CN111768620A - Road anomaly detection method based on window division and deformation clustering - Google Patents

Road anomaly detection method based on window division and deformation clustering Download PDF

Info

Publication number
CN111768620A
CN111768620A CN202010552474.2A CN202010552474A CN111768620A CN 111768620 A CN111768620 A CN 111768620A CN 202010552474 A CN202010552474 A CN 202010552474A CN 111768620 A CN111768620 A CN 111768620A
Authority
CN
China
Prior art keywords
window
sliding window
road
deformation
segment
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.)
Pending
Application number
CN202010552474.2A
Other languages
Chinese (zh)
Inventor
陈垣毅
郑增威
周铭煊
霍梅梅
陈丹
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.)
Zhejiang University City College ZUCC
Original Assignee
Zhejiang University City College ZUCC
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 Zhejiang University City College ZUCC filed Critical Zhejiang University City College ZUCC
Priority to CN202010552474.2A priority Critical patent/CN111768620A/en
Publication of CN111768620A publication Critical patent/CN111768620A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Analytical Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention relates to a road anomaly detection method based on window division and deformation clustering, which comprises the following steps: step 1, carrying out threshold detection and sliding window processing on z-axis acceleration data by a moving end, and screening a segment to be determined; step 2, the mobile terminal judges whether the road segment is an abnormal segment or not through a random forest algorithm; and 3, the cloud end deforms the data of the road abnormal fragments to a uniform length through a deformation clustering method. The invention has the beneficial effects that: the method comprises the steps of determining a possible road abnormal window by utilizing threshold detection and a sliding window, determining whether the road abnormal window is a road abnormal window according to a random forest algorithm, determining the abnormal type of the road according to an algorithm of a deformation cluster and a support vector machine, and returning the abnormal type; the method can completely intercept the road abnormal paragraphs, and can more accurately detect the abnormal conditions of the road on different data sets; the invention is superior to the existing methods in three indexes of energy consumption, network delay and network consumption.

Description

Road anomaly detection method based on window division and deformation clustering
Technical Field
The invention relates to the field of road anomaly detection, in particular to a road anomaly detection method based on window division and deformation clustering; the edge calculation part is that possible abnormal windows are firstly extracted when a mobile terminal carries out road abnormal detection, the abnormal windows and normal windows are distinguished by a random forest method, and the abnormal windows are transmitted to a cloud end; the cloud computing part obtains the abnormal type of the window through a support vector machine after the abnormal road is deformed.
Background
In order to detect abnormal conditions of roads, early research work proposed various detection methods based on professional equipment or visual information. However, dedicated sensing devices such as 3D vision, depth sensors and ground penetrating radar are expensive in cost and inconvenient to deploy on a large scale on a common vehicle, so that detection of abnormal conditions on roads is time-consuming and needs to be performed specifically for a certain road, and the device is not universal.
In recent years, with the popularization of smart phones and the development of mobile communication technologies, a large number of data mining systems based on smart phones are valued and applied to real life. Many scholars think that the road damage condition can be detected more accurately by extracting and identifying characteristics by combining time domain and frequency domain information of acceleration. The method treats road damage detection as a classification problem, and extracts identification features based on acceleration data time domain or frequency domain information. In the aspect of feature extraction, the energy value of each frequency band of the acceleration information, the mean value, the root mean square, the standard deviation, the variance and the like of the vibration of the acceleration information are widely applied. In terms of detection models, various classification clustering models are used for road damage detection, such as support vector machines, k-means clustering, decision tree classifiers, gaussian mixture models, bayesian networks, and the like.
Although the scholars at home and abroad propose various acceleration-based road damage detection models, certain defects exist. Firstly, some current methods mostly process the whole data based on a sliding window, and the road anomalies are often unevenly distributed on time sequence data, and the lengths of the road anomalies are different, so that the possibility of dividing the data of the road anomaly section exists to a great extent; secondly, some existing methods rely on data sets and features, large differences often exist on different data sets, and a multi-classification method is difficult to apply to data sets of two classifications; finally, the existing method only considers the detection effect of the road abnormity, and does not consider the consumption of the method per se in a specific application scene.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides a road anomaly detection method based on window division and deformation clustering
The road anomaly detection method based on window division and deformation clustering comprises the following steps:
step 1, carrying out threshold detection and sliding window processing on z-axis acceleration data by a moving end, and screening a segment to be determined;
step 2, the mobile terminal judges whether the road segment is an abnormal segment through a random forest algorithm, and if the road segment is the abnormal segment, the segment is transmitted to the cloud: training a random forest model by taking some standard samples as training sets, storing parameters of the random forest model and transplanting the parameters to a mobile terminal, and inputting the fragments to be determined into the random forest model by the mobile terminal for judgment;
step 3, the cloud end deforms the data of the road abnormal fragments to a uniform length through a deformation clustering method:
step 4, comparing the distance between the cloud and the standard fragment to judge the fragment type: applying deformation classification to all training sets and data transmitted by the mobile terminal; calculating Euclidean distances between the sequences after deformation is obtained by calculation and all training set deformation sequences for the data of the uneven condition of the mobile terminal, and determining the type of the uneven condition according to a KNN method;
and 5, the cloud returns the fragment types to the mobile terminal.
Preferably, the step 1 specifically comprises the following steps:
step 1.1, recording a high threshold value as TH and recording a low threshold value as TL; processing the z-axis acceleration, and recording the intersection points of the acceleration data curve and the straight line corresponding to the TH as a and b; recording the intersection points of the acceleration data curve and the straight line corresponding to the TL as c and d; setting the starting point of the segment to be determined as a point a and the end point as a point d;
after the point a is detected, the coordinate of the point a is recorded as xaThe coordinate range of the detection window is recorded as [ x ]a,xa+Range]Range is the length of the detection window; marking the point a as a left starting point of the sliding window; since the points c and d are both the points intersected with the low threshold TL, the last point intersected with the low threshold in the detection window is marked as the right starting point of the sliding window;
step 1.2, when the vehicle passes through the road and is abnormal, the acceleration sensor on the intelligent device records the change of the acceleration, the points of the start and the end of the change are shown as the point e and the point f in fig. 2, and the positioning of the segment to be determined is more accurately carried out through the sliding window processing:
step 1.2.1, recording the coordinate range of the sliding window as [ x 'for the left starting point a of the sliding window'a,xa]，x′aIs the lower bound of the sliding window, will x'aIs set to x'a＝xa-Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z1；
Step 1.2.2, updating that the lower bound of the sliding window is x'a＝x′aWindow, updating the upper bound of the sliding Window to xa＝xa-Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z2；
Step 1.2.3, judge z1And z2The size relationship of (1): if z is1>z2Then, it is determined that the z-axis acceleration in the sliding window gradually decreases, and there is a process of gradually increasing the acceleration in the actual situation, so that z is gradually increased1＝z2And returning to execute the step 1.2.2 and the step 1.2.3 if z1≤z2Then the sliding window at this time is down bound by x'aRecording as a segment starting point to be determined;
step 1.2.4, recording the coordinate range of the sliding window as x 'for the starting point d on the right side'd,xd]，x′dIs the lower bound of the sliding window, will x'dIs set to x'd＝xd-Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z3；
Step 1.2.5, updating that the lower bound of the sliding window is x'd＝x′d+ Window, updating the upper bound of the sliding Window to xd＝xd+ Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z4；
Step 1.2.6, judge z3And z4The size relationship of (1): if z is3≥z4Then the sliding window at this time is down bound by x'dRecording as a segment end point to be determined; if z is3＜z4Then, the z-axis acceleration in the sliding window is determined to gradually decrease, so as to make z3＝z4And returning to execute the step 1.2.4 and the step 1.2.5.
Preferably, the step 3 specifically comprises the following steps:
step 3.1, calculating autocorrelation coefficient rk:
Figure BDA0002543070320000031
In the above formula, k is increased from 1 to m-l (the longest length of the time series after deformation), m is the length of the time series of the road abnormal segment,
Figure BDA0002543070320000036
the average value of the time sequence data of the road abnormal segment is T ═ T1,t2,…tm) Generating a time series after deformation for the deformation classification model;
and 3.2, calculating a partial autocorrelation function (PACF) according to the following formula:
Figure BDA0002543070320000032
in the above formula, Rp＝(r1,r2,…,rp) For the top p term of the autocorrelation function (PACF),
Figure BDA0002543070320000033
is a Toepliz matrix of an autocorrelation function (PACF),
Figure BDA0002543070320000034
calculated according to the following formula:
Figure BDA0002543070320000035
and 3.3, changing the time sequence into m-l items to obtain a matrix:
Figure BDA0002543070320000041
the sequence after deformation is obtained according to the following formula:
L＝(λ1,1,λ2,2,…λm-l,m-l)。
the invention has the beneficial effects that: the method comprises the steps of determining a possible road abnormal window by utilizing threshold detection and a sliding window, determining whether the road abnormal window is a road abnormal window according to a random forest algorithm, determining the abnormal type of the road according to an algorithm of a deformation cluster and a support vector machine, and returning the abnormal type; the method can completely intercept the road abnormal paragraphs, and can more accurately detect the abnormal conditions of the road on different data sets; the invention is superior to the existing methods in three indexes of energy consumption, network delay and network consumption.
Drawings
FIG. 1 is a flow chart for generating a road anomaly detection result;
FIG. 2 is a graph of threshold detection results for z-axis acceleration data;
FIG. 3 is a graph comparing results of road pothole detection based on a data set;
FIG. 4 is a comparison graph of deceleration strip detection results based on a data set;
FIG. 5 is a graph comparing the results of metal manhole cover inspection based on a data set;
FIG. 6 is a graph of simulated delay comparison based on a data set;
FIG. 7 is a graph of simulated energy consumption comparison based on a data set;
FIG. 8 is a graph comparing simulated network usage based on a data set.
Detailed Description
The present invention will be further described with reference to the following examples. The following examples are set forth merely to aid in the understanding of the invention. It should be noted that, for a person skilled in the art, several modifications can be made to the invention without departing from the principle of the invention, and these modifications and modifications also fall within the protection scope of the claims of the present invention.
First, the overall idea of the invention:
the road abnormity detection method mainly comprises the steps of carrying out threshold detection on a moving end according to z-axis acceleration generated by an acceleration sensor, then generating a segment to be determined through a sliding window, judging whether the segment is an abnormal segment or not through a random forest algorithm, and transmitting the abnormal segment to a cloud end. And deforming the fragments to a uniform length through a deformation clustering algorithm at the cloud end, calculating a distance with the standard fragments to determine the abnormal type of the fragments, and transmitting the abnormal type back to the mobile terminal.
Secondly, as an embodiment, as shown in fig. 1, the steps are as follows:
1. the moving end carries out threshold detection and sliding window processing on z-axis acceleration data, and a segment to be determined is screened, wherein the moving end comprises the following steps:
1) the high and low thresholds are recorded as TH and TL respectively. Processing the z-axis acceleration can find the point where the dotted line represented by the threshold intersects the solid line represented by the acceleration data, as shown in fig. 2, there are typically 4 points, denoted as a, b, c, d. Because the segment to be determined only needs one starting point and one ending point, namely the point a and the point d, the marked points are screened and reduced, and the steps are as follows:
① when a point is detected, its coordinate is recorded as xaThe coordinate range of the detection window is [ x ]a,xa+Range]. Where Range is the length of the detection window. Point a is recorded as the sliding window left starting point.
And secondly, since the points c and d are the points intersected with the low threshold TL, the last point intersected with the low threshold in the detection window is marked as the right starting point of the sliding window.
2) When the vehicle passes through a road abnormality, the acceleration sensor provided in the smart device records a change in acceleration, and the points at which the change starts and ends (i.e., actual points) are generally indicated as points e and f in fig. 2. At this time, the positioning of the segment to be determined needs to be performed more accurately by a sliding window method, which includes the following steps:
① for the starting point a on the left, let us note that the sliding window coordinate range is x'a,xa]Wherein x'aIs the lower boundary of the sliding window, and the initial value is set to x'a＝xaWindow, where Window is the length of the sliding Window. Calculating the average value of the z-axis acceleration in the sliding window at the moment, and recording as z1。
② updating the lower bound of the sliding window to be x'a＝x′aWindow, upper bound xa＝xa-Window. Calculating the average value of the z-axis acceleration in the sliding window at the moment, and recording as z2。
③ determination of z1And z2The magnitude relationship of (1). If z is1>z2The z-axis acceleration in the sliding window is in a gradual decrease, and in an actual situation, a gradual acceleration increase process exists, so that the z-axis acceleration is in a gradual decrease state1＝z2Then step ② is repeated if z1≤z2Then the sliding window at this time is down bound by x'aIs recorded as the starting point of the segment to be determined.
④, for the right starting point d, processing similar to steps ① - ③ is performed, and only the updating of the sliding Window is changed to be increased by one Window each time, and the judgment result of the size relationship is reversed, so that the segment end point x 'to be determined can be obtained'd。
2. The mobile terminal judges whether the road segment is an abnormal segment or not through a random forest algorithm, if so, the segment is transmitted to the cloud terminal, and the steps are as follows:
1) training a random forest model by using a plurality of standard samples as a training set in advance.
2) And storing the model parameters and transplanting the model parameters to a mobile terminal, and inputting the fragments to be determined into a random forest by the mobile terminal for judgment.
3. The cloud end deforms the data of the road abnormal fragments to a uniform length by a deformation clustering method, and the method comprises the following steps:
for time series T ═ T (T)1,t2,…tm) The deformation classification model generates a time sequence after deformation according to the following steps:
1) the autocorrelation coefficient r is calculated as followsk:
Figure BDA0002543070320000061
Where k increases from 1 to m-l (the longest length of the time series after deformation).
2) The partial autocorrelation function (PACF) is calculated as follows:
Figure BDA0002543070320000062
wherein R isp＝(r1,r2,…,rp) For the first p terms of the autocorrelation function (ACF),
Figure BDA0002543070320000063
the Toepliz matrix, which is an autocorrelation function, is calculated as follows:
Figure BDA0002543070320000064
3) let p be m-l, i.e. the time sequence morphs into the first m-l entries, resulting in a matrix:
Figure BDA0002543070320000065
the sequence after deformation is obtained according to the following formula:
L＝(λ1,1,λ2,2,…λm-l,m-l)
4. the method for determining the abnormal type of the segment to be determined comprises the following steps:
and applying deformation classification to all training sets and data transmitted by the mobile terminal. And calculating the data of the uneven condition of the mobile terminal to obtain a sequence after deformation, calculating Euclidean distances with all the training set deformation sequences, and determining the type of the uneven condition according to a KNN method. After the type is determined, the cloud end transmits the abnormal type of the road abnormal detection result back to the mobile terminal.
Thirdly, verifying the result:
to verify the effectiveness of the method, experiments were performed on a data set with the following information as shown in table 1 below:
TABLE 1 data set Table for experimental comparison
Road pit Speed bump Metal manhole cover
Data set 236 138 135
Comparative methods table 2 below shows the road abnormality detection methods mentioned in the recent papers. In addition, the experiment also adjusts the proportion of the training set and the test set, and the proportion of the training set to the total data set is increased from 50% to 90% in 10% intervals.
TABLE 2 method of experimental comparison
Figure BDA0002543070320000071
Some parameters of the experiment were set as follows: range is 50, Window is 2, and k is 1
The evaluation of the results of the experiment was performed by comparison with F1. F1 can be calculated from the following formula:
Figure BDA0002543070320000072
for a specific type of road anomaly, tp represents the correct prediction number of the road anomaly by a certain method, fn represents the undetected number of the road anomaly by a certain method, and fp represents the number of other road anomalies divided into the road anomaly by a certain method.
Fourthly, experimental conclusion:
FIGS. 3, 4 and 5 show the F1 detection indexes of road pits, deceleration strips and metal well covers on a data set by the methods of SVM-WW, CPS-SFS, SVM-MDDP and MLM-WB in the experiment respectively compared with the QF-COTE method provided herein. It can be seen that:
1) all methods have more than 0.4 effect of identifying all irregularities on the current data set. In the existing method, the best recognition effect on three out-of-flatness conditions is SVM-MDDP, and the detection effects of road pits, deceleration strips and metal well covers of 0.765, 0.761 and 0.752 are achieved when 90% of data are used for training.
2) Compared with the existing method, the performance of the proposed QF-COTE method on detection indexes is improved to a certain extent, and the detection effect on the road pits is always over 0.8 and is improved by at least 0.1. Compared with the best method in the prior art, the method has the advantages that the promotion proportion is the largest when 70% of data are used for training, and the promotion is 18%; meanwhile, the speed bump and the metal well lid are respectively lifted by 15% and 13%.
FIGS. 6, 7 and 8 show the comparison of the simulation indexes of the SVM-WW, CPS-SFS, SVM-MDDP and MLM-WB methods in the present experiment with the QF-COTE method proposed herein on the data set 1. It can be seen that:
1) the method without data preprocessing (CPD-SFS, MLM-WB) consumes a lot in simulation because all data must be completely transmitted to the cloud end through the mobile end, and the uneven condition in the actual detection application is not large. Meanwhile, the proposed QF-COTE method does not adopt edge calculation data processing, and various losses are not superior to those of other methods.
2) In the case of processing of edge calculation, the proposed method is at least 1/2 lower than the least consumption method in terms of energy consumption, network usage and time delay, and is 1/30 of CPD-SFS and 1/3 of SVM-MDDP.

Claims (3)

1. A road anomaly detection method based on window division and deformation clustering is characterized by comprising the following steps:
step 1, carrying out threshold detection and sliding window processing on z-axis acceleration data by a moving end, and screening a segment to be determined;
step 2, the mobile terminal judges whether the road segment is an abnormal segment through a random forest algorithm, and if the road segment is the abnormal segment, the segment is transmitted to the cloud: training a random forest model by taking some standard samples as training sets, storing parameters of the random forest model and transplanting the parameters to a mobile terminal, and inputting the fragments to be determined into the random forest model by the mobile terminal for judgment;
step 3, the cloud end deforms the data of the road abnormal fragments to a uniform length through a deformation clustering method:
step 4, comparing the distance between the cloud and the standard fragment to judge the fragment type: applying deformation classification to all training sets and data transmitted by the mobile terminal; calculating Euclidean distances between the sequences after deformation is obtained by calculation and all training set deformation sequences for the data of the uneven condition of the mobile terminal, and determining the type of the uneven condition according to a KNN method;
and 5, the cloud returns the fragment types to the mobile terminal.
2. The method for detecting road anomaly based on window division and deformation clustering according to claim 1, wherein the step 1 specifically comprises the following steps:
step 1.1, recording a high threshold value as TH and recording a low threshold value as TL; processing the z-axis acceleration, and recording the intersection points of the acceleration data curve and the straight line corresponding to the TH as a and b; recording the intersection points of the acceleration data curve and the straight line corresponding to the TL as c and d; setting the starting point of the segment to be determined as a point a and the end point as a point d;
after the point a is detected, the coordinate of the point a is recorded as xaThe coordinate range of the detection window is recorded as [ x ]a,xa+Range]Range is the length of the detection window; marking the point a as a left starting point of the sliding window; recording the last point intersected with the low threshold value in the detection window as the right starting point of the sliding window;
step 1.2, when the vehicle passes through the road and is abnormal, an acceleration sensor on the intelligent device records the change of acceleration, and the positioning of the segment to be determined is more accurately carried out through the sliding window processing:
step 1.2.1, recording the coordinate range of the sliding window as [ x 'for the left starting point a of the sliding window'a,xa]，x′aIs the lower bound of the sliding window, will x'aIs set to x'a＝xa-Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z1；
Step 1.2.2, updating that the lower bound of the sliding window is x'a＝x′aWindow, updating the upper bound of the sliding Window to xa＝xa-Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z2；
Step 1.2.3, judge z1And z2The size relationship of (1): if z is1>z2Then, the z-axis acceleration in the sliding window is determined to gradually decrease, so as to make z1＝z2Return receiptIn case z is taken from step 1.2.2 and step 1.2.31≤z2Then the sliding window at this time is down bound by x'aRecording as a segment starting point to be determined;
step 1.2.4, recording the coordinate range of the sliding window as x 'for the starting point d on the right side'd,xd]，x′dIs the lower bound of the sliding window, will x'dIs set to x'd＝xd-Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z3；
Step 1.2.5, updating that the lower bound of the sliding window is x'd＝x′d+ Window, updating the upper bound of the sliding Window to xd＝xd+ Window, where Window is the length of the sliding Window; calculating the average value of the z-axis acceleration in the sliding window at the moment and recording the average value as z4；
Step 1.2.6, judge z3And z4The size relationship of (1): if z is3≥z4Then the sliding window at this time is down bound by x'dRecording as a segment end point to be determined; if z is3＜z4Then, the z-axis acceleration in the sliding window is determined to gradually decrease, so as to make z3＝z4And returning to execute the step 1.2.4 and the step 1.2.5.
3. The method for detecting road anomaly based on window division and deformation clustering according to claim 1, wherein the step 3 specifically comprises the following steps:
step 3.1, calculating autocorrelation coefficient rk:
Figure FDA0002543070310000021
In the above formula, k is increased from 1 to m-l, m is the time series length of the road abnormal segment,
Figure FDA0002543070310000022
is the average value of the time sequence data of the road abnormal segment, and the time sequence T ═t1,t2,…tm) Generating a time series after deformation for the deformation classification model;
step 3.2, calculating a partial autocorrelation function according to the following formula:
Figure FDA0002543070310000023
in the above formula, Rp＝(r1,r2,…,rp) For the first p terms of the autocorrelation function,
Figure FDA0002543070310000024
is a Toepliz matrix of the autocorrelation function,
Figure FDA0002543070310000025
calculated according to the following formula:
Figure FDA0002543070310000026
and 3.3, changing the time sequence into m-l items to obtain a matrix:
Figure FDA0002543070310000027
the sequence after deformation is obtained according to the following formula:
L＝(λ1,1,λ2,2,…λm-l,m-l)。
CN202010552474.2A 2020-06-17 2020-06-17 Road anomaly detection method based on window division and deformation clustering Pending CN111768620A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010552474.2A CN111768620A (en) 2020-06-17 2020-06-17 Road anomaly detection method based on window division and deformation clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010552474.2A CN111768620A (en) 2020-06-17 2020-06-17 Road anomaly detection method based on window division and deformation clustering

Publications (1)

Publication Number Publication Date
CN111768620A true CN111768620A (en) 2020-10-13

Family

ID=72722655

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010552474.2A Pending CN111768620A (en) 2020-06-17 2020-06-17 Road anomaly detection method based on window division and deformation clustering

Country Status (1)

Country Link
CN (1) CN111768620A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112971784A (en) * 2021-05-20 2021-06-18 浙大城市学院 Wearable bone conduction fatigue driving detection method and device
CN113660237A (en) * 2021-08-10 2021-11-16 和中通信科技有限公司 Industrial Internet data flow abnormity detection method based on dynamic sliding window, memory and processor
CN114582135A (en) * 2021-11-17 2022-06-03 北京交通发展研究院 Method, device, terminal and storage medium for detecting road surface abnormity
CN115218860A (en) * 2022-09-20 2022-10-21 四川高速公路建设开发集团有限公司 Road deformation prediction method based on Mems acceleration sensor
CN115410342A (en) * 2022-08-26 2022-11-29 安徽省地质矿产勘查局332地质队 Landslide disaster intelligent early warning method based on crack meter real-time monitoring

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104751630A (en) * 2014-12-31 2015-07-01 浙江工业大学 Road traffic state acquisition method based on Kernel-KNN matching
CN109082984A (en) * 2018-06-29 2018-12-25 浙江大学城市学院 A kind of road abnormality detection model based on window division and dynamic time warping
CN110020686A (en) * 2019-04-10 2019-07-16 浙江大学城市学院 A kind of road surface method for detecting abnormality based on intelligent perception sensing data

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104751630A (en) * 2014-12-31 2015-07-01 浙江工业大学 Road traffic state acquisition method based on Kernel-KNN matching
CN109082984A (en) * 2018-06-29 2018-12-25 浙江大学城市学院 A kind of road abnormality detection model based on window division and dynamic time warping
CN110020686A (en) * 2019-04-10 2019-07-16 浙江大学城市学院 A kind of road surface method for detecting abnormality based on intelligent perception sensing data

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
ANTHONY BAGNALL; JASON LINES; JON HILLS; AARON BOSTROM: "《Time-Series Classification with COTE: The Collective of Transformation-Based Ensembles》", 《IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING》 *
郑加柱等: "变形监测数据的时间序列分析", 《森林工程》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112971784A (en) * 2021-05-20 2021-06-18 浙大城市学院 Wearable bone conduction fatigue driving detection method and device
CN113660237A (en) * 2021-08-10 2021-11-16 和中通信科技有限公司 Industrial Internet data flow abnormity detection method based on dynamic sliding window, memory and processor
CN114582135A (en) * 2021-11-17 2022-06-03 北京交通发展研究院 Method, device, terminal and storage medium for detecting road surface abnormity
CN115410342A (en) * 2022-08-26 2022-11-29 安徽省地质矿产勘查局332地质队 Landslide disaster intelligent early warning method based on crack meter real-time monitoring
CN115410342B (en) * 2022-08-26 2023-08-11 安徽省地质矿产勘查局332地质队 Landslide hazard intelligent early warning method based on real-time monitoring of crack meter
CN115218860A (en) * 2022-09-20 2022-10-21 四川高速公路建设开发集团有限公司 Road deformation prediction method based on Mems acceleration sensor
CN115218860B (en) * 2022-09-20 2022-12-09 四川高速公路建设开发集团有限公司 Road deformation prediction method based on Mems acceleration sensor

Similar Documents

Publication Publication Date Title
CN111768620A (en) Road anomaly detection method based on window division and deformation clustering
CN103473540B (en) The modeling of intelligent transportation system track of vehicle increment type and online method for detecting abnormality
CN103488941A (en) Hardware Trojan horse detection method and hardware Trojan horse detection system
CN108596221B (en) Image recognition method and device for scale reading
CN105426840A (en) Multi-feature fusion based infrared forest fire judgment method
CN114998852A (en) Intelligent detection method for road pavement diseases based on deep learning
CN102509291A (en) Pavement disease detecting and recognizing method based on wireless online video sensor
CN112732748B (en) Non-invasive household appliance load identification method based on self-adaptive feature selection
CN105023027A (en) Sole trace pattern image retrieval method based on multi-feedback mechanism
CA2578042A1 (en) Candidate generation for lung nodule detection
CN105005565A (en) Onsite sole trace pattern image retrieval method
CN103927514A (en) Coal and rock recognition method based on random local image features
CN109082984A (en) A kind of road abnormality detection model based on window division and dynamic time warping
WO2019192252A1 (en) Method and system for monitoring personal computer usage behaviour based on acoustic channel
CN112036450B (en) High-voltage cable partial discharge mode identification method and system based on transfer learning
CN101251896A (en) Object detecting system and method based on multiple classifiers
Bull et al. The effect of environmental and operational variabilities on damage detection in wind turbine blades
CN103413131A (en) Tower crane recognition method based on spectral and geometric characteristics
CN104964736B (en) Optical fiber invasion vibration source identification method based on time-frequency characteristic maximum expected classification
CN107153881B (en) Drilling risk prediction method
CN103942576A (en) Method for identifying coal and rock through airspace multiscale random characteristics
CN114092814A (en) Unmanned plane navel orange tree image target identification and statistics method based on deep learning
CN107422373B (en) A kind of coal mine rock burst disaster on-line early warning method of feature based drift
CN105512191A (en) Industry characteristics analyzer with artificial behavior learning capability
CN116977967A (en) Fake point cloud object identification method for automatic driving system

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20201013