CN108629304B - Freezing gait online detection method - Google Patents

Freezing gait online detection method Download PDF

Info

Publication number
CN108629304B
CN108629304B CN201810386698.3A CN201810386698A CN108629304B CN 108629304 B CN108629304 B CN 108629304B CN 201810386698 A CN201810386698 A CN 201810386698A CN 108629304 B CN108629304 B CN 108629304B
Authority
CN
China
Prior art keywords
gait
frozen
offline
online
normal
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.)
Active
Application number
CN201810386698.3A
Other languages
Chinese (zh)
Other versions
CN108629304A (en
Inventor
赵金
任康
施翼
凌云
陈仲略
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Gyenno Technologies Co ltd
Original Assignee
Gyenno Technologies Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Gyenno Technologies Co ltd filed Critical Gyenno Technologies Co ltd
Priority to CN201810386698.3A priority Critical patent/CN108629304B/en
Publication of CN108629304A publication Critical patent/CN108629304A/en
Application granted granted Critical
Publication of CN108629304B publication Critical patent/CN108629304B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G06V40/23—Recognition of whole body movements, e.g. for sport training
    • G06V40/25—Recognition of walking or running movements, e.g. gait recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pathology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The invention relates to the technical field of machine learning, in particular to an on-line detection method for frozen gait. The method comprises the following steps: acquiring offline gait data and an offline gait video of a patient in an offline walking process, wherein the offline gait data comprises thigh acceleration, thigh angular velocity, shank acceleration, shank angular velocity and plantar pressure; establishing an offline sample set according to the offline gait data and the offline gait video; constructing a naive Bayes classifier of an offline normal gait and an offline frozen gait; and obtaining the probabilities of the on-line normal gait and the on-line frozen gait to obtain detection results in the on-line walking process respectively through the naive Bayes classifier of the off-line normal gait and the off-line frozen gait and the on-line gait data. The method can quickly and accurately detect the frozen gait in real time and help the sick patient to recover the walking and normal activities in time.

Description

Freezing gait online detection method
Technical Field
The invention relates to the technical field of machine learning, in particular to an on-line detection method for frozen gait.
Background
Parkinson's Disease (PD) is a common degenerative disease of the nervous system, and its main clinical features include reduced motor retardation, rigidity of limbs, and tremor, and possibly some abnormalities in posture and gait. The disease is frequently seen in the elderly, and the average age of onset is about 60 years, and in recent years, the age of onset of the disease tends to decline.
Gait Freezing (hereinafter referred to as FOG) is the most common symptom Of patients with advanced parkinson disease, and is mainly manifested as transient block Of movement, and sudden onset Of FOG can cause walking disorder and even fall Of patients, and reduce life quality Of the patients. Therefore, it is important to detect the gait of the PD patient in real time during walking and to give the PD certain intervention guidance when FOG is detected.
In the related art, there are two main detection methods for frozen gait: 1. detecting the energy ratio of the gait data sequence in a normal movement frequency band and an FOG frequency band, and judging whether the gait is normal or not according to a manually set freezing threshold value; 2. and detecting the Pearson correlation coefficient of the gait and the normal gait of the patient in the walking process, and judging whether the gait is normal or not according to the correlation.
The inventor finds that the two detection methods both need to manually preset a threshold value, and the accuracy is not high; and in the above two methods, when the feature space has a high dimension, it becomes very cumbersome to manually set the freeze threshold.
Disclosure of Invention
The invention aims to solve the technical problem that the existing detection method for frozen gait is not accurate enough, and in order to solve the technical problem, the implementation mode of the invention adopts a technical scheme that: the method for detecting the frozen gait on line comprises the following steps: acquiring offline gait data and an offline gait video of a patient in an offline walking process, wherein the offline gait data comprises thigh acceleration, thigh angular velocity, shank acceleration, shank angular velocity and plantar pressure; establishing an offline sample set according to the offline gait data and the offline gait video; respectively constructing a naive Bayes classifier of an offline normal gait and an offline frozen gait based on the offline sample set; acquiring online gait data in an online walking process; obtaining the probabilities of the online normal gait and the online frozen gait in the online walking process respectively through the naive Bayes classifier of the offline normal gait and the offline frozen gait and the online gait data; and comparing the probabilities of the online normal gait and the online frozen gait to obtain a detection result.
Optionally, the establishing an offline sample set according to the offline gait data and the offline gait video specifically includes: dividing the off-line gait data and the off-line gait video at a preset time length and a fixed time interval to obtain a plurality of window data and window videos; according to the window video corresponding to the window data, marking the gait type of the window data, wherein the gait type is a frozen gait or a normal gait; extracting corresponding feature vectors from the window data; and establishing an offline sample set according to the gait type of the window data and the characteristic vector.
Optionally, the thigh acceleration comprises: thigh X-axis acceleration, thigh Y-axis acceleration, and thigh Z-axis acceleration; the thigh angular velocity includes: thigh X-axis angular velocity, thigh Y-axis angular velocity, and thigh Z-axis angular velocity; the shank acceleration is: calf X-axis acceleration, calf Y-axis acceleration and calf Z-axis acceleration; the lower leg angular velocity comprises: the shank X-axis angular velocity, the shank Y-axis angular velocity and the shank Z-axis angular velocity; the sole pressure comprises a first node pressure, a second node pressure and a third node pressure which are uniformly distributed on the sole.
Optionally, the extracting the corresponding feature vector from the window data specifically includes:
calculating to obtain a first feature vector by the following formula;
Figure BDA0001642419730000021
wherein FI is a first feature vector; w (t, f) is a frequency domain signal after a Short Time Fourier Transform (STFT) of the thigh acceleration, the calf acceleration or the plantar pressure W (t);
calculating the energy E of the frequency domain signal in the [3Hz,8Hz ] frequency band as a second feature vector by the following formula:
Figure BDA0001642419730000031
taking the thigh angular velocity or the shank angular velocity as a third feature vector;
calculating the angle theta of the thigh or the calf respectively deviating from the vertical direction as a fourth eigenvector by the following formula
Figure BDA0001642419730000032
Wherein a is thigh acceleration or shank acceleration, and g is gravity acceleration.
Optionally, the constructing a naive bayes classifier for an offline normal gait and an offline frozen gait based on the offline sample set respectively specifically includes: counting the average value and the standard deviation of all the feature vectors of each window data; calculating prior probabilities of the normal gait and the frozen gait in the offline sample set; constructing a naive Bayes classifier of the normal gait according to the prior probability of the normal gait and the average value and the standard value of all the feature vectors of the window data marked as the normal gait; and constructing a naive Bayes classifier of the frozen gait according to the prior probability of the frozen gait and the average value and the standard value of all the feature vectors of the window data marked as the frozen gait.
Optionally, the naive bayes classifier is represented by the following equation:
Figure BDA0001642419730000033
wherein, P (x)i| c) is the conditional probability that the ith feature vector belongs to a frozen gait or a normal gait, and P (c) is the prior probability of the normal gait or the frozen gait.
Alternatively, P (x)i| c) is calculated by the following equation:
Figure BDA0001642419730000034
wherein x isiIs the ith feature vector, muc,iIs the mean value, σc,iIs the standard value.
Optionally, the acquiring online gait data in the online walking process specifically includes:
and acquiring online gait data of the detection window according to the preset time length and the fixed time interval.
Optionally, the obtaining, by the naive bayes classifier and the online gait data, probabilities of an online normal gait and an online frozen gait during an online walking process respectively includes: calculating a feature vector corresponding to the detection window; calculating the probability that the detection window is in an online normal gait through a naive Bayes classifier of the normal gait according to the feature vector; and calculating the probability that the detection window is in the online frozen state through a naive Bayes classifier of the frozen gait according to the feature vector.
Optionally, the comparing the probabilities of the online normal gait and the online frozen gait to obtain a detection result specifically includes: judging whether the probability of the online normal pace is larger than that of the online normal pace; if so, determining that the detection result is normal gait; if not, determining that the detection result is the frozen gait.
The freezing gait on-line detection method provided by the embodiment of the invention can quickly and accurately detect the freezing gait in real time through the naive Bayes classifier constructed in the off-line process, and the detection result can be used as a basis for implementing subsequent guidance measures for patients on one hand and can help the sick patients to recover walking and normal activities in time; on the other hand, the detection result can also provide the information of symptoms related to the frozen gait of the patient, and has important guiding function on the research and treatment of the frozen gait.
Drawings
Fig. 1 is a schematic flow chart of an online detection method for frozen gait according to an embodiment of the present invention;
FIG. 2 is a fixed position of two six-axis inertial sensors and three thin film pressure sensors provided by an embodiment of the present invention;
FIG. 3 is a schematic flow chart of step 12 of FIG. 1;
fig. 4 is a schematic diagram of an on-line detection process of frozen gait according to another embodiment of the invention.
Detailed Description
In order to make the objects, aspects and advantages of the present invention more apparent, the present invention will be described in further detail with reference to examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the different embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In order to better understand the frozen gait online detection method provided by the following embodiments of the invention, the following first briefly introduces the construction principle of the naive bayes classifier.
The naive Bayes classifier is an algorithm based on Bayes theory in a classification algorithm set. It is not a single occurrence, but rather a family of algorithms in which they all share common rules. For example, one of the common rules in this family of algorithms is that each classified feature vector is independent of the other feature vectors; for example, another common rule in this family of algorithms is that each feature has the same weight; thus any feature is relevant to the result and the degree of influence is the same. According to the algorithm rule, a proper feature vector is selected, and an offline sample set is established, so that the Bayesian classifier can be constructed. The embodiment of the invention mainly aims at the detection of the frozen gait, so that the selection of the proper type of feature vector and the establishment of the off-line sample according to the feature vector in the walking process of the patient are very important, and the method is an important basis for accurately detecting the frozen gait.
Therefore, the embodiment of the invention provides an online detection method for frozen gait, which comprises the steps of collecting gait data of different classifications of a patient in a walking process in an offline state, extracting feature vectors of corresponding types based on the data of the different classifications (the extracted feature vectors are independent of each other), establishing an offline sample set according to features corresponding to the feature vectors, and finally constructing a naive Bayes classifier capable of accurately detecting the frozen gait. When a patient walks in real time, gait data in the real-time walking process of the patient is collected, the probability of the frozen gait in the walking process of the patient can be accurately judged on line, and the judgment result can provide corresponding symptom information for the frozen gait of the patient on one hand and can serve as a real-time basis of subsequent guiding measures on the other hand, so that the method has an important guiding function on the research and treatment of the frozen gait.
Referring to fig. 1, fig. 1 is a schematic flow chart of an online detection method for frozen gait according to an embodiment of the present invention, as shown in fig. 1, the method includes:
step 11, acquiring off-line gait data and off-line gait videos of a patient in an off-line walking process, wherein the off-line gait data comprises thigh acceleration, thigh angular velocity, shank acceleration, shank angular velocity and plantar pressure.
The off-line gait data is data which can reflect the walking characteristics of the patient when the frozen gait occurs, and in the embodiment of the invention, the off-line gait data is divided into 15 types which are respectively used in the walking process of the patient:
1. acceleration and angular velocity of the knee in the X, Y and Z axes;
2. acceleration and angular velocity of the ankle in the X, Y and Z axes;
3. pressure of 3 nodes on the sole of the foot.
The off-line gait data can be acquired by adopting a corresponding sensor, the sensor further transmits the acquired gait data to the processor, and the processor can execute the steps of the method according to the off-line gait data.
Wherein, in order to collect the accelerations and angular velocities of the knee and ankle of the patient in the X-axis, Y-axis and Z-axis, two six-axis inertial sensors may be fixed at the knee and ankle of the right leg of the patient in the manner shown in fig. 2; in order to collect the pressure of 3 nodes on the sole of the foot during the gait and walking process of the patient, three film pressure sensors can be fixed on the right sole of the patient according to the mode shown in fig. 2.
The experimental data collected by the six-axis inertial sensor and the film pressure sensor in the off-line walking process of the patient is used as off-line gait data in the embodiment of the invention. In other embodiments, the number and the positions of the sensors can be reasonably adjusted according to the severity of the legs or the left and right legs of the patient, so as to acquire the off-line gait data more accurately.
Because the offline gait data is mainly used for constructing the naive Bayes classifier, when the offline gait data of the patient is collected, an offline gait video of the patient in an offline walking process needs to be recorded through related video equipment, and whether the gait of the patient is a frozen gait can be visually observed from the offline gait video to serve as an important parameter for subsequently constructing the naive Bayes classifier.
And step 12, establishing an offline sample set according to the offline gait data and the offline gait video.
After the offline gait data and the offline gait video are obtained, segmentation and feature vector extraction can be further performed on the offline gait data and the offline gait video respectively to establish an offline sample set, wherein the offline sample set is a basic parameter required for constructing a naive Bayes classifier.
Specifically, as shown in fig. 3, step 12 includes:
and step 121, dividing the off-line gait data and the off-line gait video according to a preset time length and a fixed time interval to obtain a plurality of window data and window videos.
The window data and the window videos are in one-to-one correspondence, and the feature vector extracted from each window data is a feature value and is used as a part for establishing an offline sample set; and each window video can intuitively judge whether the corresponding gait type is normal gait or frozen gait.
And step 122, marking the gait type of the window data according to the window video corresponding to the window data, wherein the gait type is a frozen gait or a normal gait.
The step is the process of judging and recording the gait type corresponding to each window video.
And step 123, extracting corresponding characteristic vectors from the window data.
In the step, four mutually independent characteristic vectors which are most discriminative for normal gait and frozen gait are extracted from the window data, and the calculation modes are respectively as follows:
calculating to obtain a first feature vector by the following formula;
Figure BDA0001642419730000071
wherein FI is a first feature vector; w (t, f) is a frequency domain signal after a Short Time Fourier Transform (STFT) of the thigh acceleration, the calf acceleration or the plantar pressure W (t);
calculating the energy E of the frequency domain signal in the [3Hz,8Hz ] frequency band as a second feature vector by the following formula:
Figure BDA0001642419730000072
taking the thigh angular velocity or the shank angular velocity as a third feature vector;
calculating the angle theta of the thigh or the calf respectively deviating from the vertical direction as a fourth eigenvector by the following formula
Figure BDA0001642419730000073
Wherein a is thigh acceleration or shank acceleration, and g is gravity acceleration.
And step 124, establishing an offline sample set according to the gait type of the window data and the feature vector.
The offline sample set includes an offline sample set of normal gait and an offline sample set of frozen gait. All the eigenvectors are included in the sample set, and in this embodiment, the ith eigenvector is denoted as xiThe i feature vectors are independent from each other and jointly form an offline sample set of normal gait or frozen gait, and in the embodiment, 22 feature vectors are calculated and obtained from the offline sample set of normal gait or frozen gait.
And step 13, respectively constructing a naive Bayes classifier of an offline normal gait and an offline frozen gait based on the offline sample set.
The naive bayes classifier is represented by the following equation:
Figure BDA0001642419730000081
wherein, P (x)i| c) is the conditional probability that the ith feature vector belongs to a frozen gait or a normal gait, and P (c) is the prior probability of the normal gait or the frozen gait.
Specifically, P (x)i| c) is calculated by the following equation:
Figure BDA0001642419730000082
therefore, constructing a naive Bayes classifier requires calculating normal gait window data or freezing gait window data at the ith characteristic value xiAverage value mu of the above valuesc,iAnd standard deviation σc,i。
Wherein the mean value μc,iAs shown in the following equation:
Figure BDA0001642419730000083
in the formula, sigma Vc,iThe sum of values of the normal gait window data or the abnormal gait window data on the ith characteristic value is represented, nc,iThe number of the normal gait window data or the abnormal gait window data is represented.
Wherein, the standard deviation sigmac,iAs shown in the following equation:
Figure BDA0001642419730000084
in this embodiment, the prior probability that the ith feature vector belongs to the frozen gait or the normal gait can be calculated through a window video, and the calculation formula is as follows:
Figure BDA0001642419730000091
wherein D represents the total number of window videos in the offline sample set, DcRepresenting the number of normal or abnormal gait window videos in the offline sample set.
The average value mu of the normal gait and the frozen gait is obtained by the stepsc,iStandard deviation σc,iAfter the prior probability is obtained, a naive Bayes classifier of a normal gait and a frozen gait can be respectively constructed; after the naive Bayes classifier is constructed, the naive Bayes classifier of the normal gait and the naive Bayes classifier of the frozen gait can be applied to the following online detection stage.
And step 14, acquiring online gait data in the online walking process.
The online gait data of the detection window is collected at preset time length and fixed time intervals.
And step 15, obtaining the probabilities of the online normal gait and the online frozen gait in the online walking process respectively through the naive Bayes classifier of the offline normal gait and the offline frozen gait and the online gait data.
And step 16, comparing the probabilities of the online normal gait and the online frozen gait to obtain a detection result.
Based on the naive Bayes classifier constructed in the step 13, calculating the probability that the window data at the moment is not in the frozen state
Figure BDA0001642419730000092
Is that the probability of freezing the step is
Figure BDA0001642419730000093
When P is present1>P2When the detection result is frozen gait, when P is1<P2And meanwhile, the detection result is normal gait.
The freezing gait on-line detection method provided by the embodiment of the invention can quickly and accurately detect the freezing gait of the Parkinson in real time through the naive Bayes classifier constructed in the off-line process, and the detection result can be used as a basis for implementing subsequent guidance measures for patients on one hand and can help the sick patients to recover walking and normal activities in time; on the other hand, the detection result can also provide the information of symptoms related to the frozen gait of the patient, and has important guiding function on the research and treatment of the frozen gait.
The process of on-line detection of frozen gait will be described in detail with a specific embodiment, and as shown in fig. 4, the process specifically includes: and (4) offline learning, establishing an offline sample set and online detecting to obtain a detection result.
1. The off-line learning and the establishment of the off-line sample set comprise the following processes:
1.1, 2 six-axis sensors and 3 acceleration sensors are fixed on the knee, ankle and sole of the right leg of the patient in the manner shown in fig. 2. The method comprises the steps of collecting acceleration data, angular velocity data and plantar pressure data of knees and ankles of a patient in the daily walking process as offline gait data, and simultaneously recording a process video as an offline gait video, wherein the sampling frequency of sensors is 100Hz, and 15 types of offline gait data measured by the sensors are shown in table 1.
TABLE 1 sensor data types
Acceleration n of right thigh X axis1 Acceleration n of right thigh Y axis2 Acceleration n of right thigh Z axis3
Acceleration n of right shank X axis4 Acceleration n of right shank Y axis5 Acceleration n of Z axis of right shank6
Angular velocity n of right thigh X axis7 Right thigh Y axis angular velocity n8 Right thigh Z axis angular velocity n9
Angular velocity n of right shank X axis10 Angular velocity n of right shank Y axis11 Angular velocity n of Z axis of right shank12
Right ball node 1 pressure n13 Right ball node 2 pressure n14 Right ball node 3 pressure n15
1.2, dividing the off-line gait data in the step 1.1 by a sliding time window with a time interval of 100ms and a width of 256ms to obtain 39606 window data and 39606 window videos. Then, the corresponding off-line gait video is compared, and the gait types corresponding to the window data are marked (the normal gait is marked as c)1Marker for frozen gait c2) Finally, 25536 window videos corresponding to normal gait and 14070 window videos corresponding to frozen gait are obtained.
1.3, extracting and obtaining the gait related feature vector as follows: the FI index, the energy of the frozen frequency band, the angular velocity of the big leg and the small leg, and the included angle of the big leg and the small leg from the vertical direction are calculated in the manner according to the above embodiments, which is not described herein again, the number of all the eigenvectors included in the offline sample set is calculated to be 22, and the ith eigenvector is recorded as xiThe i feature vectors are independent of each other and jointly form an offline sample set, as shown in table 2.
TABLE 2
Figure BDA0001642419730000111
And 1.4, calculating related parameters and constructing a naive Bayes classifier. The process computes two parameters of a naive bayes classifier.
The first parameter is as follows: normal gait window data and frozen gait window data are in ith characteristic value xiAverage value mu of the above valuesc,iAnd standard deviation σc,i，μc,iThe calculation method of (c) is shown by the following formula:
Figure BDA0001642419730000112
σc,ithe calculation method of (c) is shown by the following formula:
Figure BDA0001642419730000113
in the formula, sigma Vc,iRepresents the sum of values of the normal gait window data or the frozen gait window data on the ith characteristic value, nc,iRepresenting normal gait window video orNumber of abnormal gait window videos.
And a second parameter: the prior probability of normal gait and frozen gait p (c). As shown in the following equation:
Figure BDA0001642419730000114
wherein D represents the total number of windows in the offline sample set, and DcRepresenting the number of normal gait or frozen gait window videos in the offline sample set. In the above embodiment, it has been found that 25536 normal gait corresponding window videos and 14070 frozen gait corresponding window videos are obtained, and then the prior probability of the normal gait is:
Figure BDA0001642419730000121
the prior probability of freezing gait is:
Figure BDA0001642419730000122
1.5, constructing a naive Bayes classifier
Figure BDA0001642419730000123
Wherein P (x)i| c) is the class conditional probability of the ith characteristic quantity, and is shown by the following formula:
Figure BDA0001642419730000124
and constructing a calculation formula of the naive Bayes classifier according to the first parameter and the second parameter.
2. The process of obtaining the detection result by online detection comprises the following steps:
2.1, collecting window data in real time, and collecting the window data at intervals of 100ms and sliding time windows with widths of 256 ms.
2.2 calculating each eigenvector in Table 2 for the window data, and recording the value of the ith eigenvector as xi。
2.3, calculating the window data in the class c sample (normal gait c)1Freezing stepState c2) Class conditional probability P (x) at ith eigenvaluei|c)。
2.4, based on the naive Bayes classifier constructed in the step 1.5, calculating the probability that the window data at the moment is not in the frozen state as
Figure BDA0001642419730000125
Is that the probability of freezing the step is
Figure BDA0001642419730000126
When P is present1>P2When the detection result is frozen gait, when P is1<P2And meanwhile, the detection result is normal gait.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (8)

1. An on-line detection method for frozen gait, which is characterized by comprising the following steps:
acquiring offline gait data and an offline gait video of a patient in an offline walking process, wherein the offline gait data comprises thigh acceleration, thigh angular velocity, shank acceleration, shank angular velocity and plantar pressure;
dividing the off-line gait data and the off-line gait video at a preset time length and a fixed time interval to obtain a plurality of window data and window videos;
according to the window video corresponding to the window data, marking the gait type of the window data, wherein the gait type is a frozen gait or a normal gait;
calculating to obtain a first feature vector by the following formula;
Figure FDA0002701358280000011
wherein FI is a first feature vector; w (t, f) is a frequency domain signal after a Short Time Fourier Transform (STFT) of the thigh acceleration, the calf acceleration or the plantar pressure W (t);
calculating the energy E of the frequency domain signal in the [3Hz,8Hz ] frequency band as a second feature vector by the following formula:
Figure FDA0002701358280000012
taking the thigh angular velocity or the shank angular velocity as a third feature vector;
calculating the angle theta of the thigh or the calf respectively deviating from the vertical direction as a fourth eigenvector by the following formula
Figure FDA0002701358280000013
Wherein a is thigh acceleration or shank acceleration, and g is gravity acceleration;
establishing an offline sample set according to the gait type of the window data and the characteristic vector;
respectively constructing a naive Bayes classifier of an offline normal gait and an offline frozen gait based on the offline sample set;
acquiring online gait data in an online walking process;
obtaining the probabilities of the online normal gait and the online frozen gait in the online walking process respectively through the naive Bayes classifier of the offline normal gait and the offline frozen gait and the online gait data;
and comparing the probabilities of the online normal gait and the online frozen gait to obtain a detection result.
2. The on-line detection method according to claim 1, wherein the thigh acceleration includes: thigh X-axis acceleration, thigh Y-axis acceleration, and thigh Z-axis acceleration;
the thigh angular velocity includes: thigh X-axis angular velocity, thigh Y-axis angular velocity, and thigh Z-axis angular velocity;
the shank acceleration is: calf X-axis acceleration, calf Y-axis acceleration and calf Z-axis acceleration;
the lower leg angular velocity comprises: the shank X-axis angular velocity, the shank Y-axis angular velocity and the shank Z-axis angular velocity;
the sole pressure comprises a first node pressure, a second node pressure and a third node pressure which are uniformly distributed on the sole.
3. The on-line detection method according to claim 1, wherein the constructing naive bayes classifier for off-line normal gait and off-line frozen gait respectively based on the off-line sample set specifically comprises:
counting the average value and the standard deviation of all the feature vectors of each window data;
calculating prior probabilities of the normal gait and the frozen gait in the offline sample set;
constructing a naive Bayes classifier of the normal gait according to the prior probability of the normal gait and the average value and the standard value of all the feature vectors of the window data marked as the normal gait;
and constructing a naive Bayes classifier of the frozen gait according to the prior probability of the frozen gait and the average value and the standard value of all the feature vectors of the window data marked as the frozen gait.
4. The on-line detection method of claim 3, wherein the naive Bayes classifier is represented by the following equation:
Figure FDA0002701358280000021
wherein, P (x)iI c) is the conditional probability that the ith feature vector belongs to the frozen gait or the normal gait, and P (c) isA prior probability of said normal gait or said frozen gait.
5. The method of claim 4, wherein P (x)i| c) is calculated by the following equation:
Figure FDA0002701358280000031
wherein x isiIs the ith feature vector, muc,iIs the mean value, σc,iIs the standard value.
6. The method according to claim 4, wherein the acquiring online gait data during online walking comprises:
and acquiring online gait data of the detection window according to the preset time length and the fixed time interval.
7. The method according to claim 6, wherein the obtaining of the probabilities of the normal gait and the frozen gait during the walking process respectively by the naive Bayes classifier and the online gait data comprises:
calculating a feature vector corresponding to the detection window;
calculating the probability that the detection window is in an online normal gait through a naive Bayes classifier of the normal gait according to the feature vector;
and calculating the probability that the detection window is in the online frozen state through a naive Bayes classifier of the frozen gait according to the feature vector.
8. The method according to claim 7, wherein the comparing the probabilities of the online normal gait and the online frozen gait to obtain the detection result specifically comprises:
judging whether the probability of the online normal pace is larger than that of the online normal pace;
if so, determining that the detection result is normal gait;
if not, determining that the detection result is the frozen gait.
CN201810386698.3A 2018-04-26 2018-04-26 Freezing gait online detection method Active CN108629304B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810386698.3A CN108629304B (en) 2018-04-26 2018-04-26 Freezing gait online detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810386698.3A CN108629304B (en) 2018-04-26 2018-04-26 Freezing gait online detection method

