrounding magnetic field Strength Changes reflection magnetic field induction pick off and the magnetic attachment; Thereby movements of parts of the body situation in the embodiment physical activity, the magnetic field intensity perception data is also gathered in the position through pick off and magnetic attachment rationally are set, and holds the complex behavior characteristic information in the activity of falling.
3. a kind of activity detection method of falling based on the data and curves comparison according to claim 1 may further comprise the steps:
Step 1: the real time data according to multiple sensors is gathered like acceleration information, bearing data, magnetic field strength date, generates the perception data curve;
Step 2: generate hereinafter rectangular histogram in shape according to the perception data curve; Particularly, on the perception data curve, choose a characteristic point
S, be that the center generates a log-polar system of containing every other point on the curve with the characteristic point of choosing,
RBe the ultimate range of coordinate system central point to other points, with coordinate system according to log
RValue is divided into equably
KdIndividual distance range, note is done
D1,
D2 ...,
Dk, and be divided into polar coordinate central angle 2 π
KaIndividual angular range, note is done
A1,
A2 ...,
AkLog-polar system is promptly with characteristic point
SFor the center is divided into
Kd*
KaThe individual histogram that contains entire curve is removed characteristic point
SAny other points in addition all will (ρ θ) be in the some distances and angular range of coordinate system, promptly drops on characteristic point according to its polar coordinate value
SIn certain histogram for the center, the number of adding up point in each histogram just can obtain a little
SHereinafter in shape on this curve;
Step 3: the hereinafter rectangular histogram converts matrix notation in shape, i.e. the formation curve eigenmatrix; Feature points
SThe shape context-aware matrix
Ms, corresponding with histogram quantity, matrix
MsHave
Kd*
KaIndividual element, wherein element (
i,
j) value drop on coordinate system middle distance scope exactly and do
Di, angular range does
AjThe interval in the quantity put; Equally, to other characteristic points
XThe rectangular histogram of hereinafter in shape also can convert its shape context-aware matrix into
MXBecause the curved portion characteristic that it is the center that the hereinafter in shape of characteristic point has embodied with this point, so the matrix group of hereinafter in shape of a plurality of characteristic points is lumped together the characteristic that then can embody the whole piece curve; The shape context-aware matrix of all characteristic points splices and combines, and generates one
Nkd*
KaMatrix
M',
M' be the eigenmatrix of perception data curve, wherein
nBe the quantity of selected characteristic point;
Step 4: after obtaining the curvilinear characteristic matrix, can be with the comparison that relatively is converted into its eigenmatrix of curve.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011104365326A CN102551731B (en) | 2011-12-23 | 2011-12-23 | Tumbling movement detecting method based on data curve comparison |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011104365326A CN102551731B (en) | 2011-12-23 | 2011-12-23 | Tumbling movement detecting method based on data curve comparison |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102551731A true CN102551731A (en) | 2012-07-11 |
CN102551731B CN102551731B (en) | 2013-12-25 |
Family
ID=46399305
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2011104365326A Active CN102551731B (en) | 2011-12-23 | 2011-12-23 | Tumbling movement detecting method based on data curve comparison |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102551731B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107126217A (en) * | 2017-06-05 | 2017-09-05 | 河池学院 | Wearable falls down detection device |
CN108670261A (en) * | 2018-04-12 | 2018-10-19 | 深圳先进技术研究院 | Motion state detection method, wearable device and device |
CN108778120A (en) * | 2015-08-06 | 2018-11-09 | 巴黎笛卡尔大学 | Method for characterizing gait |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101584584A (en) * | 2009-07-01 | 2009-11-25 | 北京派瑞根科技开发有限公司 | Burst disease analyzing method based on motion sensor and biosensor |
WO2010126878A1 (en) * | 2009-04-27 | 2010-11-04 | Mayo Foundation For Medical Education And Research | Fall detection |
US20110201972A1 (en) * | 2008-10-17 | 2011-08-18 | Koninklijke Philips Electronics N.V. | fall detection system and a method of operating a fall detection system |
-
2011
- 2011-12-23 CN CN2011104365326A patent/CN102551731B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110201972A1 (en) * | 2008-10-17 | 2011-08-18 | Koninklijke Philips Electronics N.V. | fall detection system and a method of operating a fall detection system |
WO2010126878A1 (en) * | 2009-04-27 | 2010-11-04 | Mayo Foundation For Medical Education And Research | Fall detection |
CN101584584A (en) * | 2009-07-01 | 2009-11-25 | 北京派瑞根科技开发有限公司 | Burst disease analyzing method based on motion sensor and biosensor |
Non-Patent Citations (1)
Title |
---|
印勇: "基于改进Hu矩的异常行为识别", 《计算机技术与发展》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108778120A (en) * | 2015-08-06 | 2018-11-09 | 巴黎笛卡尔大学 | Method for characterizing gait |
CN108778120B (en) * | 2015-08-06 | 2021-03-16 | 巴黎笛卡尔大学 | Method for characterizing gait |
CN107126217A (en) * | 2017-06-05 | 2017-09-05 | 河池学院 | Wearable falls down detection device |
CN108670261A (en) * | 2018-04-12 | 2018-10-19 | 深圳先进技术研究院 | Motion state detection method, wearable device and device |
CN108670261B (en) * | 2018-04-12 | 2021-10-15 | 深圳先进技术研究院 | Motion state detection method, wearable device and apparatus |
Also Published As
Publication number | Publication date |
---|---|
CN102551731B (en) | 2013-12-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Johnston et al. | Smartwatch-based biometric gait recognition | |
CN205041401U (en) | Equipment is worn detection device and is had and wears guardianship device that detects function | |
CN105678222A (en) | Human behavior identification method based on mobile equipment | |
CN105139029B (en) | A kind of Activity recognition method and device of prison prisoner | |
CN104586398A (en) | Old man falling detecting method and system based on multi-sensor fusion | |
CN105956526A (en) | Method for identifying a microearthquake event with low signal-to-noise ratio based on multi-scale permutation entropy | |
CN101695445A (en) | Acceleration transducer-based gait identification method | |
CN106353015A (en) | Civil-engineering construction monitoring system | |
CN103886715A (en) | Human body fall detection method | |
CN105224104A (en) | Pedestrian movement's state identification method of mode is held based on smart mobile phone | |
JP2014171041A (en) | Method for determining place in which portable device is carried, portable device, system for determining place in which portable device is carried, program, and information storage medium | |
CN103559508A (en) | Video vehicle detection method based on continuous Adaboost | |
CN106650300B (en) | Old man monitoring system and method based on extreme learning machine | |
CN102551731A (en) | Tumbling movement detecting method based on data curve comparison | |
CN111475030B (en) | Micro-gesture recognition method using near-infrared sensor | |
CN107003900A (en) | A kind of switching method and portable electric appts | |
Lee et al. | A smart device using low-cost sensors to detect earthquakes | |
Tian et al. | An improved activity recognition method based on smart watch data | |
CN117441980B (en) | Intelligent helmet system and method based on intelligent computation of multi-sensor information | |
CN101782964A (en) | Weight loss feather extraction method based on acceleration transducer | |
CN101673449A (en) | Method for detecting operation of worker and alarming based on three-dimensional position sensing device | |
CN113378702A (en) | Multi-feature fusion fatigue monitoring and identifying method for pole climbing operation | |
CN103826202A (en) | Method for improving WiFi positioning result jumping based on cell phone sensor information | |
KR20170048108A (en) | Method and system for recognizing object and environment | |
CN105848101A (en) | Old people home care monitoring system based on indoor positioning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |