CN112329627B - High-altitude throwing object distinguishing method - Google Patents

High-altitude throwing object distinguishing method Download PDF

Info

Publication number
CN112329627B
CN112329627B CN202011224115.0A CN202011224115A CN112329627B CN 112329627 B CN112329627 B CN 112329627B CN 202011224115 A CN202011224115 A CN 202011224115A CN 112329627 B CN112329627 B CN 112329627B
Authority
CN
China
Prior art keywords
thrower
projectile
motion
iteration
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011224115.0A
Other languages
Chinese (zh)
Other versions
CN112329627A (en
Inventor
闫政
杜勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing Lanhui Information Technology Co ltd
Original Assignee
Chongqing Lanhui Information Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chongqing Lanhui Information Technology Co ltd filed Critical Chongqing Lanhui Information Technology Co ltd
Priority to CN202011224115.0A priority Critical patent/CN112329627B/en
Publication of CN112329627A publication Critical patent/CN112329627A/en
Application granted granted Critical
Publication of CN112329627B publication Critical patent/CN112329627B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of 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
    • 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
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Computational Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Algebra (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Operations Research (AREA)
  • Databases & Information Systems (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Aerodynamic Tests, Hydrodynamic Tests, Wind Tunnels, And Water Tanks (AREA)

Abstract

The invention discloses a method for distinguishing high-altitude throws, which comprises the following steps: (S1) a camera capturing a building floor; (S2) detecting and extracting a moving object in a camera picture using a frame difference method; (S3) extracting a motion trail coordinate sequence of the moving object; and (S4) inputting the motion trail coordinate sequence of the moving object into a pre-trained neural network discrimination model to carry out classification discrimination, and discriminating whether the moving object belongs to a high-altitude parabolic object or not. By the high-altitude projectile object identification method, high-altitude projectile object motion detection and identification are converted into a sequence classification problem, so that the detection performance is improved, and meanwhile, the complexity of a detection system is reduced.

Description

High-altitude throwing object distinguishing method
Technical Field
The invention relates to the technical field of high-altitude projectile detection, in particular to a high-altitude projectile discrimination method.
Background
High altitude parabolic is called as 'pain suspended above a city', and high altitude parabolic behaviors are paid attention to all the time, and social hazards caused by the high altitude parabolic are also great when the high altitude parabolic is taken as a city non-civilized behavior. Because most of the implementation places are high floors, few witness persons exist, the throwing time is short, and even people intentionally hide and remove the figure when throwing things, related departments are difficult to follow up legal responsibility of the thrower, and the events are frequent and not limited. Therefore, the identification and evidence collection of high-altitude parabolic behaviors have become an urgent need for current urban management and property management.
The method comprises the steps of firstly realizing the extraction of a moving target in a continuous frame through a classical moving target detection method, such as background modeling or a frame difference method, then fitting the track of the moving target in the continuous frame through the coordinates of the moving target in each frame image, and comparing the track with the known parabolic track or parabolic formula to calculate the parabolic track, thereby finally determining whether the moving track of the target is the parabolic track. In practical application, the mode has a certain problem that as the roof of the high-rise building is influenced by atmospheric activity, turbulent flow exists on the roof, and the types of parabolic targets are various, and the pneumatic shapes are quite different, the motion trail of the high-altitude parabolic targets is quite complex and is difficult to describe by a theoretical formula, and the track sample is difficult to acquire by comparing with the known high-altitude parabolic trail sample, meanwhile, the whole sample library needs to be traversed when the track comparison is carried out, the calculation complexity is high, and the system is complex; the simulation mode of the experiment is adopted, the experiment cost is high due to the limitation of the experiment field, the variable wind field conditions in the actual scene cannot be simulated, and the motion trail of the thrower with different aerodynamic shapes under different wind fields is difficult to obtain, so that the model lacks effective training sample data in the training process, and the distinguishing performance of the model is directly influenced.
Disclosure of Invention
The invention aims to provide a high-altitude projectile object identification method, which converts high-altitude projectile object motion detection and identification into a sequence classification problem, improves detection performance and reduces complexity of a detection system.
In order to achieve the above purpose, the invention provides a method for discriminating high altitude throwers, comprising the following steps:
(S1) a camera capturing a building floor;
(S2) detecting and extracting a moving object in a camera picture using a frame difference method;
(S3) extracting a motion trail coordinate sequence of the moving object;
and (S4) inputting the motion trail coordinate sequence of the moving object into a pre-trained neural network discrimination model to carry out classification discrimination, and discriminating whether the moving object belongs to a high-altitude parabolic object or not.
Further, if the moving object belongs to the high-altitude parabolic object, the moving characteristic of the moving object is added into a training set of the neural network discrimination model, and model iteration is carried out.
Further, the pre-trained neural network discrimination model is obtained by the following steps:
(S41) obtaining motion trail coordinate sequence samples of throwers with different aerodynamic shapes under different wind field simulation conditions through computer simulation;
and (S42) training the discrimination neural network by taking the sample set of the projectile motion trail coordinate sequences as a training set to obtain a trained neural network discrimination model.
Further, the method for obtaining the motion trail samples of the thrower with different aerodynamic shapes under different wind field conditions through computer simulation specifically comprises the following steps:
(S411) randomly generating 1 throwing object through computer simulation, wherein the shape of the throwing object is randomly determined in 3 shapes of a block, a rod and a plate;
(S412) initializing the motion state of the projectile and setting the iteration time step delta t and the total iteration step number for calculating the geometric center point coordinates of the projectile, calculating and recording the geometric center point coordinates on each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the projectile in the initial state and the geometric center point coordinates on each iteration time step tag Saving to a database;
(S413) obtaining a motion trail sample P by arranging throwers with different aerodynamic shapes to move under different wind field conditions tag1 、P tag2 ···P tagm Generate 1 track sample set p= { P tag1 ,P tag2 ,...,P tagm -a }; optionally, the initializing the projectile motion state specifically includes the following: randomly setting initial velocity vectors for a projectileThe angle is arbitrarily selected, and-> Is in units of m/s; projectile rotation speed +.>The simulation time is synchronous with the wind field simulation time; the initial point of the projectile motion is the center of the upper edge of the simulated wind field; wherein, the windward angle of the plate-shaped throwing object is +.>Randomly taking value, and (E) is (are) added>
Further, the geometric center point coordinates on each iteration time step are calculated and recorded, and the geometric center point coordinates of the throwing object in the initial state and the geometric center point coordinates on each iteration time step form 1 coordinate sequence P tag Saving to a database; the method specifically comprises the following steps:
(Q1) obtaining at t from the simulated wind field n-1 Wind speed vector at moment projectile motion positionCalculating t by combining with the morphological characteristic parameters of the throwing object n-1 Surface wind force of moment throwing object +.>And->Wherein,and->Respectively t n-1 Surface wind force of moment thrower in 3 dimensions x, y and z direction of space, +.>At t n-1 The throwing object is turned over at any timeForce application; wherein t is n ＝t 0 +nDeltat, n is greater than or equal to 1 and less than or equal to the total iteration step number; when n=1, t n-1 ＝t 0 ，t 0 Indicating the initial time, t, of the initial state of the projectile n Representing the current time after n iteration time steps;
(Q2) surface wind force according to throwerAnd->Calculating the acceleration of throwing objects respectively And->Wherein (1)>And->Acceleration of the projectile in the 3 dimensions x, y and z, respectively, +.>Overturning acceleration for the throwing object;
(Q3) by t n-1 Moment throwing accelerationAnd t n-1 Speed of the moment thrower in 3 dimensions x, y and z of space +.>And->Respectively calculate t n Moment throws in 3 dimensions x, y and z of spaceSpeed of the upper part->And->And (3) displacement->And->The turning speed of the throwing object>And (3) displacement->And then get t n Geometrical center point coordinates on the current iteration time step at the moment;
(Q4) when iteration starts, sequentially iterating according to a set time step delta t, repeatedly executing the steps (Q1) to (Q3) each iteration until the total iteration step number is reached, recording the geometric center point coordinates of the thrower on each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the thrower in the initial state and the geometric center point coordinates on each iteration time step tag Saving to a database.
Further, for the block thrower, a sphere, a spherical polyhedron and a cube thrower are randomly generated, the diameters of the sphere and the spherical polyhedron are randomly selected within the range of [10cm,50cm ], and the side length of the cube is randomly selected within the range of [10cm,30cm ];
for the rod-shaped throwing object, a cylinder and a polygonal prism throwing object are randomly generated, the length is randomly selected within the range of [30cm,300cm ], and the section diameter is randomly selected within the range of [5cm,20cm ];
for plate-like casts, rectangular, circular, and polygonal plate-like casts are randomly generated, the area is randomly selected within [0.1m2,1m2 ];
wherein, the mass of the throwing matter is randomly selected, m is epsilon [50,5000], and the mass unit is g.
Further, the simulated wind field generation steps are as follows:
(D1) Setting a calculation space for simulating a wind field and setting boundary layer conditions of the simulated wind field;
(D2) According to the boundary layer conditions of the set simulated wind field, calculating the t of each calculation point n Time of day, coordinate x i The wind velocity component v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) Finally get at t n The combined wind velocity V (x) at each calculation point in time i ,t n )，Wherein t is n ＝t 0 +nDeltat, n is greater than or equal to 1 and less than or equal to the total iteration step number; when n=1, t n-1 ＝t 0 ，t 0 Indicating the initial time, t, of the initial state of the projectile n Representing the current time after n iteration time steps;
(D3) Repeatedly executing the steps (D1) to (D2) in sequence according to the set time step delta t until the total iteration step number is reached, and stopping calculation to obtain a numerical model V of the simulated wind field corresponding to the boundary layer condition of the simulated wind field X,T . Optionally, the wind velocity component v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) By calculating the formula for wind speed:resolution to obtain, after resolution, v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) The formula of (2) is:
for the v direction, R (Δt) =exp (- Δt/T) Lv' )，v”(x i ,t n-1 )＝σ v′ [1-R 2 (Δt)] 1/2 ξ；
For the u direction, R (Δt) =exp (- Δt/T) Lu' )，u”(x i ,t n-1 )＝σ u′ [1-R 2 (Δt)] 1/2 ξ；
For the w direction, R (Δt) =exp (- Δt/T) Lw' )，w”(x i ,t n-1 )＝σ w′ [1-R 2 (Δt)] 1/2 ξ；
Wherein,and->At t n Wind speed average component>Wind speed average component in v, u and w directions; v "(x) i ,t n-1 )、u”(x i ,t n-1 ) And w "(x) i ,t n-1 ) At t n-1 Time wind speed pulse component V "(x) i ,t n-1 ) Wind speed pulse components in the v, u and w directions; ζ is 1 group of random numbers conforming to standard normal distribution; sigma is calculated according to the atmospheric state of the boundary layer v' 、σ u' Sum sigma w' The sigma is respectively valued in the v, u and w directions; t (T) L At t n Time wind speed V (x) i ,t n ) Is the pulsating component V' (x) i ,t n ) Lagrangian time scale, T Lv' 、T Lu' And T Lw' Respectively T L Lagrangian time scales in the v, u and w directions; Δt is the calculated time step; r (Δt) is an exponentially related coefficient.
Further, the boundary layer conditions include an unstable boundary layer, a neutral boundary layerAnd stabilizing the boundary layer, respectively obtaining a numerical model V of the simulated wind field corresponding to the unstable boundary layer, the neutral boundary layer and the stable boundary layer through the steps (D1) to (D3) X,T The method comprises the steps of carrying out a first treatment on the surface of the Alternatively, for an unstable boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝σ v′ ＝u * (12+0.5z i /|L|) 1/3 ；
T Lu′ ＝T Lv′ ＝0.15z i /σ u′ ；
for neutral boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝2u * exp(-3fz/u * )；
σ v′ ＝σ w′ ＝1.3u * exp(-2fz/u * )；
for stable boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝2u * (1-z/z i )；
σ v′ ＝σ w′ ＝1.3u * (1-z/z i )；
wherein z is the calculated point height, u * Is the friction speed, w * Is the characteristic velocity of convection. zi is the height of the mixed layer, L is the length of Moning-Obuhuff, and the value is taken according to the boundary condition. f is the length of Coriolis force, and according to the latitude distribution of China, 7.29×10 is taken -5 。
Further, the training set constructing step includes:
(S421) simulating a camera position in a simulation environment according to the position of the camera layout in the actual use environment, wherein the included angle between the optical axis of the simulation camera and the simulation floor is equal to the included angle between the optical axis of the camera and the floor in the actual environment;
(S422) simulating the motion trail of the thrower to obtain trail images of various throwers in the simulated camera under various throwing object forms and various wind field conditions;
(S423) simulating the motion trail of the non-thrower, randomly generating a transverse flying target, a longitudinal ascending target, a transverse reciprocating moving target, a longitudinal reciprocating moving target and a random flashing target, recording the motion trail of the target, and obtaining motion trail images of various non-thrower moving targets under various types of non-throwers and various wind field conditions;
(S424) mixing the thrower motion trajectory image sample with the non-thrower motion trajectory image sample in a ratio of 1:1;
(S425) mixing the mixed sample sets by the training sample number: the ratio of test sample number=7:3 is sampled to construct a training data set and a test data set.
Further, both the thrower and the non-thrower obtain motion trajectories for various types of thrower or non-thrower and under various wind field conditions by:
(F1) Let the coordinates of the camera be (x c ,y c ,z c ) The motion trail of the non-throwing object or throwing object is obtained through simulation, and the coordinates of the non-throwing object or throwing object at the moment t are set as (x) tag ,y tag ,z tag ) The coordinates (x t ' ag ,y t ' ag ，z t ' ag )，x t ' ag And y t ' ag The calculation formulas of (a) are respectively as follows:
wherein f is the focal length of the camera lens;
(F2) Repeating the step (F1) until all non-thrower or thrower position points on the 1 track are imaged in the analog camera and form a track image, and then executing the step (F3);
(F3) Repeating the step (F1) and the step (F2) to obtain track images of various non-throwers or throwers in the simulation camera under various wind field conditions of various thrower forms; optionally, the neural network discrimination model is obtained through convolutional neural network training; optionally, the neural network discrimination model is divided into an input layer, a feature extraction layer and a discrimination output layer; the input layer receives an input motion trail coordinate sequence and feeds the motion trail coordinate sequence to the feature extraction layer; the feature extraction layer is formed by stacking 3 convolution pooling layers, namely each convolution layer is connected with 1-dimensional pooling layer; the judging output layer consists of 1 full-connection layer and 1 softmax layer, and the softmax carries out two-classification judgment, namely whether the input track is a high-altitude projectile motion track or not; optionally, the length of the input motion track coordinate sequence is 125, the sequence exceeding 125 reserves data from the 1 st bit to the 125 th bit, and the sequence less than 125 complements 0 at the end of the sequence; optionally, in each convolution pooling layer, the convolution kernel size of the convolution layer is 1×5, and the step size of the convolution kernel is 1; the pooling size is 1×5, and the step size is 2.
Compared with the prior art, the invention has the following advantages:
according to the high-altitude projectile discrimination system, high-altitude projectile target motion detection and recognition are converted into a sequence classification problem, so that the complexity of a detection system is reduced while the detection performance is improved; the problem that the track sample is difficult to obtain is also solved; the system is simple, the calculation is greatly reduced, the cost is reduced, the changeable wind field conditions in the actual scene can be simulated, the motion trail of the thrower with different aerodynamic shapes under different wind fields can be obtained, the model can obtain effective training sample data in the training process, and the distinguishing performance of the model is improved.
Drawings
FIG. 1 is a flow chart of a method of determining a high altitude projectile of the present invention;
FIG. 2 is a schematic diagram of the high altitude projectile discrimination of the present invention.
In the figure:
1-a video image acquisition module; 2-a moving object detection module; 3-an image processing module; 41-a model iteration module; 4-a high-altitude parabolic judgment module; 5-network transmission module.
Detailed Description
The following describes the embodiments of the present invention further with reference to the drawings.
Referring to fig. 1, the embodiment discloses a method for discriminating a high altitude projectile, which comprises the following steps:
(S1) a camera capturing a building floor;
(S2) detecting and extracting a moving object in a camera picture using a frame difference method;
(S3) extracting a motion trail coordinate sequence of the moving object;
and (S4) inputting the motion trail coordinate sequence of the moving object into a pre-trained neural network discrimination model, and discriminating whether the moving object belongs to a high-altitude parabolic object or not.
In this embodiment, if the moving object belongs to a high-altitude parabolic object, the moving feature of the moving object is added to a training set of the neural network discriminant model, and model iteration is performed.
In this embodiment, the pre-trained neural network discrimination model is obtained by:
(S41) obtaining motion trail coordinate sequence samples of throwers with different aerodynamic shapes under different wind field simulation conditions through computer simulation;
and (S42) training the discrimination neural network by taking the sample set of the projectile motion trail coordinate sequences as a training set to obtain a trained neural network discrimination model.
In this embodiment, the computer simulation is used to obtain the motion track samples of the thrower with different aerodynamic shapes under different wind field conditions, and specifically includes the following steps:
(S411) randomly generating 1 throwing object through computer simulation, wherein the shape of the throwing object is randomly determined in 3 shapes of a block, a rod and a plate;
(S412) initializing the motion state of the projectile and setting the iteration time step delta t and the total iteration step number for calculating the geometric center point coordinates of the projectile, calculating and recording the geometric center point coordinates on each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the projectile in the initial state and the geometric center point coordinates on each iteration time step tag Saving to a database;
(S413) obtaining a motion trail sample P by arranging throwers with different aerodynamic shapes to move under different wind field conditions tag1 、P tag2 ···P tagm Generate 1 track sample set p= { P tag1 ,P tag2 ,...,P tagm }。
In this embodiment, the initializing the projectile motion state specifically includes the following: randomly setting initial velocity vectors for a projectileThe angle is arbitrarily selected, and-> Is in units of m/s; rotational speed of projectile
The simulation time is synchronous with the wind field simulation time;
the starting point of the projectile motion is the center of the upper edge of the simulated wind field. Wherein, the windward angle of the plate-shaped throwing objectRandomly taking value, and (E) is (are) added>Plate-shaped throwing object windward angle +.>Is used for inputting Fluent calculation.
In this embodiment, the geometric center point coordinates on each iteration time step are calculated and recorded, and the geometric center point coordinates of the projectile in the initial state and the geometric center point coordinates on each iteration time step are formed into 1 coordinate sequence P tag Saving to a database; the method specifically comprises the following steps:
(Q1) obtaining at t from the simulated wind field n-1 Wind speed vector at moment projectile motion positionCalculating t by combining with the morphological characteristic parameters of the throwing object n-1 Surface wind force of moment throwing object +.>And->Wherein,and->Respectively t n-1 Surface wind force of moment thrower in 3 dimensions x, y and z direction of space, +.>At t n-1 The throwing object is subjected to overturning acting force at moment; wherein t is n ＝t 0 +nDeltat, n is greater than or equal to 1 and less than or equal to the total iteration step number; when n=1, t n-1 ＝t 0 ，t 0 Indicating the initial time, t, of the initial state of the projectile n Representing the current time after n iteration time steps;
(Q2) surface wind force according to throwerAnd->Calculating the acceleration of throwing objects respectively And->Wherein (1)>And->Acceleration of the projectile in the 3 dimensions x, y and z, respectively, +.>Overturning acceleration for the throwing object;
(Q3) by t n-1 Moment throwing accelerationAnd t n-1 Throwing objects in 3 spaces at the momentSpeed in dimensions x, y and z +.>And->Respectively calculate t n Speed of the moment thrower in 3 dimensions x, y and z of space +.>And->And (3) displacement->And->The turning speed of the throwing object>And (3) displacement->And then get t n Geometrical center point coordinates on the current iteration time step at the moment;
(Q4) when iteration starts, sequentially iterating according to a set time step delta t, repeatedly executing the steps (Q1) to (Q3) in each iteration until the total iteration step number is reached, recording the geometric center point coordinates of the thrower on each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the thrower in the initial state and the geometric center point coordinates on each iteration time step tag Saving to a database; wherein, the throwing object lands or flies out of the simulated wind field boundary when the preset iteration step number is reached; throwing object falls to the ground, i.e. V tag,y,t ＝0。
In the embodiment, for the block thrower, a sphere, a spherical polyhedron and a cube thrower are randomly generated, the diameters of the sphere and the spherical polyhedron are randomly selected within the range of [10cm,50cm ], and the side length of the cube is randomly selected within the range of [10cm,30cm ];
for the rod-shaped throwing object, a cylinder and a polygonal prism throwing object are randomly generated, the length is randomly selected within the range of [30cm,300cm ], and the section diameter is randomly selected within the range of [5cm,20cm ];
for plate-like casts, rectangular, circular, and polygonal plate-like casts are randomly generated, the area is randomly selected within [0.1m2,1m2 ];
wherein, the mass of the throwing matter is randomly selected, m is epsilon [50,5000], and the mass unit is g.
In this embodiment, the surface wind forceAnd->The calculation formulas of (a) are respectively as follows:
wherein ρ is a For air density, A is the maximum reference area of the projectile,at t n-1 Moment projectile motion velocity vector, +.>Is a projectile rotation speed vector; />At t n-1 A wind speed vector at the motion position of the throwing object at the moment;and->Respectively t n-1 Surface wind coefficients of the moment thrower in the x, y and z directions of the 3 dimensions of space,at t n-1 The turning acting force applied by the throwing object at any moment corresponds to the surface wind power coefficient. In this embodiment, each iteration, the following data needs to be updated: position coordinates of the projectile; speed vector of projectile +.>Wind speed vector +.>Thrower at t n-1 Wind power coefficient of surface of throwing object at moment +.>And-> And->Is according to t n-1 Wind speed vector +.>And combining with projectile morphological characteristic parameters, adopting a RANS time-sharing stress model and CFD calculation software Fluent to calculate. t is t n Surface wind force of moment throwing object +.>And->And calculating by adopting a RANS time-sharing stress model and CFD calculation software Fluent.
In this embodiment, the projectile acceleration is calculatedAnd->The formulas of (a) are respectively as follows:
calculating the velocity of the projectile in 3 dimensions x, y and z of spaceAnd->And displacement And->The turning speed of the throwing object>Sum bitMove->The formulas of (a) are respectively as follows:
in this embodiment, the simulated wind field generation steps are as follows:
(D1) Setting a calculation space for simulating a wind field and setting boundary layer conditions of the simulated wind field;
(D2) According to the boundary layer conditions of the set simulated wind field, calculating the t of each calculation point n Time of day, coordinate x i The wind velocity component v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) Finally get at t n The combined wind velocity V (x) at each calculation point in time i ,t n )，Wherein t is n ＝t 0 +nDeltat, n is greater than or equal to 1 and less than or equal to the total iteration step number; when n=1, t n-1 ＝t 0 ，t 0 Indicating the initial time, t, of the initial state of the projectile n Representing the current time after n iteration time steps;
(D3) Repeatedly executing the steps (D1) to (D2) in sequence according to the set time step delta t until the total iteration step number is reached, and stopping calculation to obtain a numerical model V of the simulated wind field corresponding to the boundary layer condition of the simulated wind field X,T 。
In this embodiment, the size of the computation space is 120m high, 300m wide, and 300m deep; in some embodiments, the computation space may also be of other sizes, not limited herein. Dividing each calculation point: and determining the size of a wind field calculation point grid according to the calculation capability, wherein the size of the simulation grid is selectable from 1mX1m to 5mX5 m. The wind field simulation of the unstable boundary layer selects small-scale and denser grids, and the simulation of the neutral and stable boundary layers selects large-scale and sparse grids. And confirming that the wind field simulation time is the same as the total iteration step number under the projectile motion state.
In the present embodiment of the present invention, in the present embodiment,at [0m/s,5m/s]And randomly setting the time.
In the present embodiment, the wind speed component v (x) in the x, y, z direction can be deduced from the formulas (1) to (4) i ,t n )、u(x i ,t n ) And w (x) i ,t n ) The specific calculation process is as follows:
simulating the wind speed of each calculation point in the wind field at the moment t:
equation (1) can be split into 3 velocity components to represent:
the formula for generating the synthetic wind speed by the 3 velocity components is as follows:
the airflow vortex process is taken as 1 continuous process, under the condition of conforming to Markov assumption, the point wind speed is calculated by 1 time step:
V′(x i ,t n )＝R(Δt)V′(x i ,t n-1 )+V″(x i ,t n-1 ) (2)；
equation (2) can be split into 3 velocity components to represent:
R(Δt)＝exp(-Δt/T L ) (3)；
V″(x i ,t n )＝σ[1-R 2 (Δt)] 1/2 ξ (4)；
in summary, the wind speed calculation formula is obtained:the wind velocity component v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) Split v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) The formula of (2) is:
for the v direction, R (Δt) =exp (- Δt/T) Lv' )，v”(x i ,t n-1 )＝σ v′ [1-R 2 (Δt)] 1/2 ξ；
For the u direction, R (Δt) =exp (- Δt/T) Lu' )，u”(x i ,t n-1 )＝σ u′ [1-R 2 (Δt)] 1/2 ξ；
For the w direction, R (Δt) =exp (- Δt/T) Lw' )，w”(x i ,t n-1 )＝σ w′ [1-R 2 (Δt)] 1/2 ξ；
Wherein,and->At t n Wind speed average component>Wind speed average component in v, u and w directions; v "(x) i ,t n-1 )、u”(x i ,t n-1 ) And w "(x) i ,t n-1 ) At t n-1 Time wind speed pulse component V "(x) i ,t n-1 ) Wind speed pulse components in the v, u and w directions; ζ is 1 group of random numbers conforming to standard normal distribution; sigma is calculated according to the atmospheric state of the boundary layer v' 、σ u' Sum sigma w' The sigma is respectively valued in the v, u and w directions; t (T) L At t n Time wind speed V (x) i ,t n ) Is the pulsating component V' (x) i ,t n ) Lagrangian time scale, T Lv' 、T Lu' And T Lw' Respectively T L Lagrangian time scales in the v, u and w directions; Δt is the calculated time step; r (Δt) is in the form of an indexCorrelation coefficient. The average component is given here in principle by 1 diagnostic wind field software, but since the parabolic time is short, the average component of the wind speed does not change much, so in this solution, in each simulation calculation the average component of the wind speed is set and kept constant, thus->Andthe value of (2) is 1,10]Within the section (I)>The value of (2) is [0,3 ]]Within the interval. The time step is taken according to actual needs, which is not limited herein, and in this embodiment, the time step takes 1s. Computationally, the middle right second term of the above 3 equations requires the pair v' (x) i ,t n )、u'(x i ,t n )、w'(x i ,t n ) And (5) performing iteration. When described using the three formulas above. In the actual calculation, the initial value of V' is 0 or 1 small value.
In this embodiment, the boundary layer conditions include an unstable boundary layer, a neutral boundary layer, and a stable boundary layer, and the numerical model V of the simulated wind field corresponding to the unstable boundary layer, the neutral boundary layer, and the stable boundary layer is obtained through steps (D1) to (D3), respectively X,T 。
In the present embodiment, for an unstable boundary layer, σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝σ v′ ＝u * (12+0.5z i /|L|) 1/3 ；
T Lu′ ＝T Lv′ ＝0.15z i /σ u′ ；
for neutral boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝2u * exp(-3fz/u * )；
σ v′ ＝σ w′ ＝1.3u * exp(-2fz/u * )；
for stable boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝2u * (1-z/z i )；
σ v′ ＝σ w′ ＝1.3u * (1-z/z i )；
wherein z is the calculated point height, u * Is the friction speed, w * Is the characteristic velocity of convection. z i For the mixed layer height, L is the Monin-Obukhov length (Monin-Obukhov length), and is valued according to the boundary condition. f is the length of Coriolis force, and according to the latitude distribution of China, 7.29×10 is taken -5 。
In this embodiment, the training set constructing step includes:
(S421) simulating a camera position in a simulation environment according to the position of the camera layout in the actual use environment, wherein the included angle between the optical axis of the simulation camera and the simulation floor is equal to the included angle between the optical axis of the camera and the floor in the actual environment;
(S422) simulating the motion trail of the thrower to obtain trail images of various throwers in the simulated camera under various throwing object forms and various wind field conditions;
(S423) simulating the motion trail of the non-thrower, randomly generating a transverse flying target, a longitudinal ascending target, a transverse reciprocating moving target, a longitudinal reciprocating moving target and a random flashing target, recording the motion trail of the target, and obtaining motion trail images of various non-thrower moving targets under various types of non-throwers and various wind field conditions;
(S424) mixing the thrower motion trajectory image sample with the non-thrower motion trajectory image sample in a ratio of 1:1;
(S425) mixing the mixed sample sets by the training sample number: the ratio of test sample number=7:3 is sampled to construct a training data set and a test data set.
In this embodiment, both the thrower and the non-thrower obtain motion trajectories for various types of thrower or non-thrower and under various wind field conditions by:
(F1) Let the coordinates of the camera be (x c ,y c ,z c ) The motion trail of the non-throwing object or throwing object is obtained through simulation, and the coordinates of the non-throwing object or throwing object at the moment t are set as (x) tag ,y tag ,z tag ) The coordinates (x t ' ag ,y t ' ag ，z t ' ag )，x t ' ag And y t ' ag The calculation formulas of (a) are respectively as follows:
wherein f is the focal length of the camera lens; with the pinhole imaging model, no calculation is made on the data on the Z axis in the pinhole imaging model.
(F2) Repeating the step (F1) until all non-thrower or thrower position points on the 1 track are imaged in the analog camera and form a track image, and then executing the step (F3);
(F3) And (3) repeating the step (F1) and the step (F2) to obtain track images of different non-throws or throws in the simulation camera under various wind field conditions of various throws.
In this embodiment, the neural network discrimination model is obtained by convolutional neural network training. The neural network discrimination model is divided into an input layer, a feature extraction layer and a discrimination output layer; the input layer receives the input coordinate sequence track coordinate sequence and feeds the coordinate sequence to the feature extraction layer; the feature extraction layer is formed by stacking 3 convolution pooling layers, namely each convolution layer is connected with 1-dimensional pooling layer; the judging output layer consists of 1 full-connection layer and 1 softmax layer, and the softmax carries out two-class judgment, namely whether the input track is the high-altitude projectile motion track.
In this embodiment, the length of the input motion trajectory coordinate sequence is 125, the sequence exceeding 125 retains the data from the 1 st bit to the 125 th bit, and the sequence less than 125 complements 0 at the end of the sequence; optionally, in each convolution pooling layer, the convolution kernel size of the convolution layer is 1×5, and the step size of the convolution kernel is 1; the pooling size is 1×5, and the step size is 2.
In this embodiment, the neural network uses Binary Cross Entropy as the loss function:
wherein:
y is the category of the track sample;
p (yi) is the sample class classification probability given by the neural network;
n is the number of samples. The motion trail of the throwing object in the wind field is discretized into 1 trail coordinate sequence after being shot by a camera. The sequence is input into 1 discrimination network to realize the classification discrimination of the track.
Referring to fig. 2, the embodiment also discloses a high altitude projectile discrimination system, which executes the above high altitude projectile discrimination method, comprising:
the video image acquisition module 1 is used for shooting building floors;
a network transmission module 5 for transmitting a picture of a floor of a building;
a moving object detection module 2 for receiving the picture of the building floor from the network transmission module 5 and detecting and extracting a moving object in the photographed picture; the moving target detection module 2 is connected with the video image acquisition module 1;
the image processing module 3 is used for extracting a motion trail coordinate sequence of the moving object and inputting the motion trail coordinate sequence into the high-altitude parabolic distinguishing module 4; the image processing module 3 is connected with the moving object detection module 2;
the high-altitude parabolic judgment module 4 is used for judging whether the input moving object belongs to a high-altitude parabolic object, the high-altitude parabolic judgment module 4 is connected with the image processing module 3, and a neural network judgment model used for judging whether the input moving object belongs to the high-altitude parabolic object is built in the high-altitude parabolic judgment module 4.
In this embodiment, the high altitude parabolic discrimination module 4 includes a model iteration module 41 for adding the motion feature of the moving object belonging to the high altitude parabolic object to the training set of the neural network discrimination model to perform model iteration.
According to the high-altitude projectile discrimination system, high-altitude projectile target motion detection and recognition are converted into a sequence classification problem, so that the complexity of a detection system is reduced while the detection performance is improved; the problem that the track sample is difficult to obtain is also solved; the system is simple, the calculation is greatly reduced, the cost is reduced, the changeable wind field conditions in the actual scene can be simulated, the motion trail of the thrower with different aerodynamic shapes under different wind fields can be obtained, the model can obtain effective training sample data in the training process, and the distinguishing performance of the model is improved.
The foregoing describes in detail preferred embodiments of the present invention. It should be understood that numerous modifications and variations can be made in accordance with the concepts of the invention by one of ordinary skill in the art without undue burden. Therefore, all technical solutions which can be obtained by logic analysis, reasoning or limited experiments based on the prior art by the person skilled in the art according to the inventive concept shall be within the scope of protection defined by the claims.

Claims (8)

1. The method for distinguishing the high-altitude thrower is characterized by comprising the following steps of:
(S1) a camera capturing a building floor;
(S2) detecting and extracting a moving object in a camera picture using a frame difference method;
(S3) extracting a motion trail coordinate sequence of the moving object;
(S4) inputting a motion trail coordinate sequence of the moving target into a pre-trained neural network discrimination model to carry out classification discrimination, and discriminating whether the moving target belongs to a high-altitude parabolic object or not;
the pre-trained neural network discrimination model is obtained through the following steps:
(S41) obtaining motion trail coordinate sequence samples of throwers with different aerodynamic shapes under different wind field simulation conditions through computer simulation;
(S42) training the discrimination neural network by taking the sample set of the projectile motion trail coordinate sequence as a training set to obtain a trained neural network discrimination model;
the method comprises the following steps of obtaining motion track samples of throwers with different aerodynamic shapes under different wind field conditions through computer simulation:
(S411) randomly generating 1 throwing object through computer simulation, wherein the shape of the throwing object is randomly determined in 3 shapes of a block, a rod and a plate;
(S412) initializing the motion state of the projectile and setting the iteration time step Deltat and the total iteration step number for calculating the geometric center point coordinates of the projectile, calculating and recording the geometric center point coordinates on each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the projectile in the initial state and the geometric center point coordinates on each iteration time step tag Saving to a database;
(S413) obtaining a motion trail sample P by arranging throwers with different aerodynamic shapes to move under different wind field conditions tag1 、P tag2 ···P tagm Generate 1 track sample set p= { P tag1 ,P tag2 ,...,P tagm -a }; the initialization projectile motion state specifically comprises the following contents: randomly setting initial velocity vectors for a projectileThe angle of the two-dimensional optical fiber is arbitrarily selected, is in units of m/s; projectile rotation speed +.>The simulation time is synchronous with the wind field simulation time; the initial point of the projectile motion is the center of the upper edge of the simulated wind field; wherein, the windward angle of the plate-shaped throwing object is +.>The value is taken at random and the method comprises the steps of,
2. the method for determining a high altitude projectile of claim 1, whereinThe method comprises the steps of calculating and recording the geometric center point coordinates of each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the throwing object in an initial state and the geometric center point coordinates of each iteration time step tag Saving to a database; the method specifically comprises the following steps:
(Q1) obtaining at t from the simulated wind field n-1 Wind speed vector at moment projectile motion positionCalculating t by combining with the morphological characteristic parameters of the throwing object n-1 Surface wind force of moment throwing object +.>And->Wherein,and->Respectively t n-1 Surface wind force of moment thrower in 3 dimensions x, y and z direction of space, +.>At t n-1 The throwing object is subjected to overturning acting force at moment; wherein t is n ＝t 0 +nDeltat, n is greater than or equal to 1 and less than or equal to the total iteration step number; when n=1, t n-1 ＝t 0 ，t 0 Indicating the initial time, t, of the initial state of the projectile n Representing the current time after n iteration time steps;
(Q2) surface wind force according to throwerAnd->Calculating the acceleration of the throwing object respectively> And->Wherein (1)>And->Acceleration of the projectile in the 3 dimensions x, y and z, respectively, +.>Overturning acceleration for the throwing object;
(Q3) by t n-1 Moment throwing accelerationAnd t n-1 Speed of the moment thrower in 3 dimensions x, y and z of space +.>And->Respectively calculate t n Speed of the moment thrower in 3 dimensions x, y and z of space +.>And->And (3) displacement->And->The turning speed of the throwing object>And (3) displacement->And then get t n Geometrical center point coordinates on the current iteration time step at the moment;
(Q4) when iteration starts, sequentially iterating according to a set time step delta t, repeatedly executing the steps (Q1) to (Q3) in each iteration until the total iteration step number is reached, recording the geometric center point coordinates of the thrower on each iteration time step, and forming 1 coordinate sequence P by the geometric center point coordinates of the thrower in the initial state and the geometric center point coordinates on each iteration time step tag Saving to a database.
3. The method for determining a high altitude slinger according to claim 1 or 2, wherein,
for the block thrower, randomly generating a sphere, a spherical polyhedron and a cube thrower, wherein the diameters of the sphere and the spherical polyhedron are randomly selected within the range of [10cm,50cm ], and the side length of the cube is randomly selected within the range of [10cm,30cm ];
for the rod-shaped throwing object, a cylinder and a polygonal prism throwing object are randomly generated, the length is randomly selected within the range of [30cm,300cm ], and the section diameter is randomly selected within the range of [5cm,20cm ];
for plate-like casts, rectangular, circular, and polygonal plate-like casts are randomly generated, the area is randomly selected within [0.1m2,1m2 ];
wherein, the mass of the throwing matter is randomly selected, m is epsilon [50,5000], and the mass unit is g.
4. A method of determining a high altitude projectile in accordance with claim 3, wherein said simulated wind field generation step comprises:
(D1) Setting a calculation space for simulating a wind field and setting boundary layer conditions of the simulated wind field;
(D2) According to the boundary layer conditions of the set simulated wind field, calculating the t of each calculation point n Time of day, coordinate x i The wind velocity component v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) Finally get at t n The combined wind velocity V (x) at each calculation point in time i ,t n )，Wherein t is n ＝t 0 +nDeltat, n is greater than or equal to 1 and less than or equal to the total iteration step number; when n=1, t n-1 ＝t 0 ，t 0 Indicating the initial time, t, of the initial state of the projectile n Representing the current time after n iteration time steps;
(D3) Repeatedly executing the steps (D1) to (D2) in sequence according to the set time step delta t until the total iteration step number is reached, and stopping calculation to obtain a numerical model V of the simulated wind field corresponding to the boundary layer condition of the simulated wind field X,T ；
Wind velocity component v (x) i ,t n )、u(x i ,t n ) And w (x) i ,t n ) By calculating the formula for wind speed:resolution to obtain, after resolution, v (x i ,t n )、u(x i ,t n ) And w (x) i ,t n ) The formulas of (a) are respectively as follows:
for the v direction, R (Δt) =exp (- Δt/T) Lv' )，v”(x i ,t n-1 )＝σ v′ [1-R 2 (Δt)] 1/2 ξ；
For the u direction, R (Δt) =exp (- Δt/T) Lu' )，u”(x i ,t n-1 )＝σ u′ [1-R 2 (Δt)] 1/2 ξ；
For the w direction, R (Δt) =exp (- Δt/T) Lw' )，w”(x i ,t n-1 )＝σ w′ [1-R 2 (△t)] 1/2 ξ；
Wherein,and->At t n Wind speed average component>Wind speed average component in v, u and w directions; v "(x) i ,t n-1 )、u”(x i ,t n-1 ) And w "(x) i ,t n-1 ) At t n-1 Time wind speed pulse component V "(x) i ,t n-1 ) Wind speed pulse components in the v, u and w directions; ζ is 1 group of random numbers conforming to standard normal distribution; sigma is calculated according to the atmospheric state of the boundary layer v' 、σ u' Sum sigma w' The sigma is respectively valued in the v, u and w directions; t (T) L At t n Time wind speed V (x) i ,t n ) Is the pulsating component V' (x) i ,t n ) Lagrangian time scale, T Lv' 、T Lu' And T Lw' Respectively T L Lagrangian time scales in the v, u and w directions; Δt is the calculated time step; r (Δt) is an exponentially related coefficient.
5. The method according to claim 4, wherein the boundary layer conditions include an unstable boundary layer, a neutral boundary layer and a stable boundary layer, and the numerical model V of the simulated wind field corresponding to the unstable boundary layer, the neutral boundary layer and the stable boundary layer is obtained through the steps (D1) to (D3), respectively X,T The method comprises the steps of carrying out a first treatment on the surface of the For unstable boundary layers, σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝σ v′ ＝u * (12+0.5z i /|L|) 1/3 ；
T Lu′ ＝T Lv′ ＝0.15z i /σ u′ ；
for neutral boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝2u * exp(-3fz/u * )；
σ v′ ＝σ w′ ＝1.3u * exp(-2fz/u * )；
for stable boundary layer, then σ u' 、σ v' 、σ w' 、T Lu' 、T Lv' And T Lw' The formulas of (a) are respectively as follows:
σ u′ ＝2u * (1-z/z i )；
σ v′ ＝σ w′ ＝1.3u * (1-z/z i )；
wherein z is the calculated point height, u * Is the friction speed, w * For convection characteristic velocity, z i For the height of the mixed layer, L is the length of Morning-Obuhuff, the value is taken according to boundary conditions, f is the length of Coriolis force, and 7.29 multiplied by 10 is taken according to the latitude distribution of China -5 。
6. The method for determining a high altitude thrower of claim 1 or 5, wherein the training set construction step includes:
(S421) simulating a camera position in a simulation environment according to the position of the camera layout in the actual use environment, wherein the included angle between the optical axis of the simulation camera and the simulation floor is equal to the included angle between the optical axis of the camera and the floor in the actual environment;
(S422) simulating the motion trail of the thrower to obtain trail images of various throwers in the simulated camera under various throwing object forms and various wind field conditions;
(S423) simulating the motion trail of the non-thrower, randomly generating a transverse flying target, a longitudinal ascending target, a transverse reciprocating moving target, a longitudinal reciprocating moving target and a random flashing target, recording the motion trail of the target, and obtaining motion trail images of various non-thrower moving targets under various types of non-throwers and various wind field conditions;
(S424) mixing the thrower motion trajectory image sample with the non-thrower motion trajectory image sample in a ratio of 1:1;
(S425) mixing the mixed sample sets by the training sample number: the ratio of test sample number=7:3 is sampled to construct a training data set and a test data set.
7. The method for distinguishing high-altitude throws according to claim 6, wherein the thrower and the non-thrower each obtain motion trajectories of various types of thrower and non-thrower and under various wind field conditions by:
(F1) Let the coordinates of the camera be (x c ,y c ,z c ) The motion trail of the non-throwing object or throwing object is obtained through simulation, and the coordinates of the non-throwing object or throwing object at the moment t are set as (x) tag ,y tag ,z tag ) The coordinates (x 'of the non-thrower or thrower locus point on the camera imaging frame' tag ,y′ tag ，z′ tag )，x′ tag And y' tag The calculation formulas of (a) are respectively as follows:
wherein f is the focal length of the camera lens;
(F2) Repeating the step (F1) until all non-thrower or thrower position points on the 1 track are imaged in the analog camera and form a track image, and then executing the step (F3);
(F3) Repeating the step (F1) and the step (F2) to obtain track images of various non-throwers or throwers in the simulation camera under various wind field conditions of various thrower forms; the neural network discrimination model is obtained through convolutional neural network training; the neural network discrimination model is divided into an input layer, a feature extraction layer and a discrimination output layer; the input layer receives an input motion trail coordinate sequence and feeds the motion trail coordinate sequence to the feature extraction layer; the feature extraction layer is formed by stacking 3 convolution pooling layers, namely each convolution layer is connected with 1-dimensional pooling layer; the judging output layer consists of 1 full-connection layer and 1 softmax layer, and the softmax carries out two-classification judgment, namely whether the input track is a high-altitude projectile motion track or not; the length of the input motion track coordinate sequence is 125, the sequence exceeding 125 reserves the data from the 1 st bit to the 125 th bit, and the sequence less than 125 complements 0 at the end of the sequence; in each convolution pooling layer, the convolution kernel size of the convolution layer is 1 multiplied by 5, and the step length of the convolution kernel is 1; the pooling size is 1×5, and the step size is 2.
8. The method according to claim 1, 2, 5 or 7, wherein if the moving object belongs to a high-altitude parabolic object, the moving feature of the moving object is added to a training set of a neural network discrimination model, and model iteration is performed.
CN202011224115.0A 2020-11-05 2020-11-05 High-altitude throwing object distinguishing method Active CN112329627B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011224115.0A CN112329627B (en) 2020-11-05 2020-11-05 High-altitude throwing object distinguishing method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011224115.0A CN112329627B (en) 2020-11-05 2020-11-05 High-altitude throwing object distinguishing method

Publications (2)

Publication Number Publication Date
CN112329627A CN112329627A (en) 2021-02-05
CN112329627B true CN112329627B (en) 2024-02-09

Family

ID=74315896

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011224115.0A Active CN112329627B (en) 2020-11-05 2020-11-05 High-altitude throwing object distinguishing method

Country Status (1)

Country Link
CN (1) CN112329627B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114241012B (en) * 2022-02-23 2023-04-07 深圳市研超科技有限公司 High-altitude parabolic determination method and device
CN116597340B (en) * 2023-04-12 2023-10-10 深圳市明源云科技有限公司 High altitude parabolic position prediction method, electronic device and readable storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104601939A (en) * 2013-10-31 2015-05-06 深圳市东宝嘉科技有限公司 Method and system for monitoring thrown object and eliminating risks of thrown object
CN110675592A (en) * 2019-08-16 2020-01-10 重庆特斯联智慧科技股份有限公司 High-altitude parabolic early warning protection system based on target identification and control method
CN111079663A (en) * 2019-12-19 2020-04-28 深圳云天励飞技术有限公司 High-altitude parabolic monitoring method and device, electronic equipment and storage medium
CN111553256A (en) * 2020-04-26 2020-08-18 上海天诚比集科技有限公司 High-altitude parabolic early warning identification method based on object track identification
CN111639578A (en) * 2020-05-25 2020-09-08 上海中通吉网络技术有限公司 Method, device, equipment and storage medium for intelligently identifying illegal parabola
CN111723654A (en) * 2020-05-12 2020-09-29 中国电子***技术有限公司 High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104601939A (en) * 2013-10-31 2015-05-06 深圳市东宝嘉科技有限公司 Method and system for monitoring thrown object and eliminating risks of thrown object
CN110675592A (en) * 2019-08-16 2020-01-10 重庆特斯联智慧科技股份有限公司 High-altitude parabolic early warning protection system based on target identification and control method
CN111079663A (en) * 2019-12-19 2020-04-28 深圳云天励飞技术有限公司 High-altitude parabolic monitoring method and device, electronic equipment and storage medium
CN111553256A (en) * 2020-04-26 2020-08-18 上海天诚比集科技有限公司 High-altitude parabolic early warning identification method based on object track identification
CN111723654A (en) * 2020-05-12 2020-09-29 中国电子***技术有限公司 High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization
CN111639578A (en) * 2020-05-25 2020-09-08 上海中通吉网络技术有限公司 Method, device, equipment and storage medium for intelligently identifying illegal parabola

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于多层感知器和运动轨迹的海上目标类型识别;赵璐 等;《舰船电子工程》;第40卷(第3期);第36-38页 *
拉格朗日粒子模型在核事故应急中的开发与应用;杨晔 等;《原子能科学技术》;第47卷;第713-714页 *

Also Published As

Publication number Publication date
CN112329627A (en) 2021-02-05

Similar Documents

Publication Publication Date Title
CN106133756B (en) System, method and the non-transitory computer-readable medium of filtering, segmentation and identification object
CN112329627B (en) High-altitude throwing object distinguishing method
CN110222769B (en) Improved target detection method based on YOLOV3-tiny
CN109255286B (en) Unmanned aerial vehicle optical rapid detection and identification method based on deep learning network framework
CN109061703A (en) Method, apparatus, equipment and computer readable storage medium used for positioning
Josifovski et al. Object detection and pose estimation based on convolutional neural networks trained with synthetic data
CN109543285B (en) Crowd evacuation simulation method and system integrating data driving and reinforcement learning
CN105550484A (en) Dynamic data driven modeling and simulation method for crowd evacuation and transportation behaviors in sudden-onset disaster
CN108564600B (en) Moving object posture tracking method and device
CN106127125A (en) Distributed DTW human body behavior intension recognizing method based on human body behavior characteristics
Kim et al. Convolutional neural network for monocular vision-based multi-target tracking
CN113313123B (en) Glance path prediction method based on semantic inference
Li et al. Intelligent mobile drone system based on real-time object detection
CN106875403B (en) A kind of imitative hawkeye visual movement object detection method for air refuelling
CN106022266A (en) Target tracking method and target tracking apparatus
CN112347915B (en) High-altitude throwing object distinguishing system
CN112749761A (en) Enemy combat intention identification method and system based on attention mechanism and recurrent neural network
Abdullah et al. Multi-person tracking and crowd behavior detection via particles gradient motion descriptor and improved entropy classifier
CN105069829B (en) A kind of human body animation generation method based on more visually frequencies
CN114299607A (en) Human-vehicle collision risk degree analysis method based on automatic driving of vehicle
Chen et al. Continuous occupancy mapping in dynamic environments using particles
CN106127119A (en) Joint probabilistic data association method based on coloured image and depth image multiple features
CN111414720B (en) Flow field vortex detection method based on neural network
Qiwei et al. SAR image ship detection based on improved YOLOv4
CN109358618B (en) Path planning method of mobile robot

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