CN117523382A - Abnormal track detection method based on improved GRU neural network - Google Patents
Abnormal track detection method based on improved GRU neural network Download PDFInfo
- Publication number
- CN117523382A CN117523382A CN202310887092.9A CN202310887092A CN117523382A CN 117523382 A CN117523382 A CN 117523382A CN 202310887092 A CN202310887092 A CN 202310887092A CN 117523382 A CN117523382 A CN 117523382A
- Authority
- CN
- China
- Prior art keywords
- track
- sensitive area
- neural network
- convex hull
- training
- 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
Links
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 41
- 238000001514 detection method Methods 0.000 title claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000000605 extraction Methods 0.000 claims abstract description 26
- 238000010586 diagram Methods 0.000 claims description 13
- 230000005856 abnormality Effects 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 4
- 230000003213 activating effect Effects 0.000 claims description 3
- 238000000034 method Methods 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 206010034701 Peroneal nerve palsy Diseases 0.000 description 1
- 238000012098 association analyses Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- 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
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The invention provides an abnormal track detection method based on an improved GRU neural network, which comprises the following steps: sorting the track data into a track sequence, performing density clustering, sorting the result, and adding a normal or abnormal label to the track sequence; building an improved GRU neural network, and inputting a track sequence set for recording label information into the neural network; arranging a sensitive area training set by taking longitude and latitude as characteristic targets, and performing density clustering on the sensitive area training set to obtain a first sensitive area extraction result; for a given self-data training point set, sorting and recording the largest convex hull polygon to obtain a second sensitive area extraction result; based on the track model with training and the extraction result of the sensitive area, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively researched and judged according to the output result of the track model and the real-time track slope tangential direction. The invention has simple input requirement and free parameter scheduling, and supports the detection of the tendency of the sensitive area of the road network live condition, map plotting and even self-training area.
Description
Technical Field
The invention relates to the technical field of abnormal track detection, in particular to an abnormal track detection method based on an improved GRU neural network.
Background
The time attribute and the space attribute of the user are described by the time-space data in sequence, the abundant semantic information such as target behaviors, states and preferences is contained, and understanding the semantic of the time-space data expression, especially the trace semantic information, plays an important role in urban planning, traffic logistics management and emergency prevention. Abnormal trajectory detection is also one of the important research directions.
The occurrence of anomalies often predicts the occurrence of events, and thus has a higher research value. Taking urban traffic as an example, the congestion can be identified in advance through the track abnormality of an individual or a cluster. However, because the track data has the characteristics of uncertainty, sparsity, bias distribution, large scale, quick updating and the like, the development and research are relatively complex, and the traditional abnormal track detection method mainly uses space recognition and is less combined with a time sequence. And is limited by technical means, traditional methods have difficulty in dealing with large-scale data sets and handling nonlinear and complex anomaly patterns. Compared with the traditional mode, the emerging cyclic neural network technology has better basic conditions of time sequence association analysis and abnormal feature capture, so that the cyclic neural network technology has great potential in the task of detecting abnormal tracks in the time sequence direction.
Therefore, how to iterate the abnormal track detection scheme by using the cyclic neural network technology and construct a technical cluster of a system for the same is a problem to be solved at present.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide an abnormal track detection method based on an improved GRU neural network, which uses the improved GRU neural network to train a target object track model to identify track deviation abnormality, and can embed any source or even self-trained sensitive area data into a detection model to realize collaborative detection of track deviation abnormality.
In order to achieve the above object, the present invention provides the following solutions:
an abnormal track detection method based on an improved GRU neural network comprises the following steps:
acquiring recent track data of an object to be detected, and sorting the track data into a track sequence with longitude and latitude as main characteristics;
performing density clustering on the track sequence based on a DBSCAN algorithm, sorting clustering results and adding normal or abnormal labels to the track sequence;
building an improved GRU neural network, and inputting the track sequence set with the recorded label information into the improved GRU neural network so as to train a track model of an object to be detected;
arranging a sensitive area training set by taking longitude and latitude as characteristic targets, and calling a DBSCAN algorithm to perform density clustering on the sensitive area training set to obtain a first sensitive area extraction result;
for a given self-data training point set, sorting and recording the largest convex hull polygon of the self-data training point set through a polygon convex hull algorithm to obtain a second sensitive area extraction result;
based on the track model, the first sensitive area extraction result and the second sensitive area extraction result which are subjected to training, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively researched and judged according to the track model output result and the real-time track slope tangential direction.
Preferably, the own data training point set includes: road network data and an adjusted point cluster set.
Preferably, the specific steps of the clustering operation include:
defining a trajectory dataset T s (T s ＝{T 1 ,T 2 ,…T n Track class label X (x= [ X ] 1 ,x 2 ,…,x n ] T )；
Initializing a track tag setThe clustering algorithm parameters k, eps, minPts are defined and calculated by the following algorithmThe content is as follows:
k＝2*dimension(T 1 )–1；
Eps＝k-distance(T s ,k)；
MinPts＝k+1；
wherein T is n For the nth trace, x n For the label corresponding to the nth track, dimension (T 1 ) For dimension extraction algorithms, feature dimensions for calculating and returning trajectory vectors, dimension (T 1 ) Constant 2; k-distance (T) s K) is a k-field algorithm: for any theodolite T i Calculate T s The distance from each point to the kth nearest neighbor is counted down, and then the k-dist diagram is made, and the inflection point value of the k-dist diagram is the threshold value Eps.
Preferably, the specific operation steps of the polygonal convex hull algorithm include:
sorting the input points from small to large according to the x coordinate, and respectively adding the input points into a left stack and a right stack;
selecting leftmost and rightmost points from input points as initial upper convex hull endpoints, and sequentially adding the initial upper convex hull endpoints into left and right stacks;
for each unprocessed point, determining whether it is inside the upper convex hull, if so, skipping, otherwise, performing the following operations: a. starting from the stack tops of the left stack and the right stack, ejecting the points forming a convex hull with the unprocessed points in the anticlockwise direction; b. adding the unprocessed points to the stack tops of the left stack and the right stack;
repeating the steps of 'judging whether each unprocessed point is inside the upper convex hull or not, if yes, skipping, otherwise, executing the following operation' until all the points are processed;
ejecting the tops of the left stack and the right stack to obtain a lower convex part;
and adding the lower convex hull part into the upper convex hull in a reverse order to obtain the complete convex hull.
Preferably, the specific operation steps of step 3 include:
building a three-layer improved GRU network model based on the connection of two layers of GRU networks with one layer of full-connection network;
initializing network model parameters: setting the number of hidden neurons of the first layer network and the second layer network as 10, and setting the number of hidden neurons of the full-connection layer as 24; setting the learning rate as 0.06, selecting sigmod and tanh as activating functions and setting the iteration times of the juxtaposed network as 100;
inputting a track sequence set containing label information after clustering, and training and obtaining a normalized track model of the target object;
and carrying out subsequent abnormality detection work based on the fitted normalized track model.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides an abnormal track detection method based on an improved GRU neural network, which comprises the following steps: acquiring recent track data of an object to be detected, and sorting the track data into a track sequence with longitude and latitude as main characteristics; performing density clustering on the track sequence based on a DBSCAN algorithm, sorting clustering results and adding normal or abnormal labels to the track sequence; building an improved GRU neural network, and inputting the track sequence set with the recorded label information into the improved GRU neural network so as to train a track model of an object to be detected; arranging a sensitive area training set by taking longitude and latitude as characteristic targets, and calling a DBSCAN algorithm to perform density clustering on the sensitive area training set to obtain a first sensitive area extraction result; for a given self-data training point set, sorting and recording the largest convex hull polygon of the self-data training point set through a polygon convex hull algorithm to obtain a second sensitive area extraction result; based on the track model, the first sensitive area extraction result and the second sensitive area extraction result which are subjected to training, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively researched and judged according to the track model output result and the real-time track slope tangential direction. The detection means of the invention has simple input requirement and free parameter scheduling, and supports the detection of the tendency of the sensitive area of the road network live condition, map plotting and even self-training area.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a technical route provided by an embodiment of the present invention;
FIG. 3 is an exemplary diagram of a normalized trajectory model of a training target object based on an improved GRU neural network, according to an embodiment of the present invention; FIG. 3 (a) shows the normal/abnormal trajectory classification results after DBSCAN clustering; FIG. 3 (b) shows an improved network model structure;
fig. 4 is a schematic diagram of extracting a sensitive area based on road network live according to an embodiment of the present invention; FIG. 4 (a) is live information; FIG. 4 (b) is sensitive area information;
FIG. 5 is a schematic diagram of training a sensitive area based on free data according to an embodiment of the present invention; FIG. 5 (a) is a visual display of the results of the self-contained data; FIG. 5 (b) is sensitive area information;
FIG. 6 is a graph of the detection result of an abnormal track according to an embodiment of the present invention;
fig. 7 is a schematic diagram of a detection result of deviation abnormality of a sensitive area track extracted based on road condition information according to an embodiment of the present invention;
fig. 8 is a schematic diagram of detection results of deviation abnormality of a sensitive area track extracted based on own data according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide an abnormal track detection method based on an improved GRU neural network, which has the advantages of simple detection means input requirement, free parameter scheduling and support of sensitive area trend detection of road network live condition, map plotting and even self-training areas.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Fig. 1 is a flowchart of a method provided by an embodiment of the present invention, and as shown in fig. 1, the present invention provides an abnormal track detection method based on an improved GRU neural network, including:
step 100: acquiring recent track data of an object to be detected, and sorting the track data into a track sequence with longitude and latitude as main characteristics;
step 200: performing density clustering on the track sequence based on a DBSCAN algorithm, sorting clustering results and adding normal or abnormal labels to the track sequence;
step 300: building an improved GRU neural network, and inputting the track sequence set with the recorded label information into the improved GRU neural network so as to train a track model of an object to be detected;
step 400: arranging a sensitive area training set by taking longitude and latitude as characteristic targets, and calling a DBSCAN algorithm to perform density clustering on the sensitive area training set to obtain a first sensitive area extraction result;
step 500: for a given self-data training point set, sorting and recording the largest convex hull polygon of the self-data training point set through a polygon convex hull algorithm to obtain a second sensitive area extraction result;
step 600: based on the track model, the first sensitive area extraction result and the second sensitive area extraction result which are subjected to training, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively researched and judged according to the track model output result and the real-time track slope tangential direction.
Fig. 2 is a schematic diagram of a technical route provided by an embodiment of the present invention, as shown in fig. 2, an abnormal track detection method based on an improved GRU neural network, specifically, a normalized track model training and a sensitive area definition, including the following steps:
step 1: acquiring recent track data of an object to be detected, and sorting the recent track data into a track sequence taking longitude and latitude as main characteristics;
step 2: performing density clustering on the track sequence based on a DBSCAN algorithm, sorting clustering results and adding normal/abnormal labels to the tracks;
step 3: building an improved GRU neural network, inputting a track sequence set recording label information into the network to train a track model of an object to be detected; if no trace deviation abnormal detection is required, the step 6 is carried out, if the trace deviation sensitive area is public data such as road network live, the step 5 is carried out, and if the trace deviation sensitive area is required to be self-trained, the step 4 is carried out;
step 4: arranging a sensitive area training set by taking longitude and latitude as characteristic targets, calling a DBSCAN algorithm to perform density clustering on the sensitive area training set, and increasing and decreasing the point cluster set according to requirements;
step 5: for given road network data, the adjusted point cluster set or other public data, the largest convex hull polygon is tidied by a polygon convex hull algorithm and recorded;
step 6: based on the training track model and the sensitive area convex hull polygon, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively and tangentially judged according to the track model output result and the real-time track slope.
The clustering operation involved in the step 2 and the step 4 specifically comprises the following steps:
step 21: defining a trajectory dataset T s (T s ＝{T 1 ,T 2 ,…T n Track class label X (x= [ X ] 1 ,x 2 ,…,x n ] T )；
Step 22: initializing a track tag setThe clustering algorithm parameters k, eps, minPts are defined and the content is calculated by the following algorithm:
k＝2*dimension(T 1 )–1 (1)
Eps＝k-distance(T s ,k) (2)
MinPts＝k+1 (3)
wherein dimension (T) 1 ) For dimension extraction algorithm for calculating and returning feature dimension of trajectory vector, clustering operation is performed only with longitude and latitude as feature input in the present invention, so dimension (T 1 ) Constant 2; k-distance (T) s K) is a k-field algorithm: for any theodolite T i Calculate T s The distance from each point to the kth nearest neighbor is counted down, and then the k-dist diagram is made, and the inflection point value of the k-dist diagram is the threshold value Eps.
The specific operation steps of the Melkman algorithm involved in the step 5 include:
step 51: sorting the input points from small to large according to the x coordinate, and respectively adding the input points into a left stack and a right stack;
step 52: selecting leftmost and rightmost points from input points as initial upper convex hull endpoints, and sequentially adding the initial upper convex hull endpoints into left and right stacks;
step 53: for each unprocessed point, determining whether it is inside the upper convex hull, if so, skipping, otherwise, performing the following operations: a. starting from the stack tops of the left stack and the right stack, ejecting the point forming a convex hull with the point in the anticlockwise direction. b. Adding the point to the stack tops of the left stack and the right stack;
step 54: repeating the step 3 until all points are processed;
step 55: ejecting the tops of the left stack and the right stack to obtain a lower convex part;
step 56: and adding the lower convex hull part into the upper convex hull in a reverse order to obtain the complete convex hull.
The specific operation steps of the step 3 comprise:
step 31: building a three-layer improved GRU network model based on the connection of two layers of GRU networks with one layer of full-connection network;
step 32: initializing network model parameters: the number of the hidden neurons of the first layer network and the second layer network is 10, and the number of the hidden neurons of the full-connection layer is 24. Setting the learning rate as 0.06, selecting sigmod and tanh as activating functions and setting the iteration times of the juxtaposed network as 100;
step 33: inputting a track sequence set containing label information after clustering, and training and obtaining a normalized track model of the target object;
step 34: and carrying out subsequent abnormality detection work based on the fitted normalized track model.
Specifically, in the embodiment, based on calculus, linear algebra and probability theory, sequence features are automatically learned through a neural network, a normalized track model is established to identify track deviation abnormality, and meanwhile, any source of sensitive area data or even self-training area is supported to be embedded into a detection model to detect track deviation, so that integrated detection of two types of abnormality is realized.
1. Gate-controlled circulation unit
Gating loop unit (GRU) is one type of loop neural network, by introducing update gate v t Reset gate r t The two cases of gating units solve the problems of gradient disappearance, explosion and the like in the RNN, and the nonlinear problem of data can be more effectively solved by combining the full connection layer.
The results shown in FIG. 3 (a) will be input as tag features into the improved network model structure shown in FIG. 3 (b); the forward propagation function corresponding to fig. 3 (b) is organized as follows, where k is the number of layers corresponding to the layer network:
r t ＝σ(λ r ·[h t-1 ,X t ]+b r )
v t ＝σ(λ v ·[h t-1 ,X t ]+b v )
h t ＝tanh(λ h ·[r t ·Y t-1 ,X t ]+b h )
y t ＝(1-v t )·Y t-1 +v t ·h t
and selecting a proper loss function to be back-propagated after forward propagation to update model parameters, and iterating until the loss value tends to be stable, and then finishing training of the target object normalized trajectory model.
2. Sensitive area extraction based on public data such as road condition information
As shown in fig. 4, the live information shown in fig. 4 (a) can be directly obtained by an open platform such as a high-altitude map, and the sensitive area information shown in fig. 4 (b) is obtained through processing of algorithms such as hough transformation or image extraction; for the given image P containing the sensitive domain, a Hough transformation algorithm is called to extract the sensitive area network information. The algorithm involves the following formula:
ρ＝x cosθ+y sinθ
each point (x, y) in the image is converted into a sinusoid in the parameter space. The intersection of these curves in the parameter space represents a collinear point in the image space. By detecting the peak point in the parameter space, a straight line in the image space can be obtained. And any one straight line l may be defined by two points, i.e. a fixed point F outside the straight line and the foot P of l. Since the coordinates of F are fixed, the position of the straight line l is only related to the drop foot P. The units of the abscissa and the ordinate of the foot drop P are uniform, and thus, the straight line in the image space is converted into a point in the rectangular coordinate system. And finally, obtaining a sensitive area extraction result of the P profile according to the coordinates of the extreme point and the fixed point of the parameter space.
3. Sensitive area extraction based on owned data
As shown in fig. 5, the visualized display result of the self-data in fig. 5 (a) is trained by Melkman or other effective algorithms to obtain the sensitive area information shown in fig. 5 (b), and for the self-data training point set of the given longitude and latitude coordinates, the invention selects to apply the Melkman algorithm to determine the convex hull and construct and generate the sensitive area thereof. The algorithm involves the following formula:
u×v＝|u||v|sinθ
the above formula is used for calculating the vector cross product, the triangle area and the distance from the point to the straight line when the convex hull is generated, and the three are used for calculating the boundary of the convex hull.
Based on the mathematical principle, the abnormal track detection method based on the improved GRU neural network can be obtained, and comprises the following steps:
step 1: acquiring recent track data of an object to be detected, and sorting the recent track data into a track sequence taking longitude and latitude as main characteristics;
step 2: performing density clustering on the track sequence based on a DBSCAN algorithm, sorting clustering results and adding normal/abnormal labels to the tracks;
step 3: building an improved GRU neural network, inputting a track sequence set recording label information into the network to train a track model of an object to be detected; if no trace deviation abnormal detection is required, the step 6 is carried out, if the trace deviation sensitive area is public data such as road network live, the step 5 is carried out, and if the trace deviation sensitive area is required to be self-trained, the step 4 is carried out;
step 4: arranging a sensitive area training set by taking longitude and latitude as characteristic targets, calling a DBSCAN algorithm to perform density clustering on the sensitive area training set, and increasing and decreasing the point cluster set according to requirements;
step 5: for given road network data, the adjusted point cluster set or other public data, the sensitive polygon domain is arranged and recorded through a polygon convex hull or other effective algorithms;
step 6: based on the training track model and the sensitive area convex hull polygon, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively and tangentially judged according to the track model output result and the real-time track slope.
Examples
Firstly, acquiring historical track data of a target object, a real-time road condition map of a Goldmap and a self-training point set data set;
secondly, building an improved GRU neural network, and training a normalized track model of the GRU neural network aiming at a target object;
thirdly, extracting sensitive area information in road network data, a self-training point set or other public data according to actual needs, and embedding the sensitive area information into the abnormal track detection method model;
and finally, accessing instant track data of the target object, and detecting the abnormal track state according to the method. The three cases of track deviation, road condition track deviation and self-training area track deviation in this embodiment correspond to fig. 6 to 8 respectively.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (5)
1. An abnormal track detection method based on an improved GRU neural network is characterized by comprising the following steps:
acquiring recent track data of an object to be detected, and sorting the track data into a track sequence with longitude and latitude as main characteristics;
performing density clustering on the track sequence based on a DBSCAN algorithm, sorting clustering results and adding normal or abnormal labels to the track sequence;
building an improved GRU neural network, and inputting the track sequence set with the recorded label information into the improved GRU neural network so as to train a track model of an object to be detected;
arranging a sensitive area training set by taking longitude and latitude as characteristic targets, and calling a DBSCAN algorithm to perform density clustering on the sensitive area training set to obtain a first sensitive area extraction result;
for a given self-data training point set, sorting and recording the largest convex hull polygon of the self-data training point set through a polygon convex hull algorithm to obtain a second sensitive area extraction result;
based on the track model, the first sensitive area extraction result and the second sensitive area extraction result which are subjected to training, the real-time track of the object to be detected is transmitted, and track deviation or track deviation abnormal information is respectively researched and judged according to the track model output result and the real-time track slope tangential direction.
2. The improved GRU neural network based anomaly track detection method of claim 1, wherein the owned data training point set comprises: road network data and an adjusted point cluster set.
3. The abnormal trajectory detection method based on an improved GRU neural network of claim 1, wherein the clustering operation specifically comprises:
defining a trajectory dataset T s (T s ＝{T 1 ,T 2 ,…T n Track class label X (x= [ X ] 1 ,x 2 ,…,x n ] T )；
Initializing a track tag setThe clustering algorithm parameters k, eps, minPts are defined and the content is calculated by the following algorithm:
k＝2*dimension(T 1 )–1；
Eps＝k-distance(T s ,k)；
MinPts＝k+1；
wherein T is n For the nth trace, x n For the label corresponding to the nth track, dimension (T 1 ) For dimension extraction algorithms, feature dimensions for calculating and returning trajectory vectors, dimension (T 1 ) Constant 2; k-distance (T) s K) is a k-field algorithm: for any theodolite T i Calculate T s The distance from each point to the kth nearest neighbor is counted down, and then the k-dist diagram is made, and the inflection point value of the k-dist diagram is the threshold value Eps.
4. The abnormal track detection method based on the improved GRU neural network according to claim 1, wherein the specific operation steps of the polygonal convex hull algorithm include:
sorting the input points from small to large according to the x coordinate, and respectively adding the input points into a left stack and a right stack;
selecting leftmost and rightmost points from input points as initial upper convex hull endpoints, and sequentially adding the initial upper convex hull endpoints into left and right stacks;
for each unprocessed point, determining whether it is inside the upper convex hull, if so, skipping, otherwise, performing the following operations: a. starting from the stack tops of the left stack and the right stack, ejecting the points forming a convex hull with the unprocessed points in the anticlockwise direction; b. adding the unprocessed points to the stack tops of the left stack and the right stack;
repeating the steps of 'judging whether each unprocessed point is inside the upper convex hull or not, if yes, skipping, otherwise, executing the following operation' until all the points are processed;
ejecting the tops of the left stack and the right stack to obtain a lower convex part;
and adding the lower convex hull part into the upper convex hull in a reverse order to obtain the complete convex hull.
5. The abnormal trajectory detection method based on an improved GRU neural network of claim 1, wherein the specific operation steps of step 3 include:
building a three-layer improved GRU network model based on the connection of two layers of GRU networks with one layer of full-connection network;
initializing network model parameters: setting the number of hidden neurons of the first layer network and the second layer network as 10, and setting the number of hidden neurons of the full-connection layer as 24; setting the learning rate as 0.06, selecting sigmod and tanh as activating functions and setting the iteration times of the juxtaposed network as 100;
inputting a track sequence set containing label information after clustering, and training and obtaining a normalized track model of the target object;
and carrying out subsequent abnormality detection work based on the fitted normalized track model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310887092.9A CN117523382A (en) | 2023-07-19 | 2023-07-19 | Abnormal track detection method based on improved GRU neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310887092.9A CN117523382A (en) | 2023-07-19 | 2023-07-19 | Abnormal track detection method based on improved GRU neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117523382A true CN117523382A (en) | 2024-02-06 |
Family
ID=89748362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310887092.9A Pending CN117523382A (en) | 2023-07-19 | 2023-07-19 | Abnormal track detection method based on improved GRU neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117523382A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657746A (en) * | 2015-01-29 | 2015-05-27 | 电子科技大学 | Anomaly detection method based on vehicle trajectory similarity |
| US9292792B1 (en) * | 2012-09-27 | 2016-03-22 | Lockheed Martin Corporation | Classification systems and methods using convex hulls |
| US20190392205A1 (en) * | 2017-02-28 | 2019-12-26 | SZ DJI Technology Co., Ltd. | Recognition method and apparatus and mobile platform |
| WO2021115143A1 (en) * | 2019-12-13 | 2021-06-17 | 华为技术有限公司 | Motion trajectory processing method, medium, apparatus, and device |
| CN113052219A (en) * | 2021-03-16 | 2021-06-29 | 浙江大学 | Abnormal track detection method and device and electronic equipment |
| US20220172050A1 (en) * | 2020-11-16 | 2022-06-02 | UMNAI Limited | Method for an explainable autoencoder and an explainable generative adversarial network |
| CN115062713A (en) * | 2022-06-24 | 2022-09-16 | 中国矿业大学 | GCN-GRU-based open-pit mine truck stay area activity identification method |
| CN115563652A (en) * | 2022-10-31 | 2023-01-03 | 上海交通大学 | Track embedding leakage prevention method and system |
| WO2023029461A1 (en) * | 2021-08-31 | 2023-03-09 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Massive high-dimensional ais trajectory data clustering method |
-
2023
- 2023-07-19 CN CN202310887092.9A patent/CN117523382A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9292792B1 (en) * | 2012-09-27 | 2016-03-22 | Lockheed Martin Corporation | Classification systems and methods using convex hulls |
| CN104657746A (en) * | 2015-01-29 | 2015-05-27 | 电子科技大学 | Anomaly detection method based on vehicle trajectory similarity |
| US20190392205A1 (en) * | 2017-02-28 | 2019-12-26 | SZ DJI Technology Co., Ltd. | Recognition method and apparatus and mobile platform |
| WO2021115143A1 (en) * | 2019-12-13 | 2021-06-17 | 华为技术有限公司 | Motion trajectory processing method, medium, apparatus, and device |
| US20220172050A1 (en) * | 2020-11-16 | 2022-06-02 | UMNAI Limited | Method for an explainable autoencoder and an explainable generative adversarial network |
| CN113052219A (en) * | 2021-03-16 | 2021-06-29 | 浙江大学 | Abnormal track detection method and device and electronic equipment |
| WO2023029461A1 (en) * | 2021-08-31 | 2023-03-09 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Massive high-dimensional ais trajectory data clustering method |
| CN115062713A (en) * | 2022-06-24 | 2022-09-16 | 中国矿业大学 | GCN-GRU-based open-pit mine truck stay area activity identification method |
| CN115563652A (en) * | 2022-10-31 | 2023-01-03 | 上海交通大学 | Track embedding leakage prevention method and system |
Non-Patent Citations (2)
| Title |
|---|
| GUOHAO TANG ET AL: "Low‑cost and high‑performance abnormal trajectory detection based on the GRU model with deep spatiotemporal sequence analysis in cloud computing", 《CLOUD COMPUT》, 5 March 2024 (2024-03-05), pages 1 - 17, XP021333422, DOI: 10.1186/s13677-024-00611-1 * |
| 权宇澄: "基于改进DBSCAN 方法的移动对象轨迹停留分析", 《测绘与空间地理信息》, vol. 39, no. 10, 31 October 2016 (2016-10-31), pages 1 - 5 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977918B (en) | Target detection positioning optimization method based on unsupervised domain adaptation | |
| Jeong et al. | Image preprocessing for efficient training of YOLO deep learning networks | |
| CN111079674B (en) | Target detection method based on global and local information fusion | |
| CN110569843B (en) | Intelligent detection and identification method for mine target | |
| CN104992223A (en) | Dense population estimation method based on deep learning | |
| CN113486726A (en) | Rail transit obstacle detection method based on improved convolutional neural network | |
| Ren et al. | A novel squeeze YOLO-based real-time people counting approach | |
| CN112861917B (en) | Weak supervision target detection method based on image attribute learning | |
| CN113850242B (en) | Storage abnormal target detection method and system based on deep learning algorithm | |
| CN112508014A (en) | Improved YOLOv3 target detection method based on attention mechanism | |
| CN114600130A (en) | Process for learning new image classes without labels | |
| Melek et al. | Object detection in shelf images with YOLO | |
| CN111382709A (en) | Insulator image detection method based on unmanned aerial vehicle inspection | |
| CN111611933A (en) | Information extraction method and system for document image | |
| Wu et al. | Single shot multibox detector for vehicles and pedestrians detection and classification | |
| CN115019133A (en) | Method and system for detecting weak target in image based on self-training and label anti-noise | |
| CN110347853A (en) | A kind of image hash code generation method based on Recognition with Recurrent Neural Network | |
| CN113129336A (en) | End-to-end multi-vehicle tracking method, system and computer readable medium | |
| Kajabad et al. | YOLOv4 for urban object detection: Case of electronic inventory in St. Petersburg | |
| CN111539390A (en) | Small target image identification method, equipment and system based on Yolov3 | |
| INTHIYAZ et al. | YOLO (YOU ONLY LOOK ONCE) Making Object detection work in Medical Imaging on Convolution detection System. | |
| CN117523382A (en) | Abnormal track detection method based on improved GRU neural network | |
| CN112668643B (en) | Semi-supervised significance detection method based on lattice tower rule | |
| Wen et al. | A comprehensive review of deep reinforcement learning for object detection | |
| CN111273779B (en) | Dynamic gesture recognition method based on self-adaptive space supervision |
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 |