Publications (2)

Publication Number Publication Date
CN108629304A CN108629304A (en) 2018-10-09
CN108629304B true CN108629304B (en) 2020-12-08

Family

ID=63694729

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810386698.3A Active CN108629304B (en) 2018-04-26 2018-04-26 Freezing gait online detection method

Country Status (1)

Country Link
CN (1) CN108629304B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109480857B (en) * 2018-12-29 2021-09-14 中国科学院合肥物质科学研究院 Device and method for detecting frozen gait of Parkinson disease patient
CN110151190A (en) * 2019-05-23 2019-08-23 西南科技大学 A kind of orthopaedics postoperative rehabilitation monitoring method and system
CN115171886B (en) * 2022-07-25 2023-03-24 北京戴来科技有限公司 Frozen gait detection method, device, equipment and storage medium based on random forest algorithm
CN115188468B (en) * 2022-07-25 2023-04-25 北京戴来科技有限公司 Freezing gait detection method, device, equipment and storage medium based on support vector machine
CN117298449B (en) * 2023-10-31 2024-04-09 首都医科大学宣武医院 Closed-loop DBS regulation and control method and system based on wearable equipment

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012167328A1 (en) * 2011-06-10 2012-12-13 Bright Devices Group Pty Ltd Freezing of gait cue apparatus
CN103886341A (en) * 2014-03-19 2014-06-25 国家电网公司 Gait behavior recognition method based on feature combination
CN104834888A (en) * 2014-12-04 2015-08-12 龙岩学院 Abnormal gait identification method capable of facilitating screening Parkinsonism
CN105142714A (en) * 2013-01-21 2015-12-09 卡拉健康公司 Devices and methods for controlling tremor
CN107361773A (en) * 2016-11-18 2017-11-21 深圳市臻络科技有限公司 For detecting, alleviating the device of Parkinson's abnormal gait

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102578261B1 (en) * 2016-09-05 2023-09-13 삼성전자주식회사 Method for walking assist, and devices operating the same

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012167328A1 (en) * 2011-06-10 2012-12-13 Bright Devices Group Pty Ltd Freezing of gait cue apparatus
CN105142714A (en) * 2013-01-21 2015-12-09 卡拉健康公司 Devices and methods for controlling tremor
CN103886341A (en) * 2014-03-19 2014-06-25 国家电网公司 Gait behavior recognition method based on feature combination
CN104834888A (en) * 2014-12-04 2015-08-12 龙岩学院 Abnormal gait identification method capable of facilitating screening Parkinsonism
CN107361773A (en) * 2016-11-18 2017-11-21 深圳市臻络科技有限公司 For detecting, alleviating the device of Parkinson's abnormal gait

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Online Detection of Freezing of Gait with Smartphones and Machine Learning Techniques;Sinziana Mazilu etc;《2012 6th Interational Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth) and Workshops》;20120713;正文125-126、128页 *

Also Published As

Publication number Publication date
CN108629304A (en) 2018-10-09

Similar Documents

Publication Publication Date Title
CN108629304B (en) Freezing gait online detection method
EP3468450B1 (en) Method and system for analyzing human gait
Zhang et al. Beyond the standard clinical rating scales: fine-grained assessment of post-stroke motor functionality using wearable inertial sensors
JP6951516B2 (en) Methods and systems for detecting human pace
Otanasap Pre-impact fall detection based on wearable device using dynamic threshold model
Jalloul et al. Activity recognition using complex network analysis
KR20110125656A (en) System and method for detecting the walk of a person
CN112115827B (en) Falling behavior identification method based on human body posture dynamic characteristics
Whelan et al. Leveraging IMU data for accurate exercise performance classification and musculoskeletal injury risk screening
Xie et al. Human activity recognition method based on inertial sensor and barometer
Papageorgiou et al. Experimental validation of human pathological gait analysis for an assisted living intelligent robotic walker
Stone et al. Extracting footfalls from voxel data
Lueken et al. Estimation of stride time variability in unobtrusive long-term monitoring using inertial measurement sensors
Farashi Distinguishing between Parkinson’s disease patients and healthy individuals using a comprehensive set of time, frequency and time-frequency features extracted from vertical ground reaction force data
Jarchi et al. Validation of the e-AR sensor for gait event detection using the parotec foot insole with application to post-operative recovery monitoring
Saad et al. Sensoring and features extraction for the detection of Freeze of Gait in Parkinson disease
Lueken et al. Peak detection algorithm for gait segmentation in long-term monitoring for stride time estimation using inertial measurement sensors
WO2018092219A1 (en) Information processing system, information processing device, information processing method and information processing program
Pham et al. An anomaly detection technique in wearable wireless monitoring systems for studies of gait freezing in Parkinson's disease
Gradl et al. Movement speed estimation based on foot acceleration patterns
Islam et al. Neurodegenerative disease classification using gait signal features and random Forest classifier
Santhiranayagam et al. Estimation of end point foot clearance points from inertial sensor data
US11216074B2 (en) Motion classification user library
Saponaro et al. Estimating Physical Activity Intensity And Energy Expenditure Using Computer Vision On Videos
Heydarov et al. Performance comparison of ML methods applied to motion sensory information for identification of vestibular system disorders

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant