CN117372969B - Monitoring scene-oriented abnormal event detection method - Google Patents
Monitoring scene-oriented abnormal event detection method Download PDFInfo
- Publication number
- CN117372969B CN117372969B CN202311675550.9A CN202311675550A CN117372969B CN 117372969 B CN117372969 B CN 117372969B CN 202311675550 A CN202311675550 A CN 202311675550A CN 117372969 B CN117372969 B CN 117372969B
- Authority
- CN
- China
- Prior art keywords
- entity
- detected
- information
- attribute information
- relation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 50
- 238000001514 detection method Methods 0.000 title claims abstract description 41
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 38
- 238000004422 calculation algorithm Methods 0.000 claims description 33
- 238000000034 method Methods 0.000 claims description 8
- 230000009191 jumping Effects 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 3
- 238000002372 labelling Methods 0.000 abstract description 5
- 206010000117 Abnormal behaviour Diseases 0.000 abstract description 3
- 230000011218 segmentation Effects 0.000 description 9
- 230000033001 locomotion Effects 0.000 description 6
- 230000005856 abnormality Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive 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
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a monitoring scene-oriented abnormal event detection method, which comprises the following steps: acquiring attribute information of each entity based on historical monitoring data in a certain monitoring scene; splitting the monitoring scene into a plurality of areas; acquiring each dimension probability distribution model of each entity based on the attribute information of each entity and the plurality of areas; acquiring attribute information of each entity to be detected based on the monitored data to be detected in the monitored scene; acquiring each dimension probability value of each entity to be detected based on attribute information of each entity to be detected and each dimension probability distribution model of each entity; calculating each dimension information quantity of each entity to be detected based on each dimension probability value of each entity to be detected; and judging whether the entities to be detected are abnormal or not based on the information quantity of each dimension of the entities to be detected. The invention can not only reduce the dependence on the labeling data and expand the detectable abnormal category, but also detect the unknown abnormal behavior.
Description
Technical Field
The invention relates to the technical field of anomaly detection, in particular to a monitoring scene-oriented anomaly event detection method.
Background
At present, most of anomaly detection algorithms of a monitoring scene are based on technologies such as image processing, pattern recognition, machine learning and the like, and abnormal events are detected and recognized by analyzing object behaviors and characteristics. However, these algorithms have some problems in terms of data dependence, detection category, generalization, etc., which limit their application and development in the field of monitoring scene anomaly detection.
1. Data dependence:
Deep learning based anomaly detection algorithms typically require a large amount of annotated training data to build accurate anomaly models. However, acquiring large scale annotation data is an expensive and time consuming task. This may be challenging for certain scenarios or specific anomaly categories, limiting the scope and effectiveness of the algorithm.
2. Detection category and generalization:
Many anomaly detection algorithms are modeled during the training phase based only on certain specific and known anomaly samples, which are limited not only in anomaly detection categories, but also do not allow for efficient detection of unknown anomalies (which results in poor generalization of the algorithm). For some complex and diversified monitoring scenes, the existing anomaly detection algorithm cannot be accurately adapted and processed.
Therefore, how to provide a monitoring scene-oriented abnormal event detection method, which can not only reduce the dependence on labeling data and expand the detectable abnormal category, but also detect unknown abnormal behaviors is a problem to be solved by those skilled in the art.
Disclosure of Invention
In view of the above, the present invention is directed to a monitoring scene-oriented abnormal event detection method.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a monitoring scene-oriented abnormal event detection method comprises the following steps:
S1: acquiring attribute information of each entity based on historical monitoring data in a certain monitoring scene;
S2: splitting the monitoring scene into a plurality of areas;
S3: acquiring each dimension probability distribution model of each entity based on the attribute information of each entity and the plurality of areas;
S4: acquiring attribute information of each entity to be detected based on the monitored data to be detected in the monitored scene;
s5: acquiring each dimension probability value of each entity to be detected based on attribute information of each entity to be detected and each dimension probability distribution model of each entity;
S6: calculating each dimension information quantity of each entity to be detected based on each dimension probability value of each entity to be detected;
S7: and judging whether the entities to be detected are abnormal or not based on the information quantity of each dimension of the entities to be detected.
Preferably, S1 further comprises:
Acquiring position attribute information and category attribute information of each entity in the historical monitoring data based on a target detection algorithm;
And/or acquiring speed attribute information of each entity in the history monitoring data based on a track tracking algorithm; and/or obtaining mask graph attribute information of each entity in the history monitoring data based on an instance segmentation algorithm;
preferably, S3 further comprises:
each dimension probability distribution model of each type of entity includes one or more of a position probability distribution model, a velocity probability distribution model, a mask map probability distribution model, a direction jump probability distribution model, a containment relationship probability distribution model, and a containment relationship jump probability distribution model.
Preferably, S3 further comprises:
modeling the position attribute information of each entity based on the category attribute information of each entity and the Gaussian mixture model to obtain a position probability distribution model of each entity;
Modeling the speed attribute information of each entity based on the category attribute information and the Gaussian model of each entity to obtain a speed probability distribution model of each entity;
Modeling the mask map attribute information of each entity based on the category attribute information and the Gaussian model of each entity to obtain mask map probability distribution models of various entities;
Obtaining direction information of each entity based on the position attribute information of each entity, and obtaining a direction jump probability distribution model of each entity based on the direction information statistics of each entity;
Obtaining a containing relation probability distribution model of each entity based on the position attribute information, the category attribute information and the plurality of region statistics of each entity;
and obtaining the containment relation jump probability distribution model of each entity based on the position attribute information, the category attribute information and the plurality of region statistics of each entity.
Preferably, S5 further comprises:
Inputting the position attribute information of each entity to be detected into a position probability distribution model of the category to which the position attribute information belongs to obtain a position probability value of each entity to be detected;
inputting the speed attribute information of each entity to be detected into a speed probability distribution model of the category to which the speed attribute information belongs to obtain a speed probability value of each entity to be detected;
inputting the mask map attribute related information of each entity to be detected into a mask map probability distribution model of the category to which the information belongs to obtain mask map probability values of each entity to be detected;
The method comprises the steps of obtaining direction information of each entity to be detected, and inputting the direction information into a direction jump probability distribution model of the category to which the direction information belongs to obtain a direction jump probability value of each entity to be detected;
Acquiring the containing relation information of each entity to be detected, and inputting the containing relation information into a containing relation probability distribution model of the category to which the containing relation information belongs to acquire the containing relation probability value of each entity to be detected; wherein the inclusion relation probability value represents a probability value that an entity to be detected belongs to a certain area;
Acquiring the relation jump information of each entity to be detected, and inputting the relation jump information into a relation jump probability distribution model of the category to which the relation jump probability distribution model belongs to acquire a relation jump probability value of each entity to be detected; wherein the inclusion relation jump probability value represents a probability value that an entity to be detected jumps from one area to another area;
Preferably, S6 further comprises:
Wherein, Representing the pixel position of the entity to be detected,/>A position probability value representing the entity to be detected,Representing the amount of location information of the entity to be detected; /(I)Representing the speed of the entity to be detected,/>Representing a velocity probability value of the entity to be detected,/>Representing the speed information quantity of the entity to be detected; /(I)Indicating that the entity to be detected is in the area/>The probability value of the occurrence is calculated,Representing the information quantity of the inclusion relation of the entity to be detected; /(I)Representing the detected entity from the area/>To region/>I.e. the probability value of the entity to be detected, including the relation jump probability value,/>Representing the information quantity of the relation jump of the entity to be detected; /(I)Representing the direction/>, from the entity to be detectedJump to direction/>I.e. the probability value of the direction jump of the entity to be detected,/>Indicating the direction jump information quantity of the entity to be detected; /(I)Mask map probability value representing entity to be detected,/>The mask map information amount representing the detection entity.
Preferably, proximity relation information of each entity is obtained based on the position attribute information of each entity;
obtaining a proximity relation probability distribution model based on the proximity relation information statistics of each entity;
Acquiring each piece of proximity relation information of each entity to be detected based on the position attribute information of each entity to be detected;
inputting each proximity relation information of each entity to be detected into a proximity relation probability distribution model to obtain each proximity relation probability of each entity to be detected;
calculating each proximity relation probability information amount of each entity to be detected based on each proximity relation probability of each entity to be detected ；
；
Wherein,Representing an i-th class entity,/>Represents the/>Class entity,/>；/>Represents the/>Class entity/>And/>Class entity/>Is a proximity relation probability of (a).
Preferably, S7 further comprises:
Calculating the sum of the information quantity of each dimension of each entity to be detected, and judging the entity to be detected as abnormal if the sum of the information quantity of each dimension is larger than the sum threshold of the information quantity;
Preferably, track information volume density I of each entity to be detected is calculated respectively, and if the track information volume density I is larger than a track information volume density threshold value, the entity to be detected is judged to be abnormal;
Wherein, ，/>Representing the sum of the information amounts of the entities to be detected in a certain frame.
Preferably, S7 further comprises:
Calculating to obtain scene information volume density L based on each dimension information volume of each entity to be detected, and if the scene information volume density L is larger than a scene information volume density threshold value, judging that the monitored scene is abnormal;
；
wherein m represents the number of entities to be detected in the monitoring scene, Representing the sum of the information amounts of the mth entity to be detected.
Preferably, S2 further comprises: and splitting the monitoring scene into a plurality of areas by adopting an AOG graph modeling mode.
Compared with the prior art, the invention discloses a monitoring scene-oriented abnormal event detection method, which has the following beneficial technical effects:
1. The invention does not need to carry out independent labeling on training data (namely history monitoring data), namely the invention does not depend on large-scale labeling data, and can reduce the cost of data labeling;
2. The invention judges whether the abnormality is caused by calculating the information quantity of the entity, and does not limit the detection of certain specific abnormality categories like an abnormality detection algorithm, and can detect more abnormality categories.
3. The invention judges whether the abnormal state is abnormal or not by calculating the information quantity of the entity, and even unusual abnormal behaviors can be detected because the information quantity is larger than that of normal data.
4. The target detection algorithm, the track tracking algorithm and the instance segmentation algorithm of the invention do not depend on a fixed monitoring scene as a general module, so that the invention can be quickly migrated to a new monitoring scene.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of the overall method of the present invention.
FIG. 2 is a traffic scene graph in an embodiment of the invention;
FIG. 3 is a schematic diagram of a vehicle to be detected in an embodiment of the present invention;
FIG. 4 is a graph of a vehicle position probability distribution model in an embodiment of the invention;
FIG. 5 is a graph of a vehicle inclusion relationship probability distribution model in an embodiment of the invention;
FIG. 6 is a graph of a vehicle inclusion relationship jump probability distribution model in an embodiment of the invention;
FIG. 7 is a constructed ST-AOG model in an embodiment of the 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 can be applied to any monitoring scene needing to detect abnormal events: such as traffic monitoring scenarios, campus monitoring scenarios.
The technical scheme of the invention is described below by taking a traffic monitoring scene as an example.
As shown in fig. 1, the embodiment of the invention discloses a traffic scene-oriented abnormal event detection method, which comprises the following steps:
S1: acquiring attribute information of each entity based on historical monitoring data in a certain traffic scene;
In one embodiment, the attribute information includes: location attribute information, category attribute information, speed attribute information, mask map attribute information, and the like; the entities include people, automobiles, bicycles and the like; the invention obtains one or more of the attribute information for each person, each car and each bicycle according to the specific traffic scene.
In one embodiment:
Location attribute information and category attribute information for each entity in the historical monitoring data is obtained based on a target detection algorithm (Yolov). Through an efficient target detection algorithm, the position of each entity can be accurately calibrated and distinguished from the background.
Acquiring speed attribute information of each entity in the history monitoring data based on a track tracking algorithm (Deep source); the track tracking algorithm tracks the entity in the time dimension, i.e. associates and tracks the motion track of the same entity between successive frames. And obtaining the speed attribute information of the entity by modeling and predicting the motion state of the entity.
Obtaining mask map attribute information of each entity in the history monitoring data based on an instance segmentation algorithm (SOLOv 2); the example segmentation algorithm is used for carrying out pixel-level segmentation on the entity, and generating a mask image of the entity, namely accurately segmenting the entity from the image. By means of an example segmentation algorithm we can obtain accurate contour and shape information for each entity.
S2: splitting the traffic scene into a plurality of areas;
In one embodiment, the traffic scene is split into several regions by means of AOG (AND-OR Graph) modeling.
Specific: it is assumed that the traffic scene mainly includes 5 areas of sidewalks, roadways, parking lanes, buildings, and the like. Each area is an AND node, AND the traffic scene can be split intoDescription is made. Each region has a moving entity, each entity being movable over a different region, each entity containing attribute information describing a feature of the entity. The entity attribute information may include: ID. Location, area, speed, object type, mask, etc. The upper limit of the number of examples is the sum N of objects that can be accommodated at the viewing angle, for example, the upper limit of the number n=20 is set.
FIG. 7 is a ST-AOG (space-time and or graph) model constructed for 5 regions;
S3: acquiring each dimension probability distribution model of each entity based on the attribute information of each entity and the plurality of areas;
In one embodiment, each dimensional probability distribution model of each type of entity includes one or more of a position probability distribution model, a velocity probability distribution model, a mask map probability distribution model, a direction jump probability distribution model, a containment relationship probability distribution model, and a containment relationship jump probability distribution model.
Wherein, all kinds of entities include: people, automobiles and bicycles, namely, the invention respectively constructs one or more of the probability distribution models for the people, the automobiles and the bicycles according to specific traffic scenes.
In one embodiment:
modeling the position attribute information of each entity based on the category attribute information of each entity and the Gaussian mixture model to obtain a position probability distribution model of each entity;
Modeling the speed attribute information of each entity based on the category attribute information and the Gaussian model of each entity to obtain a speed probability distribution model of each entity;
Modeling the mask map attribute information of each entity based on the category attribute information and the Gaussian model of each entity to obtain mask map probability distribution models of various entities; by counting the distribution condition of each entity mask graph of the history monitoring data, the area and color distribution characteristics of each entity in the image can be known, and the detection of abnormal appearance can be performed.
It should be noted that: the concrete steps for constructing the mask map probability distribution model are as follows:
1. obtaining a mask image of each entity in each frame of picture through an example segmentation algorithm;
2. Calculating the similarity of each two adjacent frame mask images in each entity, such as calculating the Euclidean distance of the two adjacent frame mask images;
3. Obtaining the similarity of mask graphs of various entities based on the category attribute information of the various entities;
4. modeling the mask graph similarity of various entities based on the Gaussian model to obtain mask graph probability distribution models of various entities.
Obtaining direction information of each entity based on the position attribute information of each entity, and obtaining a direction jump probability distribution model of each entity based on the direction information statistics of each entity;
it should be noted that: the direction information refers to the direction of the movement of the entity, and the direction information of each entity is obtained based on the position attribute information of each entity;
Specific: the direction is calculated by subtracting the coordinate position at the previous time from the coordinate position at the current time, and the direction of movement is determined.
Obtaining a relationship-containing probability distribution model of each entity based on the position attribute information, the category attribute information and the statistics of the plurality of areas (namely according to the constructed ST-AOG model);
And obtaining the containment relation jump probability distribution model of each entity based on the position attribute information, the category attribute information and the statistics of the plurality of areas (namely according to the constructed ST-AOG model).
S4: acquiring attribute information of each entity to be detected based on the monitoring data to be detected in the traffic scene;
in one embodiment:
And obtaining the position attribute information and the category attribute information of each entity to be detected in the monitoring data to be detected based on a target detection algorithm (Yolov).
Acquiring speed attribute information of each entity to be detected in the monitoring data to be detected based on a track tracking algorithm (Deep source);
obtaining mask graph attribute information of each entity to be detected in the monitoring data to be detected based on an example segmentation algorithm (SOLOv 2);
s5: acquiring each dimension probability value of each entity to be detected based on attribute information of each entity to be detected and each dimension probability distribution model of each entity;
in one embodiment:
Inputting the position attribute information of each entity to be detected into a position probability distribution model of the category to which the position attribute information belongs to obtain a position probability value of each entity to be detected;
inputting the speed attribute information of each entity to be detected into a speed probability distribution model of the category to which the speed attribute information belongs to obtain a speed probability value of each entity to be detected;
inputting the mask map attribute related information of each entity to be detected into a mask map probability distribution model of the category to which the information belongs to obtain mask map probability values of each entity to be detected;
it should be noted that: the mask map attribute related information of the entity to be detected is a similarity value (such as euclidean distance) of mask maps in two frames before and after the entity to be detected.
The method comprises the steps of obtaining direction information of each entity to be detected, and inputting the direction information into a direction jump probability distribution model of the category to which the direction information belongs to obtain a direction jump probability value of each entity to be detected;
It should be noted that: the direction information refers to the direction of the movement of the entity, and the direction information of each entity to be detected is obtained based on the position attribute information of each entity to be detected;
Specific: the direction is calculated by subtracting the coordinate position at the previous time from the coordinate position at the current time, and the direction of movement is determined.
Acquiring the inclusion relation information of each entity to be detected (determining the area of the entity to be detected based on the position attribute information of the current frame of the entity to be detected), and inputting the information into an inclusion relation probability distribution model of the category to which the information belongs to obtain the inclusion relation probability value of each entity to be detected; wherein the inclusion relation probability value represents a probability value that an entity to be detected belongs to a certain area;
Acquiring the information of the relation skip of each entity to be detected (determining the information of the area skip of the entity to be detected based on the position attribute information of the last frame and the current frame of the entity to be detected, wherein the area skip represents the skip from one area to another area), and inputting the information into the relation skip probability distribution model of the category to which the information belongs to obtain the relation skip probability value of each entity to be detected; wherein the inclusion relation jump probability value represents a probability value that an entity to be detected jumps from one area to another area;
S6: calculating each dimension information quantity of each entity to be detected based on each dimension probability value of each entity to be detected;
in one embodiment, S6 further comprises:
Wherein, Representing the pixel position of the entity to be detected,/>Representing a position probability value of an entity to be detected,/>Representing the amount of location information of the entity to be detected; /(I)Representing the speed of the entity to be detected,/>Representing a velocity probability value of the entity to be detected,/>Representing the speed information quantity of the entity to be detected; /(I)Indicating that the entity to be detected is in the area/>Probability value of occurrence,/>Representing the information quantity of the inclusion relation of the entity to be detected; /(I)Representing the detected entity from the area/>To region/>I.e. the probability value of the entity to be detected, including the relation jump probability value,/>Representing the information quantity of the relation jump of the entity to be detected; /(I)Representing the direction/>, from the entity to be detectedJump to direction/>I.e. the probability value of the direction jump of the entity to be detected,/>Indicating the direction jump information quantity of the entity to be detected; /(I)Mask map probability value representing entity to be detected,/>The mask map information amount representing the detection entity.
