CN113537323A - Indoor track error evaluation method based on LSTM neural network - Google Patents
Indoor track error evaluation method based on LSTM neural network Download PDFInfo
- Publication number
- CN113537323A CN113537323A CN202110752788.1A CN202110752788A CN113537323A CN 113537323 A CN113537323 A CN 113537323A CN 202110752788 A CN202110752788 A CN 202110752788A CN 113537323 A CN113537323 A CN 113537323A
- Authority
- CN
- China
- Prior art keywords
- track
- training
- step number
- reconstructed
- sequence number
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 28
- 238000011156 evaluation Methods 0.000 title claims abstract description 24
- 239000013598 vector Substances 0.000 claims abstract description 135
- 238000003062 neural network model Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims description 257
- 238000000034 method Methods 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 6
- 238000010801 machine learning Methods 0.000 abstract description 4
- 238000013507 mapping Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000005021 gait Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses an indoor track error evaluation method based on an LSTM neural network, which comprises the following steps: acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number; constructing a multi-dimensional feature vector of the reconstructed track point, obtaining a multi-dimensional feature vector set according to the multi-dimensional feature vector of the reconstructed track point, and abstracting the indoor track into a sequence; and for each target track step number sequence number, inputting a multi-dimensional feature vector set corresponding to the target track step number sequence number into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence is used for representing track step number sequence numbers larger than or equal to a preset value K, automatically learning and evaluating the mapping relation between the feature vector of each track step number sequence number of the indoor track and the real track through a machine learning model, and obtaining indoor track error evaluation results with finer granularity and higher precision.
Description
Technical Field
The invention relates to the technical field of geographic information, in particular to an indoor track error evaluation method based on an LSTM neural network.
Background
The evaluation of the error of the indoor track is an important research problem in the fields of track analysis, human movement pattern research and the like.
In terms of technical means, a common track error evaluation method is to establish an oval uncertain region on the basis of two reference points; the uncertain region established by the method is difficult to draw a complex shape between reference points, and the size of the elliptical region is also difficult to determine, so that the method has a limited effect in practical application.
Thus, there is still a need for improvement and development of the prior art.
Disclosure of Invention
The invention aims to solve the technical problem that aiming at the defects in the prior art, the invention provides an indoor track error evaluation method based on an LSTM neural network, aiming at solving the problem that in the track error evaluation method in the prior art, an oval uncertain area is mostly established on the basis of two reference points; the uncertain area established by the method is difficult to draw a complex shape between reference points, and the size of the elliptical area is also difficult to determine, so that the problem of limited action in practical application is caused.
The technical scheme adopted by the invention for solving the problems is as follows:
in a first aspect, an embodiment of the present invention provides an indoor trajectory error estimation method based on an LSTM neural network, where the method includes:
acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number;
constructing a multi-dimensional feature vector of the reconstructed track point, and obtaining a multi-dimensional feature vector set of the reconstructed track point according to the multi-dimensional feature vector of the reconstructed track point;
and for each target track step number sequence number, inputting the multi-dimensional feature vector set corresponding to the target track step number sequence number into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence number is used for representing the track step number sequence number which is greater than or equal to a preset value K.
In one implementation, the constructing the multi-dimensional feature vector of the reconstructed trajectory point includes:
obtaining the motion step length of the reconstructed track point, the motion direction of the reconstructed track point and the total track step number;
for each track step number, dividing the track step number by the total track step number to obtain a track step quotient corresponding to each track step number;
and for each track step number sequence number, the motion step length of the reconstructed track point, the motion direction of the reconstructed track point and the track step quotient form a multi-dimensional feature vector of the reconstructed track point.
In an implementation manner, the obtaining, according to the multidimensional feature vector of the reconstructed trajectory point, a multidimensional feature vector set of the reconstructed trajectory point includes:
and for each target track step number sequence number, combining the multidimensional characteristic vector corresponding to the first K-1 track step number sequence numbers adjacent to the target track step number sequence number and the multidimensional characteristic vector corresponding to the target track step number sequence number in sequence to obtain the multidimensional characteristic vector set of the reconstructed track point.
In one implementation, the training process of the LSTM neural network model specifically includes:
acquiring a training data set; wherein the training data set comprises a training multi-dimensional feature vector set and a training deviation value;
inputting the training multi-dimensional feature vector set to a preset initial network model to obtain a prediction deviation value;
obtaining a loss function according to the training deviation value and the prediction deviation value;
and training the initial network model according to the loss function to obtain an LSTM neural network model.
In one implementation, the obtaining the training data set includes:
acquiring a training track step number sequence number, training reconstruction track points corresponding to the training track step number sequence number and training real track points corresponding to the training track step number sequence number;
obtaining a training deviation value according to the training reconstructed track point and the training real track point;
obtaining training multidimensional characteristic vectors of the training reconstructed trajectory points according to the training reconstructed trajectory points;
and obtaining a training multi-dimensional feature vector set according to the training multi-dimensional feature vector.
In an implementation manner, the obtaining a training deviation value according to the training reconstructed trajectory point and the training real trajectory point includes:
for each training track step number sequence number, subtracting the abscissa of the training real track point from the abscissa of the training reconstructed track point to obtain a track abscissa difference value;
for each training track step number sequence number, subtracting the ordinate of the training real track point from the ordinate of the training reconstructed track point to obtain a track ordinate difference value;
summing the square of the track horizontal coordinate difference value and the square of the track vertical coordinate difference value to obtain a sum value;
and squaring the sum to obtain a training deviation value.
In an implementation manner, the obtaining a training multidimensional feature vector of the training reconstructed trajectory point according to the training reconstructed trajectory point includes:
acquiring a training motion step length of the training reconstructed track point, a motion direction of the training reconstructed track point and a total step number of a training track;
for each training track step number sequence number, dividing the training track step number sequence number by the total training track step number to obtain a training track step number quotient corresponding to each training track step number sequence number;
and for each training track step number sequence number, the training multi-dimensional feature vector of the training reconstructed track point is formed by the motion step length of the training reconstructed track point, the motion direction of the training reconstructed track point and the training track step number quotient.
In one implementation, the obtaining a training multi-dimensional feature vector set according to the training multi-dimensional feature vector includes:
taking the training track step number sequence number which is greater than or equal to the preset value K as a target training track step number sequence number;
and for each target training track step number sequence number, combining training multidimensional feature vectors corresponding to the first K-1 training track step number sequences adjacent to the target training track step number sequence number with training multidimensional feature vectors corresponding to the target training track step number sequence number in sequence to obtain a training multidimensional feature vector set of the target training track step number sequence number.
In a second aspect, an embodiment of the present invention further provides an indoor trajectory error estimation apparatus based on an LSTM neural network, where the apparatus includes:
the reconstruction track point acquisition module is used for acquiring track step number serial numbers and reconstruction track points corresponding to the track step number serial numbers;
the multi-dimensional feature vector set acquisition module is used for constructing the multi-dimensional feature vectors of the reconstructed track points and acquiring the multi-dimensional feature vector set of the reconstructed track points according to the multi-dimensional feature vectors of the reconstructed track points;
and the indoor track deviation value acquisition module is used for inputting the multi-dimensional feature vector set corresponding to the target track step number sequence number into the trained LSTM neural network model for each target track step number sequence number to obtain the indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence number is used for representing the track step number sequence number which is greater than or equal to the preset value K.
In a third aspect, an embodiment of the present invention further provides an intelligent terminal, including a memory, and one or more programs, where the one or more programs are stored in the memory, and configured to be executed by one or more processors includes a processor configured to execute the LSTM neural network-based indoor trajectory error estimation method described in any one of the above.
In a fourth aspect, embodiments of the present invention also provide a non-transitory computer-readable storage medium, wherein instructions of the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the LSTM neural network-based indoor trajectory error assessment method as described in any one of the above.
The invention has the beneficial effects that: firstly, acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number; then, constructing a multi-dimensional feature vector of the reconstructed track point, obtaining a multi-dimensional feature vector set of the reconstructed track point according to the multi-dimensional feature vector of the reconstructed track point, and abstracting the indoor track into a sequence; finally, for each target track step number sequence number, inputting a multi-dimensional feature vector set corresponding to the target track step number sequence number into a trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence is used for representing track step number sequences larger than or equal to a preset value K, and automatically learning and evaluating the mapping relation between the feature vector of each track step number sequence number of the indoor track and the real track through a machine learning model; and an indoor track error evaluation result with finer granularity and higher precision can be obtained.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of an indoor trajectory error evaluation method based on an LSTM neural network according to an embodiment of the present invention.
Fig. 2 is a schematic flowchart of an implementation manner provided in the embodiment of the present invention.
Fig. 3 is a diagram of an LSTM neural network architecture according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of experimental results obtained by using real indoor reconstructed trajectory data according to an embodiment of the present invention.
Fig. 5 is a schematic block diagram of an indoor trajectory error evaluation apparatus based on an LSTM neural network according to an embodiment of the present invention.
Fig. 6 is a schematic block diagram of an internal structure of an intelligent terminal according to an embodiment of the present invention.
Detailed Description
The invention discloses an indoor track error evaluation method based on an LSTM neural network, and in order to make the purpose, technical scheme and effect of the invention clearer and clearer, the invention is further described in detail below by referring to the attached drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
In the prior art, a common track error evaluation method is mostly used for establishing an elliptic uncertain region on the basis of two reference points. The uncertain region established by the method is difficult to draw a complex shape between reference points, and the size of the elliptical region is also difficult to determine, so that the method has a limited effect in practical application.
In order to solve the problems in the prior art, the embodiment provides an indoor track error evaluation method based on an LSTM neural network, and a mapping relation between a step number feature vector of each track step of an indoor track and a real track is automatically learned and evaluated through a machine learning model, so that an indoor track error evaluation result with finer granularity and higher precision can be obtained. In specific implementation, firstly, a track step number and a reconstruction track point corresponding to the track step number are obtained; then, constructing a multi-dimensional feature vector of the reconstructed track point, obtaining a multi-dimensional feature vector set of the reconstructed track point according to the multi-dimensional feature vector of the reconstructed track point, and abstracting the indoor track into a sequence; and finally, inputting the multi-dimensional feature vector set corresponding to the step number of the target track into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to the step number of the target track, wherein the step number of the target track is used for representing the step number of the track which is greater than or equal to a preset value K.
Exemplary method
The embodiment provides an indoor track error evaluation method based on an LSTM neural network, and the method can be applied to intelligent terminals in the field of geographic information. As shown in fig. 1 in detail, the method includes:
s100, acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number;
specifically, as shown in fig. 2, in the indoor trajectory error evaluation, the gait of the trajectory is detected, and the gait detection includes a trajectory step number, a reconstructed trajectory and a real trajectory. The reconstructed track is a motion track constructed according to position information fed back by the object at intervals, for example, the vehicle uploads the position of the vehicle to the server through a GPS at intervals, so that GPS data of the vehicle is stored in the server, and the motion track of the vehicle within a certain time is constructed by using the data through a track reconstruction function, and the motion track is the reconstructed track. The real trajectory is a total path that the object actually moves through, for example, a total path that the vehicle actually passes through while traveling. In practice, each step corresponds to a track step number, and each track step number corresponds to a reconstructed track point and a real track point. In order to find the corresponding relationship between the reconstructed track point and the real track point subsequently, the track step number sequence number and the reconstructed track point corresponding to the track step number sequence number need to be obtained first.
After the reconstructed trace points are obtained, the following steps can be performed as shown in fig. 1: s200, constructing a multi-dimensional feature vector of the reconstructed track point, and obtaining a multi-dimensional feature vector set of the reconstructed track point according to the multi-dimensional feature vector of the reconstructed track point;
specifically, a multidimensional feature vector can be constructed according to the relevant parameters of the reconstructed trajectory point, and the multidimensional feature vector is used for reflecting the characteristics of multiple dimensions of the reconstructed trajectory point. Because during error evaluation, the error of current track point can not directly be obtained through the error of current track point, need obtain through the accumulative error of several preceding steps of current track point, so need according to the multidimension feature vector of rebuilding track point obtains the multidimension feature vector set of rebuilding track point to obtain the accumulative error of several preceding steps of current track point, finally obtain the accurate error of current track point.
In an implementation manner of the present invention, the constructing the multidimensional feature vector of the reconstructed trajectory point includes the following steps:
s201, obtaining the motion step length of the reconstructed track point, the motion direction of the reconstructed track point and the total track step number;
s202, dividing the track step number by the total track step number to obtain a track step quotient corresponding to each track step number;
and S203, for each track step number sequence, forming the multi-dimensional characteristic vector of the reconstructed track point by the motion step length of the reconstructed track point, the motion direction of the reconstructed track point and the track step number quotient.
Specifically, the motion step length of the reconstructed track point and the motion direction of the reconstructed track point are the step length and the direction of motion from the positioning moment of the corresponding reconstructed point to the adjacent positioning moment, and for each step RT.p of the reconstructed track RTiStride is the reconstruction trajectory at RT.piCorresponding step length is processed, and the header is the reconstructed track at the RT.piCorresponding course is processed, and for each track step number, the track step number is divided by the total track step number to obtain a track step quotient value corresponding to each track step number; for example, i is the sequence number of the current step, i.e. the step number; n is the total number of steps of this track, and i/n is every the track step number quotient value that track step number corresponds is to every track step number will rebuild the motion step length of track point rebuild the motion direction of track point with track step number quotient value constitutes rebuild the multi-dimensional eigenvector of track point. For example, a multidimensional vector corresponding to each track step number is constructed through parameters corresponding to each track step number:
in an implementation manner of the present invention, obtaining the multi-dimensional feature vector set of the reconstructed trajectory points according to the multi-dimensional feature vectors of the reconstructed trajectory points includes the following steps:
and S204, for each target track step number sequence number, sequentially combining the multidimensional characteristic vector corresponding to the previous K-1 track step number sequence numbers adjacent to the target track step number sequence number and the multidimensional characteristic vector corresponding to the target track step number sequence number to obtain the multidimensional characteristic vector set of the reconstruction track point.
Specifically, in order to obtain the error of the current reconstructed trajectory point, the accumulated error of the K-1 step adjacent to and before the current step needs to be considered, so in this embodiment, the error estimation cannot be performed on the reconstructed trajectory points of the 0 th to the K-1 th steps. That is, the sequence numbers of the target track steps are the sequence numbers of all track steps starting from the Kth step. In actual processing, for example, when a first target track step number sequence number is processed, the multidimensional feature vector corresponding to the target track step number sequence number and the multidimensional feature vector corresponding to K-1 track step number sequences located before the first target track step number sequence number and adjacent to the target track step number sequence number are combined in sequence to form the multidimensional feature vector set of the reconstructed track point, for example, the track step number sequence number is 1,2,3,4,5,6,7,8,9, when K is 5, only the error of the 5 th reconstructed track point starting (the track step number sequence number is 5) can be evaluated, the error of the 1 st to 4 th reconstructed track points cannot be evaluated, the multidimensional feature vectors of the 1 st to 5 th reconstructed track points can be obtained by the foregoing method, the multidimensional feature vectors of the 1 st to 5 th reconstructed track points are combined according to the sequence relationship of the track step numbers, a multi-dimensional feature vector set of the 5 th reconstructed trajectory point can be obtained, and by analogy, a multi-dimensional feature vector set of the 6 th, 7 th, 8 th, 9 th, and 10 th.
After obtaining the multi-dimensional feature vector set of the reconstructed trajectory points, the following steps as shown in fig. 1 may be performed: s300, inputting the multi-dimensional feature vector set corresponding to the step number of each target track into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to the step number of each target track, wherein the step number of each target track is used for representing the step number of the track larger than or equal to a preset value K.
Specifically, since each reconstructed trajectory point has an error corresponding to the reconstructed trajectory point, that is, an indoor trajectory deviation value, from the kth reconstructed trajectory point, an indoor trajectory deviation value corresponding to each target trajectory step number can be obtained through the trained LSTM neural network model, and an input item of the trained LSTM neural network model is a multi-dimensional feature vector set corresponding to the target trajectory step number.
In an implementation manner of the present invention, the acquiring the training data set includes the following steps: acquiring a training track step number sequence number, training reconstruction track points corresponding to the training track step number sequence number and training real track points corresponding to the training track step number sequence number; obtaining a training deviation value according to the training reconstructed track point and the training real track point; obtaining training multidimensional characteristic vectors of the training reconstructed trajectory points according to the training reconstructed trajectory points; and obtaining a training multi-dimensional feature vector set according to the training multi-dimensional feature vector.
Specifically, a training track step number sequence number, training reconstruction track points corresponding to the training track step number sequence number and training real track points corresponding to the training track step number sequence number are obtained; for example, for one training reconstruction trajectory RT ═ RT1,RT.p2,...,RT.pn]And its corresponding training true trajectory GT ═ GT1,GT.p2,...,GT.pn]I is a training track step number sequence number, and then a training deviation value is obtained according to the training reconstruction track point and the training real track point; correspondingly, the step of obtaining the training deviation value according to the training reconstructed track point and the training real track point comprises the following steps: for each training track step number sequence number, subtracting the abscissa of the training real track point from the abscissa of the training reconstructed track point to obtain a track abscissa difference value; for each training track step number sequence number, subtracting the ordinate of the training real track point from the ordinate of the training reconstructed track point to obtain a track ordinate difference value; summing the square of the track horizontal coordinate difference value and the square of the track vertical coordinate difference value to obtain a sum value; and squaring the sum to obtain a training deviation value. For example, the calculation formula for calculating the training deviation value of the step number of each training track is as follows:
then according to whatThe training reconstructed trajectory points are obtained, and training multidimensional feature vectors of the training reconstructed trajectory points are obtained; correspondingly, obtaining the training multidimensional characteristic vector of the training reconstructed trajectory point according to the training reconstructed trajectory point comprises the following steps: acquiring a training motion step length of the training reconstructed track point, a motion direction of the training reconstructed track point and a total step number of a training track; for each training track step number sequence number, dividing the training track step number sequence number by the total training track step number to obtain a training track step number quotient corresponding to each training track step number sequence number; and for each training track step number sequence number, the training multi-dimensional feature vector of the training reconstructed track point is formed by the motion step length of the training reconstructed track point, the motion direction of the training reconstructed track point and the training track step number quotient. For example, the number of steps per training trajectory RT.p 'for the training reconstructed trajectory RT'iStride 'is the training reconstruction trajectory at RT.p'iCorresponding motion step length is processed, and the leading 'is the training reconstruction track at RT.p'iCorresponding movement direction, i' is the sequence number of the current training track step number, namely the step number; n is the total step number of the training track of the reconstruction track,
for training the trajectory step quotient, the following training multidimensional feature vectors are constructed:
and after the training multi-dimensional feature vector is obtained, obtaining a training multi-dimensional feature vector set according to the training multi-dimensional feature vector. Correspondingly, the step of obtaining a training multi-dimensional feature vector set according to the training multi-dimensional feature vector comprises the following steps: taking the training track step number sequence number which is greater than or equal to the preset value K as a target training track step number sequence number; for each step number sequence number of the target training track, combining training multidimensional feature vectors corresponding to the step number sequences of the first K-1 training tracks adjacent to the step number sequence number of the target training track with training multidimensional feature vectors corresponding to the step number sequences of the target training track in sequence to obtain a target training trackAnd training multi-dimensional feature vector set marking the step number sequence number of the training track.
Specifically, in the training process, errors cannot be obtained for the first K-1 training reconstructed trajectory points, calculation can only be started from the kth training reconstructed trajectory point, and for convenience of calculation, the sequence number of the training trajectory steps greater than or equal to the preset value K is used as the sequence number of the target training trajectory steps; when the Kth training reconstruction track point is calculated, training multidimensional feature vectors corresponding to the Kth training reconstruction track point can be obtained through the method, meanwhile, training multidimensional feature vectors corresponding to the first K-1 training reconstruction track points are respectively obtained through the method, and the training multidimensional feature vectors corresponding to the Kth training reconstruction track point and the training multidimensional feature vectors corresponding to the first K-1 training reconstruction track points of the Kth training reconstruction track point are combined according to the front-back relation of the step number of the target training track to obtain a training multidimensional feature vector set of the step number of the Kth target training reconstruction track. After the training multi-dimensional feature vector corresponding to the K +1 th training reconstruction track point is calculated, the training multi-dimensional feature vector corresponding to the K +1 th training reconstruction track point and the training multi-dimensional feature vector corresponding to the K-1 th training reconstruction track point can be combined according to the front-back relation of the step number sequence of the target training track, so that the training multi-dimensional feature vector set of the step number sequence of the K +1 th target training track is obtained, and the like.
In an implementation manner of the present invention, the training process of the LSTM neural network model specifically includes the following steps: acquiring a training data set; wherein the training data set comprises a training multi-dimensional feature vector set and a training deviation value; inputting the training multi-dimensional feature vector set to a preset initial network model to obtain a prediction deviation value; obtaining a loss function according to the training deviation value and the prediction deviation value; and training the initial network model according to the loss function to obtain an LSTM neural network model.
Inputting a training multi-dimensional feature vector set in a training data set to a preset initial network model to obtain a prediction deviation value corresponding to the training multi-dimensional feature vector set, wherein the training data set comprises the training multi-dimensional feature vector set and a training deviation value; in this embodiment, as shown in fig. 3, the initial network model is composed of two parts, the first part is an LSTM structure, and k (for example, k is 10) LSTM units are connected in series; the second part is a fully-connected multi-layer neural network which accepts as input the output of the last LSTM unit of the previous layer. Obtaining a loss function according to the training deviation value and the prediction deviation value; the loss function can be a cross entropy loss function or a mean square error value, after the loss function is obtained through calculation, whether the loss function is smaller than or equal to a preset threshold value or not is judged, and if the loss function is smaller than or equal to the preset threshold value, the training is ended; if the loss function is larger than a preset threshold value, judging whether the training times of the initial network model reach a preset time threshold value, if the training times of the initial network model do not reach the preset time threshold value, correcting the network parameters of the initial network model according to the loss function, and if the training times of the initial network model reach the preset time threshold value, ending the training to obtain the LSTM neural network model. The preset threshold and the preset time threshold may be determined according to the model accuracy for obtaining the LSTM neural network model, which is not described in detail herein. The preset number threshold may be a maximum training number of the initial network model, for example, 5000 times. Therefore, whether the initial network model training is finished or not is judged through the loss function and the training times, and the phenomenon that the initial network model training enters a dead cycle due to the fact that the loss function cannot reach the condition smaller than or equal to the preset threshold value can be avoided.
In order to illustrate the effect of the indoor trajectory error evaluation method provided by the embodiment of the invention, the invention uses real data to perform experiments. Fig. 4 is an experimental result obtained by using real pieces of indoor reconstructed trajectory data according to the present invention. The error of the reconstructed track and the real track is compared with the error predicted by the method, and the visual result shows that the error predicted by the method is very close to the actual error. The effectiveness of the method in predicting the indoor track error is proved.
Exemplary device
As shown in fig. 5, an indoor trajectory error evaluation apparatus based on an LSTM neural network includes a reconstructed trajectory point obtaining module 401, a multidimensional feature vector set obtaining module 402, and an indoor trajectory deviation value obtaining module 403, where:
a reconstructed track point obtaining module 401, configured to obtain a track step number sequence number and a reconstructed track point corresponding to the track step number sequence number;
a multidimensional feature vector set obtaining module 402, configured to construct a multidimensional feature vector of the reconstructed trajectory point, and obtain a multidimensional feature vector set of the reconstructed trajectory point according to the multidimensional feature vector of the reconstructed trajectory point;
and an indoor track deviation value obtaining module 403, configured to, for each target track step number, input the multidimensional feature vector set corresponding to the target track step number into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number, where the target track step number is used to represent a track step number greater than or equal to a preset value K.
Based on the above embodiment, the present invention further provides an intelligent terminal, and a schematic block diagram thereof may be as shown in fig. 6. The intelligent terminal comprises a processor, a memory, a network interface, a display screen and a temperature sensor which are connected through a system bus. Wherein, the processor of the intelligent terminal is used for providing calculation and control capability. The memory of the intelligent terminal comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the intelligent terminal is used for being connected and communicated with an external terminal through a network. The computer program is executed by a processor to implement an indoor trajectory error estimation method based on an LSTM neural network. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the intelligent terminal is arranged inside the intelligent terminal in advance and used for detecting the operating temperature of internal equipment.
Those skilled in the art will appreciate that the schematic diagram of fig. 6 is merely a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation of the intelligent terminal to which the solution of the present invention is applied, and a specific intelligent terminal may include more or less components than those shown in the figure, or combine some components, or have different arrangements of components.
In one embodiment, an intelligent terminal is provided that includes a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for:
acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number;
constructing a multi-dimensional feature vector of the reconstructed track point, and obtaining a multi-dimensional feature vector set of the reconstructed track point according to the multi-dimensional feature vector of the reconstructed track point;
and for each target track step number sequence number, inputting the multi-dimensional feature vector set corresponding to the target track step number sequence number into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence number is used for representing the track step number sequence number which is greater than or equal to a preset value K.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
In summary, the present invention discloses an indoor trajectory error assessment method, an intelligent terminal, and a storage medium based on an LSTM neural network, wherein the method includes: acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number; constructing a multi-dimensional feature vector of the reconstructed track point, obtaining a multi-dimensional feature vector set according to the multi-dimensional feature vector of the reconstructed track point, and abstracting the indoor track into a sequence; and for each target track step number sequence number, inputting a multi-dimensional feature vector set corresponding to the target track step number sequence number into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence is used for representing track step number sequence numbers larger than or equal to a preset value K, automatically learning and evaluating the mapping relation between the feature vector of each track step number sequence number of the indoor track and the real track through a machine learning model, and obtaining indoor track error evaluation results with finer granularity and higher precision.
Based on the above embodiments, the present invention discloses an indoor trajectory error estimation method based on LSTM neural network, it should be understood that the application of the present invention is not limited to the above examples, and it will be apparent to those skilled in the art that modifications and variations can be made in the light of the above description, and all such modifications and variations are within the scope of the appended claims.
Claims (10)
1. An indoor trajectory error evaluation method based on an LSTM neural network is characterized by comprising the following steps:
acquiring a track step number sequence number and a reconstruction track point corresponding to the track step number sequence number;
constructing a multi-dimensional feature vector of the reconstructed track point, and obtaining a multi-dimensional feature vector set of the reconstructed track point according to the multi-dimensional feature vector of the reconstructed track point;
and for each target track step number sequence number, inputting the multi-dimensional feature vector set corresponding to the target track step number sequence number into the trained LSTM neural network model to obtain an indoor track deviation value corresponding to each target track step number sequence number, wherein the target track step number sequence number is used for representing the track step number sequence number which is greater than or equal to a preset value K.
2. The LSTM neural network-based indoor trajectory error assessment method of claim 1, wherein the constructing the multi-dimensional feature vector of the reconstructed trajectory points comprises:
obtaining the motion step length of the reconstructed track point, the motion direction of the reconstructed track point and the total track step number;
for each track step number, dividing the track step number by the total track step number to obtain a track step quotient corresponding to each track step number;
and for each track step number sequence number, the motion step length of the reconstructed track point, the motion direction of the reconstructed track point and the track step quotient form a multi-dimensional feature vector of the reconstructed track point.
3. The LSTM neural network-based indoor trajectory error assessment method of claim 1, wherein obtaining the multi-dimensional feature vector set of the reconstructed trajectory points according to the multi-dimensional feature vectors of the reconstructed trajectory points comprises:
and for each target track step number sequence number, combining the multidimensional characteristic vector corresponding to the first K-1 track step number sequence numbers adjacent to the target track step number sequence number and the multidimensional characteristic vector corresponding to the target track step number sequence number in sequence to obtain the multidimensional characteristic vector set of the reconstructed track point.
4. The LSTM neural network-based indoor trajectory error assessment method of claim 1, wherein the training process of the LSTM neural network model specifically comprises:
acquiring a training data set; wherein the training data set comprises a training multi-dimensional feature vector set and a training deviation value;
inputting the training multi-dimensional feature vector set to a preset initial network model to obtain a prediction deviation value;
obtaining a loss function according to the training deviation value and the prediction deviation value;
and training the initial network model according to the loss function to obtain an LSTM neural network model.
5. The LSTM neural network-based indoor trajectory error assessment method of claim 4, wherein said obtaining a training data set comprises:
acquiring a training track step number sequence number, training reconstruction track points corresponding to the training track step number sequence number and training real track points corresponding to the training track step number sequence number;
obtaining a training deviation value according to the training reconstructed track point and the training real track point;
obtaining training multidimensional characteristic vectors of the training reconstructed trajectory points according to the training reconstructed trajectory points;
and obtaining a training multi-dimensional feature vector set according to the training multi-dimensional feature vector.
6. The LSTM neural network-based indoor trajectory error assessment method of claim 5, wherein the deriving of the training deviation values from the training reconstructed trajectory points and the training real trajectory points comprises:
for each training track step number sequence number, subtracting the abscissa of the training real track point from the abscissa of the training reconstructed track point to obtain a track abscissa difference value;
for each training track step number sequence number, subtracting the ordinate of the training real track point from the ordinate of the training reconstructed track point to obtain a track ordinate difference value;
summing the square of the track horizontal coordinate difference value and the square of the track vertical coordinate difference value to obtain a sum value;
and squaring the sum to obtain a training deviation value.
7. The indoor trajectory error evaluation method based on the LSTM neural network according to claim 5, wherein the obtaining of the training multidimensional feature vector of the training reconstructed trajectory points according to the training reconstructed trajectory points comprises:
acquiring a training motion step length of the training reconstructed track point, a motion direction of the training reconstructed track point and a total step number of a training track;
for each training track step number sequence number, dividing the training track step number sequence number by the total training track step number to obtain a training track step number quotient corresponding to each training track step number sequence number;
and for each training track step number sequence number, the training multi-dimensional feature vector of the training reconstructed track point is formed by the motion step length of the training reconstructed track point, the motion direction of the training reconstructed track point and the training track step number quotient.
8. The LSTM neural network-based indoor trajectory error assessment method of claim 4, wherein the deriving a training multi-dimensional feature vector set according to the training multi-dimensional feature vector comprises:
taking the training track step number sequence number which is greater than or equal to the preset value K as a target training track step number sequence number;
and for each target training track step number sequence number, combining training multidimensional feature vectors corresponding to the first K-1 training track step number sequences adjacent to the target training track step number sequence number with training multidimensional feature vectors corresponding to the target training track step number sequence number in sequence to obtain a training multidimensional feature vector set of the target training track step number sequence number.
9. An intelligent terminal comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory, and wherein the one or more programs being configured to be executed by the one or more processors comprises instructions for performing the method of any of claims 1-8.
10. A non-transitory computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the method of any of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110752788.1A CN113537323B (en) | 2021-07-02 | 2021-07-02 | Indoor track error assessment method based on LSTM neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110752788.1A CN113537323B (en) | 2021-07-02 | 2021-07-02 | Indoor track error assessment method based on LSTM neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113537323A true CN113537323A (en) | 2021-10-22 |
| CN113537323B CN113537323B (en) | 2023-11-07 |
Family
ID=78126693
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110752788.1A Active CN113537323B (en) | 2021-07-02 | 2021-07-02 | Indoor track error assessment method based on LSTM neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537323B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101354787A (en) * | 2008-08-18 | 2009-01-28 | 浙江大学 | Intelligent method for extracting target movement track characteristic in vision monitoring searches |
| CN105674989A (en) * | 2016-01-27 | 2016-06-15 | 西北大学 | Indoor target motion track estimation method based on mobile phone built-in sensors |
| CN107976193A (en) * | 2017-11-21 | 2018-05-01 | 出门问问信息科技有限公司 | A kind of pedestrian's flight path estimating method, device, flight path infer equipment and storage medium |
| CN108444473A (en) * | 2018-03-20 | 2018-08-24 | 南京华苏科技有限公司 | Track localization method in a kind of pedestrian room |
| CN110264721A (en) * | 2019-07-01 | 2019-09-20 | 北京理工大学 | A kind of urban intersection nearby vehicle trajectory predictions method |
| CN110334167A (en) * | 2019-06-26 | 2019-10-15 | 浪潮软件集团有限公司 | Positional shift method for early warning and device based on neural network track |
| CN110715654A (en) * | 2019-10-10 | 2020-01-21 | 北京无限光场科技有限公司 | Motion track determination method and device of terminal equipment and electronic equipment |
| CN111208545A (en) * | 2018-11-22 | 2020-05-29 | 广州幻境科技有限公司 | Motion trail recording method and system |
| CN111754605A (en) * | 2020-06-28 | 2020-10-09 | 深圳壹账通智能科技有限公司 | User trajectory drawing method and device, computer equipment and storage medium |
| CN111783492A (en) * | 2020-06-09 | 2020-10-16 | 北京无限向溯科技有限公司 | Indoor positioning method and device based on two-dimensional code, mobile terminal and storage medium |
| CN112016400A (en) * | 2020-08-04 | 2020-12-01 | 香港理工大学深圳研究院 | Single-class target detection method and device based on deep learning and storage medium |
-
2021
- 2021-07-02 CN CN202110752788.1A patent/CN113537323B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101354787A (en) * | 2008-08-18 | 2009-01-28 | 浙江大学 | Intelligent method for extracting target movement track characteristic in vision monitoring searches |
| CN105674989A (en) * | 2016-01-27 | 2016-06-15 | 西北大学 | Indoor target motion track estimation method based on mobile phone built-in sensors |
| CN107976193A (en) * | 2017-11-21 | 2018-05-01 | 出门问问信息科技有限公司 | A kind of pedestrian's flight path estimating method, device, flight path infer equipment and storage medium |
| CN108444473A (en) * | 2018-03-20 | 2018-08-24 | 南京华苏科技有限公司 | Track localization method in a kind of pedestrian room |
| CN111208545A (en) * | 2018-11-22 | 2020-05-29 | 广州幻境科技有限公司 | Motion trail recording method and system |
| CN110334167A (en) * | 2019-06-26 | 2019-10-15 | 浪潮软件集团有限公司 | Positional shift method for early warning and device based on neural network track |
| CN110264721A (en) * | 2019-07-01 | 2019-09-20 | 北京理工大学 | A kind of urban intersection nearby vehicle trajectory predictions method |
| CN110715654A (en) * | 2019-10-10 | 2020-01-21 | 北京无限光场科技有限公司 | Motion track determination method and device of terminal equipment and electronic equipment |
| CN111783492A (en) * | 2020-06-09 | 2020-10-16 | 北京无限向溯科技有限公司 | Indoor positioning method and device based on two-dimensional code, mobile terminal and storage medium |
| CN111754605A (en) * | 2020-06-28 | 2020-10-09 | 深圳壹账通智能科技有限公司 | User trajectory drawing method and device, computer equipment and storage medium |
| CN112016400A (en) * | 2020-08-04 | 2020-12-01 | 香港理工大学深圳研究院 | Single-class target detection method and device based on deep learning and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| SHENGHAO QIN ET AL.: "Recurrent Attentive Neural Process for Sequential Data", 《ARXIV》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113537323B (en) | 2023-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109389030B (en) | Face characteristic point detection method and device, computer equipment and storage medium | |
| CN109063742B (en) | Butterfly identification network construction method and device, computer equipment and storage medium | |
| CN109840588B (en) | Neural network model training method, device, computer equipment and storage medium | |
| CN109285105A (en) | Method of detecting watermarks, device, computer equipment and storage medium | |
| CN109472213A (en) | Palm grain identification method, device, computer equipment and storage medium | |
| CN110824587A (en) | Image prediction method, image prediction device, computer equipment and storage medium | |
| CN111915650B (en) | Target tracking method and system based on improved twin network | |
| CN110633751A (en) | Training method of car logo classification model, car logo identification method, device and equipment | |
| CN111144398A (en) | Target detection method, target detection device, computer equipment and storage medium | |
| CN110796634A (en) | Dam state detection method and device, computer equipment and readable storage medium | |
| CN112001399B (en) | Image scene classification method and device based on local feature saliency | |
| CN111062324A (en) | Face detection method and device, computer equipment and storage medium | |
| CN112417988A (en) | Video multi-target tracking method, device and equipment based on deep learning | |
| CN111707294B (en) | Pedestrian navigation zero-speed interval detection method and device based on optimal interval estimation | |
| CN111491204A (en) | Video repair method, video repair device, electronic equipment and computer-readable storage medium | |
| CN111160303A (en) | Eye movement response information detection method and device, mobile terminal and storage medium | |
| CN110472588B (en) | Anchor point frame determining method and device, computer equipment and storage medium | |
| CN111754429A (en) | Motion vector post-processing method and device, electronic device and storage medium | |
| CN111178126A (en) | Target detection method, target detection device, computer equipment and storage medium | |
| CN110990135A (en) | Spark operation time prediction method and device based on deep migration learning | |
| CN112629565B (en) | Method, device and equipment for calibrating rotation relation between camera and inertial measurement unit | |
| CN111738126B (en) | Driver fatigue detection method and device based on Bayesian network and HMM | |
| CN113256683A (en) | Target tracking method and related equipment | |
| CN110929779B (en) | Reconstruction neuron quality detection method, ordered point cloud classification method and device | |
| CN113537323A (en) | Indoor track error evaluation method based on LSTM neural network |
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 |