In one embodiment:
acquiring proximity relation information of each entity based on the position attribute information of each entity;
obtaining a proximity relation probability distribution model based on the proximity relation information statistics of each entity;
It should be noted that: the definition of proximity is: in a period of time t1, the pixel distance d between the entity 1 and the entity 2 is smaller than the set threshold d1 (always smaller in the period of time t 1), and the first-time proximity relation is the one that satisfies the above time and space conditions.
The specific acquisition method of the proximity relation probability distribution model comprises the following steps:
Assume that the dataset includes category 1, category 2, category N;
S1, counting the number of neighbors of the category 1 and the category 2;
Counting the adjacent quantity … of the category 2 and the category 3, and counting the adjacent quantity of the category 2 and the category N;
...
counting the number of the adjacent categories N-1 and N;
and counting the number of the neighbors of the category N and the category N.
S2, calculating the adjacent probability of the category X and the category Y, wherein the adjacent probability of the category X and the category Y is = (the adjacent quantity of the category X and the category Y)/(the adjacent quantity in the step 1), wherein，/>；
If there are 3 entities in one dataset, each is human O1 (category 1), human O2 (category 1), vehicle O3 (category 2). According to the calculation: o1 is adjacent to O2, O1 is adjacent to O3, and O2 is adjacent to O3. And (3) statistical acquisition: all neighbors are 3, category 1 and category 1 are 2, category 1 and category 2 are 1.
Thus: the proximity probability of the category 1 and the category 1 (namely, the human proximity probability) is 1/3, and the proximity probability of the category 1 and the category 2 (namely, the human proximity probability) is 2/3.
Acquiring each piece of proximity relation information of each entity to be detected based on the position attribute information of each entity to be detected;
inputting each proximity relation information of each entity to be detected into a proximity relation probability distribution model to obtain each proximity relation probability of each entity to be detected;
Specific: one entity to be detected may correspond to a plurality of proximity relation probabilities, taking entity to be detected a as an example, inputting the category of the entity to be detected a and another entity to be detected B and the position information in the continuous t1 time, and calculating whether the two entities are adjacent. And if the proximity relation is satisfied once, querying the proximity relation probability model to obtain proximity relation probability values of the two entities to be detected. Thus, a proximity probability of the entity a to be detected can be obtained.
Taking the entity A to be detected as an example, inputting the category of the entity A to be detected and the other entity C to be detected and the position information in the continuous t1 time, and calculating whether the two entities are adjacent. And if the proximity relation is satisfied once, querying the proximity relation probability model to obtain proximity relation probability values of the two entities to be detected. So far, another proximity probability of the entity a to be detected can be obtained.
Calculating each proximity relation probability information amount of each entity to be detected based on each proximity relation probability of each entity to be detected；
；
Wherein,Representing an i-th class entity,/>Represents the/>Class entity,/>；/>Represents the/>Class entity/>And/>Class entity/>Proximity relation probability,/>Including the probability of proximity between people, the probability of proximity between people and cars, the probability of proximity between cars, and so forth.
S7: and judging whether the entities to be detected are abnormal or not based on the information quantity of each dimension of the entities to be detected.
In one embodiment, S7 further comprises:
Calculating the sum of the information quantity of each dimension of each entity to be detected, and judging the entity to be detected as abnormal if the sum of the information quantity of each dimension is larger than the sum threshold of the information quantity;
Preferably, track information volume density I of each entity to be detected is calculated respectively, and if the track information volume density I is larger than a track information volume density threshold value, the entity to be detected is judged to be abnormal;
Wherein, ，/>Representing the sum of the information amounts of the entities to be detected in a certain frame.
In one embodiment, S7 further comprises:
Calculating to obtain scene information volume density L based on each dimension information volume of each entity to be detected, and if the scene information volume density L is larger than a scene information volume density threshold value, the traffic scene is abnormal;
；
wherein m represents the number of entities to be detected in the traffic scene, Representing the sum of the information amounts of the mth entity to be detected.
In one embodiment, S1 further comprises:
Acquiring position attribute information and category attribute information of each entity in the historical monitoring data based on a target detection algorithm;
And/or acquiring speed attribute information of each entity in the history monitoring data based on a track tracking algorithm; and/or obtaining mask graph attribute information of each entity in the history monitoring data based on an instance segmentation algorithm;
the technical scheme of the invention is described below with reference to a specific traffic scene:
As shown in fig. 2, the traffic scene is STREET SCENE as a dataset, and the following steps are adopted to obtain the total information of the car in the red frame in fig. 3:
1) The location attribute information for each car in the dataset is obtained STREET SCENE based on the target detection algorithm.
2) The traffic scene is split into ten AND nodes of A1, A2, A3, A4, A5, A6, A7, A8, A9, a10 (representing left sidewalk, right sidewalk, left motor vehicle roadway, right motor vehicle roadway, left bike lane, right bike lane, left parking lane, right parking lane, bush AND other areas, respectively) AND an ST-AOG (space-time AND or graph) model is constructed.
3) Performing mixed Gaussian model modeling on the position attribute of each automobile in STREET SCENE dataset to obtain an automobile position probability distribution model (shown in figure 4);
based on the ST-AOG (space-time and or graph) model, counting the occurrence probability of each automobile in each area, and obtaining an automobile inclusion relation probability model (shown in figure 5);
Based on the ST-AOG (space-time and or graph) model, counting the probability of each automobile jumping from one area to another area, and obtaining an automobile inclusion relation jumping probability model (shown in FIG. 6);
4) Obtaining position attribute information of the automobile in the red frame based on the target detection algorithm, and obtaining position probability of the automobile in the red frame based on the automobile position probability distribution model of fig. 4 ；
Determining the area of the current frame of the automobile based on the position attribute information of the current frame of the automobile in the red frame: as can be seen from FIG. 2, the vehicle is located in the left motor lane, and the inclusion probability value can be obtained by the vehicle inclusion probability model of FIG. 5；
Determining which region to jump to by which region (in this embodiment, the car is located in the left motor lane in the previous frame and the current frame) based on the position attribute information of the current frame and the position attribute information of the previous frame of the car in the red frame, and then obtaining the inclusion relation jump value of the car in the red frame according to the car inclusion relation jump probability model of fig. 6；
5) Calculating the position information quantity of the automobile in the red frameContains relation information quantity/>And contains the relation jump information quantity：
；
Wherein,Representing the pixel position of the car in the red frame;
；
；
6) Calculating the total information quantity H of the automobile in the red frame:
；
7) Selecting a reasonable threshold value As the information amount sum threshold. Meanwhile, through evaluating the test data, the corresponding information quantity sum with the highest AUC value in the test data is searched out and used as the information quantity sum threshold/>。
8) If it isAnd indicating that the automobile in the red frame is abnormal.
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. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (9)
1. The monitoring scene-oriented abnormal event detection method is characterized by comprising the following steps of:
S1: acquiring attribute information of each entity based on historical monitoring data in a certain monitoring scene;
the attribute information includes: one or more of location attribute information, category attribute information, speed attribute information, mask map attribute information;
S2: splitting the monitoring scene into a plurality of areas;
S3: acquiring each dimension probability distribution model of each entity based on the attribute information of each entity and the plurality of areas;
Each dimension probability distribution model of each entity comprises one or more of a position probability distribution model, a speed probability distribution model, a mask map probability distribution model, a direction jump probability distribution model, a relation-containing probability distribution model and a relation jump probability distribution model;
S4: acquiring attribute information of each entity to be detected based on the monitored data to be detected in the monitored scene;
s5: acquiring each dimension probability value of each entity to be detected based on attribute information of each entity to be detected and each dimension probability distribution model of each entity;
S6: calculating each dimension information quantity of each entity to be detected based on each dimension probability value of each entity to be detected;
S7: and judging whether the entities to be detected are abnormal or not based on the information quantity of each dimension of the entities to be detected.
2. The monitoring-scene-oriented abnormal event detection method according to claim 1, wherein S1 further comprises:
Acquiring position attribute information and category attribute information of each entity in the historical monitoring data based on a target detection algorithm;
and/or acquiring speed attribute information of each entity in the history monitoring data based on a track tracking algorithm;
and/or obtaining mask map attribute information of each entity in the history monitoring data based on a track tracking algorithm.
3. The monitoring-scene-oriented abnormal event detection method according to claim 2, wherein S3 further comprises:
modeling the position attribute information of each entity based on the category attribute information of each entity and the Gaussian mixture model to obtain a position probability distribution model of each entity;
Modeling the speed attribute information of each entity based on the category attribute information and the Gaussian model of each entity to obtain a speed probability distribution model of each entity;
Modeling the mask map attribute information of each entity based on the category attribute information and the Gaussian model of each entity to obtain mask map probability distribution models of various entities;
Obtaining direction information of each entity based on the position attribute information of each entity, and obtaining a direction jump probability distribution model of each entity based on the direction information statistics of each entity;
Obtaining a containing relation probability distribution model of each entity based on the position attribute information, the category attribute information and the plurality of region statistics of each entity;
and obtaining the containment relation jump probability distribution model of each entity based on the position attribute information, the category attribute information and the plurality of region statistics of each entity.
4. The monitoring-scene-oriented abnormal event detection method according to claim 3, wherein S5 further comprises:
Inputting the position attribute information of each entity to be detected into a position probability distribution model of the category to which the position attribute information belongs to obtain a position probability value of each entity to be detected;
inputting the speed attribute information of each entity to be detected into a speed probability distribution model of the category to which the speed attribute information belongs to obtain a speed probability value of each entity to be detected;
inputting the mask map attribute related information of each entity to be detected into a mask map probability distribution model of the category to which the information belongs to obtain mask map probability values of each entity to be detected;
The method comprises the steps of obtaining direction information of each entity to be detected, and inputting the direction information into a direction jump probability distribution model of the category to which the direction information belongs to obtain a direction jump probability value of each entity to be detected;
Acquiring the containing relation information of each entity to be detected, and inputting the containing relation information into a containing relation probability distribution model of the category to which the containing relation information belongs to acquire the containing relation probability value of each entity to be detected; wherein the inclusion relation probability value represents a probability value that an entity to be detected belongs to a certain area;
acquiring the relation jump information of each entity to be detected, and inputting the relation jump information into a relation jump probability distribution model of the category to which the relation jump probability distribution model belongs to acquire a relation jump probability value of each entity to be detected; wherein the inclusive relationship jump probability value represents a probability value that an entity to be detected jumps from one region to another region.
5. The method for detecting an abnormal event oriented to a monitored scene as set forth in claim 4, wherein S6 further comprises:
h1＝-log2P(x，y)；
h2＝-log2P(v)；
h3＝-∑log2P(ci)；
h4＝-Σlog2P(cj|ci)；
h5＝-Σlog2P(ej|ei)；
h6＝-log2Pmask；
Wherein (x, y) represents the pixel position of the entity to be detected, P (x, y) represents the position probability value of the entity to be detected, and h 1 represents the position information quantity of the entity to be detected; v represents the speed of the entity to be detected, P (v) represents the speed probability value of the entity to be detected, and h 2 represents the speed information quantity of the entity to be detected; p (c i) represents a probability value of the entity to be detected appearing in the region c i, and h 3 represents the information quantity containing relation of the entity to be detected; p (c j|ci) represents a probability value of the entity to be detected from the region c i to the region c j, namely a containment relationship skip probability value of the entity to be detected, and h 4 represents a containment relationship skip information amount of the entity to be detected; p (e j|ei) represents a probability value of the to-be-detected entity jumping from the direction e i to the direction e j, namely a direction jumping probability value of the to-be-detected entity, and h 5 represents a direction jumping information amount of the to-be-detected entity; p mask represents a mask map probability value of the entity to be detected, and h 6 represents a mask map information amount of the entity to be detected.
6. The monitoring-scene-oriented abnormal event detection method according to claim 2 or 5, further comprising:
acquiring proximity relation information of each entity based on the position attribute information of each entity;
obtaining a proximity relation probability distribution model based on the proximity relation information statistics of each entity;
Acquiring each piece of proximity relation information of each entity to be detected based on the position attribute information of each entity to be detected;
inputting each proximity relation information of each entity to be detected into a proximity relation probability distribution model to obtain each proximity relation probability of each entity to be detected;
Calculating each proximity relation probability information quantity h 7 of each entity to be detected based on each proximity relation probability of each entity to be detected;
h7＝-log P(mj,mi)；
Wherein m i represents an i-th class entity, m j represents a j-th class entity, i=j or i+.j; p (m j,mi) represents the proximity probability of the i-th class entity m i and the j-th class entity m j.
7. The monitoring-scene-oriented abnormal event detection method according to claim 1, wherein S7 further comprises:
and respectively calculating the sum of the information quantity of each dimension of each entity to be detected, and judging the entity to be detected as abnormal if the sum of the information quantity of each dimension is larger than the sum threshold of the information quantity.
8. The monitoring scene oriented abnormal event detection method according to claim 1 or 7, wherein;
respectively calculating the track information volume density I of each entity to be detected, and judging the entity to be detected as abnormal if the track information volume density I is larger than the track information volume density threshold;
Wherein, H (t) represents the sum of the information amounts of the entities to be detected in a certain frame.
9. The monitoring-scene-oriented abnormal event detection method according to claim 1 or 7, wherein S7 further comprises:
Calculating to obtain scene information volume density L based on each dimension information volume of each entity to be detected, and if the scene information volume density L is larger than a scene information volume density threshold value, judging that the monitored scene is abnormal;
Wherein m represents the number of the to-be-detected entities in the monitoring scene, and H m represents the sum of the information quantity of the mth to-be-detected entities.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311675550.9A CN117372969B (en) | 2023-12-08 | 2023-12-08 | Monitoring scene-oriented abnormal event detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311675550.9A CN117372969B (en) | 2023-12-08 | 2023-12-08 | Monitoring scene-oriented abnormal event detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117372969A CN117372969A (en) | 2024-01-09 |
| CN117372969B true CN117372969B (en) | 2024-05-10 |
Family
ID=89404520
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311675550.9A Active CN117372969B (en) | 2023-12-08 | 2023-12-08 | Monitoring scene-oriented abnormal event detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117372969B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117877124A (en) * | 2024-02-22 | 2024-04-12 | 暗物质(北京)智能科技有限公司 | Identification method and device for abnormal personnel behaviors, electronic equipment and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108688A (en) * | 2017-12-18 | 2018-06-01 | 青岛联合创智科技有限公司 | A kind of limbs conflict behavior detection method based on the extraction of low-dimensional space-time characteristic with theme modeling |
| CN109643485A (en) * | 2016-12-30 | 2019-04-16 | 同济大学 | A kind of urban highway traffic method for detecting abnormality |
| CN111368650A (en) * | 2020-02-18 | 2020-07-03 | 北京交通大学 | Crowd crowding behavior identification method based on track data |
| CN112508100A (en) * | 2020-12-07 | 2021-03-16 | 国网浙江省电力有限公司双创中心 | Method and device for monitoring running state of intelligent community public equipment |
| CN114663710A (en) * | 2022-04-25 | 2022-06-24 | 电信科学技术第五研究所有限公司 | Track recognition method, device, equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8892484B2 (en) * | 2012-09-28 | 2014-11-18 | Sphere Of Influence, Inc. | System and method for predicting events |
-
2023
- 2023-12-08 CN CN202311675550.9A patent/CN117372969B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109643485A (en) * | 2016-12-30 | 2019-04-16 | 同济大学 | A kind of urban highway traffic method for detecting abnormality |
| CN108108688A (en) * | 2017-12-18 | 2018-06-01 | 青岛联合创智科技有限公司 | A kind of limbs conflict behavior detection method based on the extraction of low-dimensional space-time characteristic with theme modeling |
| CN111368650A (en) * | 2020-02-18 | 2020-07-03 | 北京交通大学 | Crowd crowding behavior identification method based on track data |
| CN112508100A (en) * | 2020-12-07 | 2021-03-16 | 国网浙江省电力有限公司双创中心 | Method and device for monitoring running state of intelligent community public equipment |
| CN114663710A (en) * | 2022-04-25 | 2022-06-24 | 电信科学技术第五研究所有限公司 | Track recognition method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117372969A (en) | 2024-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368687B (en) | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation | |
| Chen et al. | Real time object detection, tracking, and distance and motion estimation based on deep learning: Application to smart mobility | |
| Lin et al. | A Real‐Time Vehicle Counting, Speed Estimation, and Classification System Based on Virtual Detection Zone and YOLO | |
| Li et al. | Traffic light recognition for complex scene with fusion detections | |
| CN111652912B (en) | Vehicle counting method and system, data processing equipment and intelligent shooting equipment | |
| CN111554105B (en) | Intelligent traffic identification and statistics method for complex traffic intersection | |
| Mallikarjuna et al. | Traffic data collection under mixed traffic conditions using video image processing | |
| Shirazi et al. | Vision-based turning movement monitoring: count, speed & waiting time estimation | |
| CN107730903A (en) | Parking offense and the car vision detection system that casts anchor based on depth convolutional neural networks | |
| Wang et al. | Probabilistic inference for occluded and multiview on-road vehicle detection | |
| CN117372969B (en) | Monitoring scene-oriented abnormal event detection method | |
| CN107315998A (en) | Vehicle class division method and system based on lane line | |
| CN112818935B (en) | Multi-lane congestion detection and duration prediction method and system based on deep learning | |
| Abdel-Aty et al. | Advances and applications of computer vision techniques in vehicle trajectory generation and surrogate traffic safety indicators | |
| CN117456482B (en) | Abnormal event identification method and system for traffic monitoring scene | |
| Ashraf et al. | HVD-net: a hybrid vehicle detection network for vision-based vehicle tracking and speed estimation | |
| Philipp et al. | Automated 3d object reference generation for the evaluation of autonomous vehicle perception | |
| CN116824520A (en) | Vehicle track prediction method and system based on ReID and graph convolution network | |
| Delavarian et al. | Multi‐camera multiple vehicle tracking in urban intersections based on multilayer graphs | |
| Dey et al. | Co-Optimizing Sensing and Deep Machine Learning in Automotive Cyber-Physical Systems | |
| CN112734835A (en) | High-speed intersection control method and device based on vehicle activity prediction | |
| CN114999183B (en) | Traffic intersection vehicle flow detection method | |
| CN109063675A (en) | Vehicle density calculation method, system, terminal and computer readable storage medium | |
| Gilroy et al. | Replacing the human driver: An objective benchmark for occluded pedestrian detection | |
| Bharadwaj et al. | Lane, Car, Traffic Sign and Collision Detection in Simulated Environment Using GTA-V |
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 |