CN108805907A - A kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method - Google Patents
A kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method Download PDFInfo
- Publication number
- CN108805907A CN108805907A CN201810578415.5A CN201810578415A CN108805907A CN 108805907 A CN108805907 A CN 108805907A CN 201810578415 A CN201810578415 A CN 201810578415A CN 108805907 A CN108805907 A CN 108805907A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- ant
- value
- image
- chicken
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION methods, including following steps：Step 1：Build pedestrian sample image data base；Step 2：Pedestrian image frame in pedestrian sample database is pre-processed, and to pretreated image setting pedestrian detection frame, pedestrian target mark and pedestrian position label vector；Step 3：Build the pedestrian detection model based on extreme learning machine；Step 4：Build the pedestrian tracking model based on BP neural network；Step 5：Pedestrian track real-time tracking recognizes；This method uses the method based on BP neural network in pedestrian's tracing detection, pedestrian can be realized and quickly and effectively examine and mark, it disclosure satisfy that the requirement identified immediately to emergency in actual traffic environment, it is also applied for intelligent chemical plant, laboratory, under the complex environments such as robot delivery, be conducive to Modern Traffic intelligence, the raising of industrial intelligent.
Description
Technical field
The invention belongs to artificial intelligence field, more particularly to a kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method.
Background technology
With the rapid development of science and technology, answered extensively using the pedestrian detection technology of computer vision the relevant technologies
For the various aspects of life, such as intelligent train, Vehicular automatic driving field.The safety of traffic is eternal topic.In vehicle
Collision class accident in, the collision between vehicle and pedestrian also accounts for prodigious proportion.Nowadays, safety belt, air bag etc.
Conventional security technology popularize comprehensively, however these are all passive guard methods.It is desirable to work out vehicle
Security system is actively protected, and it is the emphasis studied that pedestrian, which is accurately identified and tracked,.
The pedestrian tracting method mostly used at present is description, i.e., is used as macroscopic features of pedestrian such as clothes color etc. and sentences
Disconnected feature extracts the color histogram of image, and then calculates similarity, such method robust by Euclidean distance or Pasteur's distance
Property is relatively low, and effect is undesirable.Also scholar proposes to carry out match cognization, but shallow-layer to pedestrian with the description method of multiple features fusion
Feature is vulnerable to the limitation of descriptive power, and has higher subjective factor.
Chinese patent CN201610317720 discloses a kind of multi-object tracking method based on Recognition with Recurrent Neural Network, including
Following steps：1：Structure is labelled with the monitor video data set of every frame pedestrian position；2：To being labelled with the prison of every frame pedestrian position
Control sets of video data is manually expanded, and training set sample is obtained；3：Training set sample is grouped, multiple training are obtained
Group；4：Build multiple target tracking network；5：Each training group is inputted multiple target tracking network as unit of sequence to be trained；6：
By the multiple target tracking network after video data to be measured input training, propagated forward is carried out, the movement locus of multiple targets is obtained.
There are problems that the following in scheme described in the patent：1. pedestrian's of short duration disappearance weight in video is not considered in pedestrian tracking system
Now or midway has the situation that new pedestrian enters, both the above situation that may lead to system error in judgement；2. needing to pedestrian
Data set is manually expanded so that judges that system is cumbersome, efficiency declines；3. being susceptible to office using Recognition with Recurrent Neural Network algorithm
It restrains in portion.
Invention content
The present invention provides a kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION methods, it is intended that overcoming in the prior art
It is not high to pedestrian track identification precision in monitor video, and the problem that efficiency is low.
A kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method, includes the following steps：
Step 1：Build pedestrian sample image data base；
The pedestrian sample image data base is that continuous pedestrian image frame is extracted from the monitor video of crossing, obtains three classes
Image group；
The three classes image group is respectively free of the negative sample of pedestrian, more proper manners sheets comprising multiple pedestrians and only wraps
Single pedestrian sample containing same a group traveling together includes at least 300 frame images per class image group；
Step 2：Pedestrian image frame in pedestrian sample database is pre-processed, and pretreated image is arranged
Pedestrian detection frame, pedestrian target mark and pedestrian position label vector；
The pedestrian detection frame is the minimum enclosed rectangle of pedestrian image frame middle row people's profile；
The pedestrian target mark is the unique mark P of the different pedestrians occurred in all pedestrian image frames；
The expression-form of the pedestrian position label vector is [t, x, y, a, b], and t indicates that current pedestrian's picture frame belongs to prison
The t frames in video are controlled, x and y indicate the abscissa and ordinate in the lower left corner of the pedestrian detection frame in pedestrian image frame respectively,
A and b indicates that pedestrian detection frame is long and wide respectively；
In the different frame image of monitor video, the target identification of the same pedestrian is identical；
Step 3：Build the pedestrian detection model based on extreme learning machine；
Pass through pretreated image as input data, corresponding pedestrian using pedestrian image frame in pedestrian sample database
Location tags vector sum pedestrian quantity is trained extreme learning machine, obtains based on extreme learning machine as output data
Pedestrian detection model；
For the image not comprising pedestrian, pedestrian's quantity and location tags vector are sky；For more pedestrian samples, pedestrian
Location tags vector number is identical with the number of pedestrian's quantity；
The input layer number of the extreme learning machine is the pixel number s of input picture, hidden layer Wavelet Element
Number is 2s-1, and output layer node number is 4, and the maximum iteration in training process is set as 2000, and training learning rate is
0.01, threshold value 0.00005；
Step 4：Build the pedestrian tracking model based on BP neural network；
It will scheme successively by the pedestrian tracking detection in pretreated adjacent two field pictures and using based on extreme learning machine
Pedestrian detection model extraction correspondence pedestrian position label vector as input layer data, with the row in previous frame line people image
Tracking result of the people in a later frame pedestrian image is output layer data, is trained, is based on to BP neural network model
The pedestrian tracking model of BP neural network；
The pedestrian tracking detection figure as input layer data refers to from a frame by will be single in pretreated image
What a pedestrian contour figure was extracted from figure, it is assumed that as soon as there are 4 pedestrians in frame, then there are 4 pedestrian tracking detection figures；
Appearance result of the pedestrian in a later frame pedestrian image in the former frame pedestrian image refers to if former frame
Pedestrian in pedestrian image occurs in a later frame pedestrian image, then the tracking result of the pedestrian is 1, is otherwise 0；If pedestrian
Tracking result is 1, then the correspondence pedestrian position label vector occurred in a later frame pedestrian image is added in pedestrian track, institute
The initial value for stating pedestrian track is the pedestrian position label vector occurred for the first time in monitor video in the picture；
Trace model is handled two field pictures every time, only judges the pedestrian in former frame pedestrian image,
Whether occur in a later frame pedestrian image, if there is the label vector of people in the second frame is then added to people in first frame
In record；
When using the model, all pedestrian tracking detection figures of former frame and a later frame are combined one by one, as defeated
Enter layer data, matched, if the pedestrian that the pedestrian occurred in the second frame image and first frame occur is same people, by first
The pedestrian target mark occurred in frame, which assigns in the second frame, corresponds to pedestrian, while by the corresponding pedestrian position label vector of the second frame
Charge to the target following track；If the pedestrian occurred in the second frame image does not match with the arbitrary pedestrian occurred in first frame,
New target identification is arranged to the pedestrian occurred in the second frame image；
Step 5：Pedestrian track real-time tracking recognizes；
From real time monitoring video, the adjacent pedestrian image of two frames is extracted successively, and input is described based on extreme learning machine
In pedestrian detection model, the detection of pedestrian position label vector and pedestrian's quantity in two field pictures is carried out, then by two field pictures
In the pedestrian tracking detection figure input pedestrian tracking model based on BP neural network, to occurring in former frame pedestrian image
Pedestrian carry out pedestrian track tracking, obtain in monitor video, the pursuit path of all pedestrians.
Further, following pretreatment is carried out to pedestrian sample image and the monitoring image acquired in real time：
Step A1：Uniform sizes cutting is carried out to the picture frame extracted from the monitor video of crossing；
Step A2：Gray processing processing is carried out to the image after cutting, then picture contrast is adjusted using Gamma correction methods；
Step A3：The histograms of oriented gradients feature of the image after contrast adjustment is extracted, and using PCA to direction
Histogram of gradients feature carries out dimension-reduction treatment；
Step A4：Utilize the histograms of oriented gradients threshold value of histograms of oriented gradients feature and setting after dimensionality reduction, extraction
More than the histograms of oriented gradients feature after the dimensionality reduction of the histograms of oriented gradients threshold value of setting, corresponding pedestrian area is obtained；
Step A5：Smoothing denoising processing is carried out to pedestrian area, and extracts largest connected domain as pedestrian contour region, is obtained
To the image for the people for being more advantageous to neural network recognization；
Step A6：With the maximum width and maximum height in pedestrian contour region, width and height as pedestrian detection frame.
Further, using chicken group's algorithm to the power of the extreme learning machine in the pedestrian detection model based on extreme learning machine
Value and threshold value optimize, and are as follows：
Step B1：Using chicken group body position as the weights of extreme learning machine and threshold value, initiation parameter；
Population scale M=[50,200], search space dimension are j, the value of j be required optimization extreme learning machine weights and
The sum of number of parameters of threshold value, it is maximum to count T=[500,800], iterations t, initial value 0, cock ratio Pg=repeatly
20%, hen ratio Pm=70%, chicken ratio Px=10% randomly choose female godmother chicken, ratio Pd=10% from hen；
Step B2：Fitness function is set, and enables iterations t=1；
The chicken group corresponding weights in body position and threshold value are substituted into the pedestrian detection model based on extreme learning machine successively,
And using the pedestrian detection model based on extreme learning machine of the individual location determination of chicken group, to the mark of the pedestrian in input picture
Label vector is detected, by the detected value for all pedestrian's label vectors for including in input picture and corresponding pedestrian's label vector
The inverse of the absolute value of the sum of the difference of actual value is as the first fitness function f1(x)；
Step B3：Build chicken group subgroup；
It is ranked up according to all ideal adaptation angle value, the chicken group's individual for choosing the preceding M*Pg of fitness value row is determined as public affairs
Chicken, header of the every cock as a sub-group；The chicken group's individual for choosing M*Px after fitness value is arranged is determined as chicken；Other
Chicken group's individual is determined as hen；
Chicken group is divided into, subgroup is divided according to cock number, if a subgroup includes a cock, several chickens and fundatrix
Chicken, and each chicken randomly chooses a hen in population and builds mother-child relationship (MCR)；
Step B4：The individual location updating of chicken group and the fitness for calculating current each individual；
Cock location update formula：
Wherein,Indicate position of the cock i individuals in j dimension spaces in the t times iteration,Corresponding cock individual
The new position in the t+1 times iteration, r (0, σ2) be obey mean value be 0, standard deviation σ2Normal distribution N (0, σ2)；
Hen location update formula：
Wherein,For in the t times iteration hen g in the position of j dimension spaces,For the hen g in the t times iteration
Unique cock i of place subgroup1A body position,For the random public affairs except subgroup where the hen i in the t times iteration
Chicken i2A body position, rand (0,1) are random function, uniformly random value, L between (0,1)1、L2It is hen i by place
The location updating coefficient that group and other subgroups influence, L1Value range [0.3,0.6], L2Value range [0.2,0.4].
Chicken location update formula：
Wherein,For in the t times iteration chicken l in the position of j dimension spaces,For the chicken l in the t times iteration
Female godmother chicken g of corresponding mother-child relationship (MCR)mA body position,For unique cock individual in subgroup where the chicken in the t times iteration
Position, ω, α, β are respectively chicken self-renewing coefficient [0.2,0.7], follow female godmother chicken coefficient [0.5,0.8], follow cock
Coefficient [0.8,1.5]；
Step B5：Personal best particle and all personal best particles of chicken group are updated according to fitness function, is judged whether
Reach maximum iteration, is exited if meeting, otherwise, enable t=t+1, be transferred to step B3, until meeting maximum iteration,
The weights and threshold value for exporting the corresponding extreme learning machine in optimal chicken group body position, obtain the pedestrian detection based on extreme learning machine
Model.
Further, the weights of the BP neural network in the pedestrian tracking model based on BP neural network and threshold value use
Ant lion algorithm optimizes, and is as follows：
Step C1：Using each body position in ant lion group and ant swarm as based in BP neural network pedestrian tracking model
BP neural network weights and threshold value, parameter and population initialization；
Ant lion and ant number are N, and value range is [40,100], maximum iteration T, value range be [600,
2000], the lower border value of parametric variable to be optimized is set as lb, and upper boundary values are set as ub, all weights variables it is upper following
Dividing value is [0.01,0.6], and the value range of the up-and-down boundary of all thresholding variables is [0.0001,0.001]；
Step C2：Initialize the position of all ant lions and ant in ant lion group and ant swarm；
The initial position of ant and ant lion random initializtion in search space, formula are as follows：
Wherein,The position of i-th of individual when for iterations being 1；Rand (0,1) is that rand (0,1) is random function,
The uniformly random value between (0,1)；
Step C3：Fitness function is set, and calculates the fitness of each individual, elite ant is selected according to fitness value
Lion enables iterations t=1；
Ant lion group penalty coefficient corresponding with each body position in ant swarm and nuclear parameter are substituted into and be based on BP nerve nets
In the pedestrian tracking model of network, the pedestrian tracking model based on BP neural network of ant lion and ant individual location determination is obtained
Pedestrian tracking result and actual tracking result between absolute value of the difference inverse as the second fitness function f2(x)；
The big ant lion of fitness value or ant individual are outstanding；
Step C4：It is elite ant lion that the maximum individual of the second fitness function value is selected from ant swarm and ant lion group, is connect
It and is arranged from big to small according to fitness, ant lion is selected as by preceding N-1, it is rear N number of as ant；
Step C5：Ant and ant lion individual location updating calculate the second fitness function value of each individual；
(2) it enables ant individual carry out random walk, and utilizes boundary and the ant lion body position pair chosen using roulette
Ant body position after random walk is normalized；
Wherein, aiAnd biThe minimum value and maximum value on the boundaries in entire walk process corresponding ant individual i, ci tAnd di tIt is right
The minimum value and maximum value on boundary, value are influenced by ant lion position when answering the t times iteration： For the t times iteration when by ant individual i from ant lion group in randomly selected ant lion individual s position, ubtAnd lbtTable respectively
Show the coboundary in the t times iteration and lower boundary；
(2) ant lion preys on ant, update ant lion body position；
If carrying out the fitness that the ant body position after migration is more than the ant lion body position chosen using roulette,
Ant lion preys on the ant, and corresponding ant lion body position is substituted using the ant body position after migration；
(3) updated ant lion body position and elite ant lion position are utilized, formicivorous a body position of being caught is updated；
Indicate the position of the ant individual n of the prey obtained after the t times iteration,WithTable respectively
Ant lion individual s and elite ant lion when the t times iteration when showing the t times iteration；
(4) ant migration bounds are updated；
Wherein, ubtAnd lbtCoboundary and lower boundary, ω when being illustrated respectively in the t times iteration have with current iteration number
It closes,
(5) the second fitness function value of all individuals is calculated；
Step C6：Judge whether to meet maximum iteration, if not satisfied, then t=t+1, return to step C4, until meeting
After maximum iteration, with corresponding elite ant lion body position when the second fitness function value maximum, determine based on BP nerves
The weights and threshold value of the pedestrian tracking model of network.
Advantageous effect
The present invention provides a kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION methods, including following steps：Step 1：Structure
Build pedestrian sample image data base；Step 2：Pedestrian image frame in pedestrian sample database is pre-processed, and to locating in advance
Image setting pedestrian detection frame, pedestrian target mark and pedestrian position label vector after reason；Step 3：Structure is based on the limit
The pedestrian detection model of learning machine；Step 4：Build the pedestrian tracking model based on BP neural network；Step 5：Pedestrian track is real
When tracking identification；
This method compared to existing technologies, has the following advantages：
1. detection is accurate：The method based on neural network is used in pedestrian detection, pedestrian can be realized and quickly and effectively be examined
And label, it disclosure satisfy that the requirement identified immediately to emergency in actual traffic environment, be also applied for intelligent chemical plant, test
Room, robot delivery etc. are conducive to Modern Traffic intelligence, the raising of industrial intelligent under complex environments；
2. discrimination is high again, by neural network automatically to tracking Objective extraction higher level of abstraction feature, realize to tracking target
Efficiently quickly matching identifies again；
3. carrying out tuning processing to neural network parameter using optimization algorithm, the operational efficiency of neural network can be improved,
The precision that neural network identifies pedestrian is improved, copes with the traffic pedestrian of big flow, robustness is good.
Pedestrian is detached from image using image processing method 4. novelty proposes, carry out frame by frame comparison carry out with
Track ensure that the stability of pedestrian tracking, it is not easy to lose target.
5. remote processing image data saves equipment cost it is not necessary that street corner equipment is added or updated.
Description of the drawings
Fig. 1 is the flow diagram of the method for the invention.
Specific implementation mode
Below in conjunction with attached drawing and example, the present invention is described further.
As shown in Figure 1, a kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method, includes the following steps：
Step 1：Build pedestrian sample image data base；
The pedestrian sample image data base is that continuous pedestrian image frame is extracted from the monitor video of crossing, obtains three classes
Image group；
The three classes image group is respectively free of the negative sample of pedestrian, more proper manners sheets comprising multiple pedestrians and only wraps
Single pedestrian sample containing same a group traveling together includes at least 300 frame images per class image group；
Step 2：Pedestrian image frame in pedestrian sample database is pre-processed, and pretreated image is arranged
Pedestrian detection frame, pedestrian target mark and pedestrian position label vector；
The pedestrian detection frame is the minimum enclosed rectangle of pedestrian image frame middle row people's profile；
The pedestrian target mark is the unique mark P of the different pedestrians occurred in all pedestrian image frames；
The expression-form of the pedestrian position label vector is [t, x, y, a, b], and t indicates that current pedestrian's picture frame belongs to prison
The t frames in video are controlled, x and y indicate the abscissa and ordinate in the lower left corner of the pedestrian detection frame in pedestrian image frame respectively,
A and b indicates that pedestrian detection frame is long and wide respectively；
In the different frame image of monitor video, the target identification of the same pedestrian is identical；
Following pretreatment is carried out to pedestrian sample image and the monitoring image acquired in real time：
Step A1：Uniform sizes cutting is carried out to the picture frame extracted from the monitor video of crossing；
Step A2：Gray processing processing is carried out to the image after cutting, then picture contrast is adjusted using Gamma correction methods；
Step A3：The histograms of oriented gradients feature of the image after contrast adjustment is extracted, and using PCA to direction
Histogram of gradients feature carries out dimension-reduction treatment；
Step A4：Utilize the histograms of oriented gradients threshold value of histograms of oriented gradients feature and setting after dimensionality reduction, extraction
More than the histograms of oriented gradients feature after the dimensionality reduction of the histograms of oriented gradients threshold value of setting, corresponding pedestrian area is obtained；
Step A5：Smoothing denoising processing is carried out to pedestrian area, and extracts largest connected domain as pedestrian contour region；
Obtain the image for the people for being more advantageous to neural network recognization；
Step A6：With the maximum width and maximum height in pedestrian contour region, width and height as pedestrian detection frame.
Step 3：Build the pedestrian detection model based on extreme learning machine；
Pass through pretreated image as input data, corresponding pedestrian using pedestrian image frame in pedestrian sample database
Location tags vector sum pedestrian quantity is trained extreme learning machine, obtains based on extreme learning machine as output data
Pedestrian detection model；
For the image not comprising pedestrian, pedestrian's quantity and location tags vector are sky；For more pedestrian samples, pedestrian
Location tags vector number is identical with the number of pedestrian's quantity；
The input layer number of the extreme learning machine is the pixel number s of input picture, hidden layer Wavelet Element
Number is 2s-1, and output layer node number is 4, and the maximum iteration in training process is set as 2000, and training learning rate is
0.01, threshold value 0.00005；
Chicken group's algorithm basic principle is as follows：
1) there is several subgroups in entire chicken group, if each subgroup is by a cock, dried hen and some chicken groups
At.
2) how chicken group is divided into several subgroups and how to determine that the type of chicken depends on the fitness value of chicken itself.Chicken
In group, the best several body of fitness value is as cock, and every cock is all the head of a subgroup；With worst suitable
Answer the several body of angle value as chicken；Remaining individual is just used as hen.At will which subgroup selection belongs to hen, hen and
The mother-child relationship (MCR) of chicken is also to establish at random.
3) cock search of food of the individual in each subgroup in this subgroup, can also prevent other individuals from robbing
Take the food of oneself by force；And assume that chicken can steal the food that Shiqi his individual has been found that at random, every chicken follows them
Mother's search of food together.The individual with ascendancy is with good competitive advantage in chicken group, they are prior to other
Body finds food.
Female godmother chicken is that the hen in the randomly selected sub-group of chicken is used as female godmother chicken, carries out following study；
Using chicken group's algorithm to the weights and threshold value of the extreme learning machine in the pedestrian detection model based on extreme learning machine
It optimizes, is as follows：
Step B1：Using chicken group body position as the weights of extreme learning machine and threshold value, initiation parameter；
Population scale M=[50,200], search space dimension are j, the value of j be required optimization extreme learning machine weights and
The sum of number of parameters of threshold value, it is maximum to count T=[500,800], iterations t, initial value 0, cock ratio Pg=repeatly
20%, hen ratio Pm=70%, chicken ratio Px=10% randomly choose female godmother chicken, ratio Pd=10% from hen；
Step B2：Fitness function is set, and enables iterations t=1；
The chicken group corresponding weights in body position and threshold value are substituted into the pedestrian detection model based on extreme learning machine successively,
And using the pedestrian detection model based on extreme learning machine of the individual location determination of chicken group, to the mark of the pedestrian in input picture
Label vector is detected, by the detected value for all pedestrian's label vectors for including in input picture and corresponding pedestrian's label vector
The inverse of the absolute value of the sum of the difference of actual value is as the first fitness function f1(x)；
Step B3：Build chicken group subgroup；
It is ranked up according to all ideal adaptation angle value, the chicken group's individual for choosing the preceding M*Pg of fitness value row is determined as public affairs
Chicken, header of the every cock as a sub-group；The chicken group's individual for choosing M*Px after fitness value is arranged is determined as chicken；Other
Chicken group's individual is determined as hen；
Chicken group is divided into, subgroup is divided according to cock number, if a subgroup includes a cock, several chickens and fundatrix
Chicken, and each chicken randomly chooses a hen in population and builds mother-child relationship (MCR)；
Step B4：The individual location updating of chicken group and the fitness for calculating current each individual；
Cock location update formula：
Wherein,Indicate position of the cock i individuals in j dimension spaces in the t times iteration,Corresponding cock individual
The new position in the t+1 times iteration, r (0, σ2) be obey mean value be 0, standard deviation σ2Normal distribution N (0, σ2)；
Hen location update formula：
Wherein,For in the t times iteration hen g in the position of j dimension spaces,For the hen g in the t times iteration
Unique cock i of place subgroup1A body position,For the random public affairs except subgroup where the hen i in the t times iteration
Chicken i2A body position, rand (0,1) are random function, uniformly random value, L between (0,1)1、L2It is hen i by place
The location updating coefficient that group and other subgroups influence, L1Value range [0.3,0.6], L2Value range [0.2,0.4].
Chicken location update formula：
Wherein,For in the t times iteration chicken l in the position of j dimension spaces,For the chicken l in the t times iteration
Female godmother chicken g of corresponding mother-child relationship (MCR)mA body position,For unique cock individual in subgroup where the chicken in the t times iteration
Position, ω, α, β are respectively chicken self-renewing coefficient [0.2,0.7], follow female godmother chicken coefficient [0.5,0.8], follow cock
Coefficient [0.8,1.5]；
Step B5：Personal best particle and all personal best particles of chicken group are updated according to fitness function, is judged whether
Reach maximum iteration, is exited if meeting, otherwise, enable t=t+1, be transferred to step B3, until meeting maximum iteration,
The weights and threshold value for exporting the corresponding extreme learning machine in optimal chicken group body position, obtain the pedestrian detection based on extreme learning machine
Model.
Step 4：Build the pedestrian tracking model based on BP neural network；
It will scheme successively by the pedestrian tracking detection in pretreated adjacent two field pictures and using based on extreme learning machine
Pedestrian detection model extraction correspondence pedestrian position label vector as input layer data, with the row in previous frame line people image
Tracking result of the people in a later frame pedestrian image is output layer data, is trained, is based on to BP neural network model
The pedestrian tracking model of BP neural network；
Ant lion algorithm basic principle is as follows：
Ant lion optimization algorithm be in by nature ant lion catch ant hunting mechanism inspire and propose a kind of new group
Body intelligent optimization algorithm.Ant lion is moved in sand along circular trace in nature, and the circle of a trapping ant is dug out using lower jaw
Conical pit, when the ant of random movement is absorbed in hole, it is that ant lion just preys on and again repair hole wait for next prey (ant
Ant).
Ant lion algorithm is exactly to imitate this interaction of ant lion and ant to realize the optimization to problem：Ant is by enclosing
The exploration to search space is realized in random walk around ant lion, and is learnt to ant lion and elite to ensure the diversity of population
With the optimizing performance of algorithm；Ant lion is equivalent to the solution of problem, and pairing approximation optimal solution is realized by catching the high ant of fitness
Update and preservation.
The weights and threshold value of BP neural network in the pedestrian tracking model based on BP neural network use ant lion algorithm
It optimizes, is as follows：
Step C1：Using each body position in ant lion group and ant swarm as based in BP neural network pedestrian tracking model
BP neural network weights and threshold value, parameter and population initialization；
Ant lion and ant number are N, and value range is [40,100], maximum iteration T, value range be [600,
2000], the lower border value of parametric variable to be optimized is set as lb, and upper boundary values are set as ub, all weights variables it is upper following
Dividing value is [0.01,0.6], and the value range of the up-and-down boundary of all thresholding variables is [0.0001,0.001]；
Step C2：Initialize the position of all ant lions and ant in ant lion group and ant swarm；
The initial position of ant and ant lion random initializtion in search space, formula are as follows：
Wherein,The position of i-th of individual when for iterations being 1；Rand (0,1) is that rand (0,1) is random function,
The uniformly random value between (0,1)；
Step C3：Fitness function is set, and calculates the fitness of each individual, elite ant is selected according to fitness value
Lion enables iterations t=1；
Ant lion group penalty coefficient corresponding with each body position in ant swarm and nuclear parameter are substituted into and be based on BP nerve nets
In the pedestrian tracking model of network, the pedestrian tracking model based on BP neural network of ant lion and ant individual location determination is obtained
Pedestrian tracking result and actual tracking result between absolute value of the difference inverse as the second fitness function f2(x)；
The big ant lion of fitness value or ant individual are outstanding；
Step C4：It is elite ant lion that the maximum individual of the second fitness function value is selected from ant swarm and ant lion group, is connect
It and is arranged from big to small according to fitness, ant lion is selected as by preceding N-1, it is rear N number of as ant；
Step C5：Ant and ant lion individual location updating calculate the second fitness function value of each individual；
(3) it enables ant individual carry out random walk, and utilizes boundary and the ant lion body position pair chosen using roulette
Ant body position after random walk is normalized；
Ant carries out random walk behavior, and specific formula is as follows：
Wherein cumsum is to calculate accumulated value, and T is maximum iteration, and t is current iteration number, and r (t) is one random
Function, formula are as follows：
Ant random walk in order to prevent is crossed the border, and according to boundary, ant random walk is normalized：
Wherein, aiAnd biThe minimum value and maximum value on the boundaries in entire walk process corresponding ant individual i, ci tAnd di tIt is right
The minimum value and maximum value on boundary, value are influenced by ant lion position when answering the t times iteration： For the t times iteration when by ant individual i from ant lion group in randomly selected ant lion individual s position；ubtAnd lbtTable respectively
Show the coboundary in the t times iteration and lower boundary；
(2) ant lion preys on ant, update ant lion body position；
If carrying out the fitness that the ant body position after migration is more than the ant lion body position chosen using roulette,
Ant lion preys on the ant, and corresponding ant lion body position is substituted using the ant body position after migration；
(3) updated ant lion body position and elite ant lion position are utilized, formicivorous a body position of being caught is updated；
Indicate the position of the ant individual n of the prey obtained after the t times iteration,WithTable respectively
Ant lion individual s and elite ant lion when the t times iteration when showing the t times iteration；
(4) ant migration bounds are updated；
Wherein, ubtAnd lbtCoboundary and lower boundary, ω when being illustrated respectively in the t times iteration have with current iteration number
It closes,
(5) the second fitness function value of all individuals is calculated；
Step C6：Judge whether to meet maximum iteration, if not satisfied, then t=t+1, return to step C4, until meeting
After maximum iteration, with corresponding elite ant lion body position when the second fitness function value maximum, determine based on BP nerves
The weights and threshold value of the pedestrian tracking model of network.
The pedestrian tracking detection figure as input layer data refers to from a frame by will be single in pretreated image
What a pedestrian contour figure was extracted from figure, it is assumed that as soon as there are 4 pedestrians in frame, then there are 4 pedestrian tracking detection figures；
Appearance result of the pedestrian in a later frame pedestrian image in the former frame pedestrian image refers to if former frame
Pedestrian in pedestrian image occurs in a later frame pedestrian image, then the tracking result of the pedestrian is 1, is otherwise 0；If pedestrian
Tracking result is 1, then the correspondence pedestrian position label vector occurred in a later frame pedestrian image is added in pedestrian track, institute
The initial value for stating pedestrian track is the pedestrian position label vector occurred for the first time in monitor video in the picture；
Trace model is handled two field pictures every time, only judges the pedestrian in former frame pedestrian image,
Whether occur in a later frame pedestrian image, if there is the label vector of people in the second frame is then added to people in first frame
In record；
When using the model, all pedestrian tracking detection figures of former frame and a later frame are combined one by one, as defeated
Enter layer data, matched, if the pedestrian that the pedestrian occurred in the second frame image and first frame occur is same people, by first
The pedestrian target mark occurred in frame, which assigns in the second frame, corresponds to pedestrian, while by the corresponding pedestrian position label vector of the second frame
Charge to the target following track；If the pedestrian occurred in the second frame image does not match with the arbitrary pedestrian occurred in first frame,
New target identification is arranged to the pedestrian occurred in the second frame image；
Step 5：Pedestrian track real-time tracking recognizes；
From real time monitoring video, the adjacent pedestrian image of two frames is extracted successively, and input is described based on extreme learning machine
In pedestrian detection model, the detection of pedestrian position label vector and pedestrian's quantity in two field pictures is carried out, then by two field pictures
In the pedestrian tracking detection figure input pedestrian tracking model based on BP neural network, to occurring in former frame pedestrian image
Pedestrian carry out pedestrian track tracking, obtain in monitor video, the pursuit path of all pedestrians.
Specific embodiment described herein is only an example for the spirit of the invention.Technology belonging to the present invention is led
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Claims (4)
1. a kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method, which is characterized in that include the following steps：
Step 1：Build pedestrian sample image data base；
The pedestrian sample image data base is that continuous pedestrian image frame is extracted from the monitor video of crossing, obtains three classes image
Group；
The three classes image group, is respectively free of the negative sample of pedestrian, more proper manners sheets comprising multiple pedestrians and includes only same
Single pedestrian sample of a group traveling together includes at least 300 frame images per class image group；
Step 2：Pedestrian image frame in pedestrian sample database is pre-processed, and pedestrian is arranged to pretreated image
Detection block, pedestrian target mark and pedestrian position label vector；
The pedestrian detection frame is the minimum enclosed rectangle of pedestrian image frame middle row people's profile；
The pedestrian target mark is the unique mark P of the different pedestrians occurred in all pedestrian image frames；
The expression-form of the pedestrian position label vector is [t, x, y, a, b], and t indicates that current pedestrian's picture frame belongs to monitoring and regards
T frames in frequency, x and y indicate the abscissa and ordinate in the lower left corner of the pedestrian detection frame in pedestrian image frame, a and b respectively
Indicate that pedestrian detection frame is long and wide respectively；
Step 3：Build the pedestrian detection model based on extreme learning machine；
Pass through pretreated image as input data, corresponding pedestrian position using pedestrian image frame in pedestrian sample database
Label vector and pedestrian's quantity are trained extreme learning machine as output data, obtain the pedestrian based on extreme learning machine
Detection model；
The input layer number of the extreme learning machine is the pixel number s of input picture, and hidden layer Wavelet Element number is
2s-1, output layer node number are 4, and the maximum iteration in training process is set as 2000, and training learning rate is 0.01, threshold
Value is 0.00005；
Step 4：Build the pedestrian tracking model based on BP neural network；
It will scheme and utilize the row based on extreme learning machine by the pedestrian tracking detection in pretreated adjacent two field pictures successively
The correspondence pedestrian position label vector of people's detection model extraction is existed as input layer data with the pedestrian in previous frame line people image
Tracking result in a later frame pedestrian image is output layer data, is trained to BP neural network model, is obtained based on BP god
Pedestrian tracking model through network；
Appearance result of the pedestrian in a later frame pedestrian image in the former frame pedestrian image refers to if former frame pedestrian
Pedestrian in image occurs in a later frame pedestrian image, then the tracking result of the pedestrian is 1, is otherwise 0；If pedestrian tracking
As a result it is 1, then the correspondence pedestrian position label vector occurred in a later frame pedestrian image is added in pedestrian track, the row
The initial value of people track is the pedestrian position label vector occurred for the first time in monitor video in the picture；
Step 5：Pedestrian track real-time tracking recognizes；
From real time monitoring video, the adjacent pedestrian image of two frames is extracted successively, inputs the pedestrian based on extreme learning machine
In detection model, the detection of pedestrian position label vector and pedestrian's quantity in two field pictures is carried out, it then will be in two field pictures
The pedestrian tracking detection figure input pedestrian tracking model based on BP neural network, to the row occurred in former frame pedestrian image
People carries out pedestrian track tracking, obtains in monitor video, the pursuit path of all pedestrians.
2. according to the method described in claim 1, it is characterized in that, to pedestrian sample image and the monitoring image acquired in real time into
The following pretreatment of row：
Step A1：Uniform sizes cutting is carried out to the picture frame extracted from the monitor video of crossing；
Step A2：Gray processing processing is carried out to the image after cutting, then picture contrast is adjusted using Gamma correction methods；
Step A3：The histograms of oriented gradients feature of the image after contrast adjustment is extracted, and using PCA to direction gradient
Histogram feature carries out dimension-reduction treatment；
Step A4：Using the histograms of oriented gradients threshold value of histograms of oriented gradients feature and setting after dimensionality reduction, extraction is more than
Histograms of oriented gradients feature after the dimensionality reduction of the histograms of oriented gradients threshold value of setting, obtains corresponding pedestrian area；
Step A5：Smoothing denoising processing is carried out to pedestrian area, and extracts largest connected domain as pedestrian contour region；
Step A6：With the maximum width and maximum height in pedestrian contour region, width and height as pedestrian detection frame.
3. according to the method described in claim 1, it is characterized in that, being examined to the pedestrian based on extreme learning machine using chicken group's algorithm
The weights and threshold value for the extreme learning machine surveyed in model optimize, and are as follows：
Step B1：Using chicken group body position as the weights of extreme learning machine and threshold value, initiation parameter；
Population scale M=[50,200], search space dimension are j, and the value of j is the weights and threshold value of required optimization extreme learning machine
The sum of number of parameters, maximum to count T=[500,800] repeatly, iterations t, initial value 0, cock ratio Pg=20%,
Hen ratio Pm=70%, chicken ratio Px=10% randomly choose female godmother chicken, ratio Pd=10% from hen；
Step B2：Fitness function is set, and enables iterations t=1；
The chicken group corresponding weights in body position and threshold value are substituted into the pedestrian detection model based on extreme learning machine successively, and profit
With the pedestrian detection model based on extreme learning machine of the individual location determination of chicken group, to the label of the pedestrian in input picture to
Amount is detected, by the detected value for all pedestrian's label vectors for including in input picture and corresponding pedestrian's label vector reality
The inverse of the absolute value of the sum of the difference of value is as the first fitness function f1(x)；
Step B3：Build chicken group subgroup；
It is ranked up according to all ideal adaptation angle value, the chicken group's individual for choosing the preceding M*Pg of fitness value row is determined as cock, often
Header of the cock as a sub-group；The chicken group's individual for choosing M*Px after fitness value is arranged is determined as chicken；Other chickens group
Individual is determined as hen；
Chicken group is divided into, subgroup is divided according to cock number, if a subgroup includes a cock, several chickens and dried hen, and
And each chicken randomly chooses a hen in population and builds mother-child relationship (MCR)；
Step B4：The individual location updating of chicken group and the fitness for calculating current each individual；
Cock location update formula：
Wherein,Indicate position of the cock i individuals in j dimension spaces in the t times iteration,The corresponding cock individual is the
New position in t+1 iteration, r (0, σ2) be obey mean value be 0, standard deviation σ2Normal distribution N (0, σ2)；
Hen location update formula：
Wherein,For in the t times iteration hen g in the position of j dimension spaces,It is sub where the hen g in the t times iteration
Unique cock i of group1A body position,For the random cock i except subgroup where the hen i in the t times iteration2It is a
Body position, rand (0,1) are random function, uniformly random value, L between (0,1)1、L2It is hen i by place subgroup and its
The location updating coefficient that his subgroup influences, L1Value range [0.3,0.6], L2Value range [0.2,0.4].
Chicken location update formula：
Wherein,For in the t times iteration chicken l in the position of j dimension spaces,For in the t times iteration chicken l correspond to
Female godmother chicken g of mother-child relationship (MCR)mA body position,For unique cock position in subgroup where the chicken in the t times iteration
It sets, ω, α, β are respectively chicken self-renewing coefficient [0.2,0.7], follow female godmother chicken coefficient [0.5,0.8], follow the cock to be
Number [0.8,1.5]；
Step B5：Personal best particle and all personal best particles of chicken group are updated according to fitness function, judges whether to reach
Maximum iteration exits if meeting, otherwise, enables t=t+1, be transferred to step B3, until meeting maximum iteration, exports
The weights and threshold value of the corresponding extreme learning machine in optimal chicken group body position, obtain the pedestrian detection mould based on extreme learning machine
Type.
4. according to claim 1-3 any one of them methods, which is characterized in that described to be based on BP neural network pedestrian tracking mould
The weights and threshold value of BP neural network in type are optimized using ant lion algorithm, are as follows：
Step C1：Using each body position in ant lion group and ant swarm as based on the BP in BP neural network pedestrian tracking model
The weights and threshold value of neural network, parameter and population initialization；
Ant lion and ant number are N, and value range is [40,100], maximum iteration T, value range be [600,
2000], the lower border value of parametric variable to be optimized is set as lb, and upper boundary values are set as ub, all weights variables it is upper following
Dividing value is [0.01,0.6], and the value range of the up-and-down boundary of all thresholding variables is [0.0001,0.001]；
Step C2：Initialize the position of all ant lions and ant in ant lion group and ant swarm；
The initial position of ant and ant lion random initializtion in search space, formula are as follows：
Wherein,The position of i-th of individual when for iterations being 1；Rand (0,1) is that rand (0,1) is random function,
(0,1) uniformly random value between；
Step C3：Fitness function is set, and calculates the fitness of each individual, selects elite ant lion according to fitness value, enables
Iterations t=1；
Ant lion group penalty coefficient corresponding with each body position in ant swarm and nuclear parameter are substituted into based on BP neural network
In pedestrian tracking model, row that the pedestrian tracking model based on BP neural network of ant lion and ant individual location determination is obtained
The inverse of absolute value of the difference is as the second fitness function f between people's tracking result and actual tracking result2(x)；
Step C4：It is elite ant lion that the maximum individual of the second fitness function value is selected from ant swarm and ant lion group, is then pressed
It is arranged from big to small according to fitness, ant lion is selected as by preceding N-1, it is rear N number of as ant；
Step C5：Ant and ant lion individual location updating calculate the second fitness function value of each individual；
(1) ant individual is enabled to carry out random walk, and using boundary and the ant lion body position chosen using roulette to random
Ant body position after migration is normalized；
Wherein, aiAnd biThe minimum value and maximum value on the boundaries in entire walk process corresponding ant individual i, cit and dit are corresponded to
The minimum value and maximum value on boundary, value are influenced by ant lion position when t times iteration： For the t times iteration when by ant individual i from ant lion group in randomly selected ant lion individual s position；ubtAnd lbtTable respectively
Show the coboundary in the t times iteration and lower boundary；
(2) ant lion preys on ant, update ant lion body position；
If carrying out the fitness that the ant body position after migration is more than the ant lion body position chosen using roulette, ant lion
The ant is preyed on, corresponding ant lion body position is substituted using the ant body position after migration；
(3) updated ant lion body position and elite ant lion position are utilized, formicivorous a body position of being caught is updated；
Indicate the position of the ant individual n of the prey obtained after the t times iteration,WithT is indicated respectively
Ant lion individual s and elite ant lion when the t times iteration when secondary iteration；
(4) ant migration bounds are updated；
Wherein, ubtAnd lbtCoboundary when being illustrated respectively in the t times iteration and lower boundary, ω is related with current iteration number,
(5) the second fitness function value of all individuals is calculated；
Step C6：Judge whether to meet maximum iteration, if not satisfied, then t=t+1, return to step C4, maximum until meeting
After iterations, with corresponding elite ant lion body position when the second fitness function value maximum, determines and be based on BP neural network
Pedestrian tracking model weights and threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810578415.5A CN108805907B (en) | 2018-06-05 | 2018-06-05 | Pedestrian posture multi-feature intelligent identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810578415.5A CN108805907B (en) | 2018-06-05 | 2018-06-05 | Pedestrian posture multi-feature intelligent identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108805907A true CN108805907A (en) | 2018-11-13 |
| CN108805907B CN108805907B (en) | 2022-03-29 |
Family
ID=64088732
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810578415.5A Active CN108805907B (en) | 2018-06-05 | 2018-06-05 | Pedestrian posture multi-feature intelligent identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108805907B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712160A (en) * | 2018-12-26 | 2019-05-03 | 桂林电子科技大学 | Improved lion group algorithm is combined to realize carrying out image threshold segmentation method based on generalized entropy |
| CN110288606A (en) * | 2019-06-28 | 2019-09-27 | 中北大学 | A kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization |
| CN112836721A (en) * | 2020-12-17 | 2021-05-25 | 北京仿真中心 | Image identification method and device, computer equipment and readable storage medium |
| CN113239900A (en) * | 2021-06-17 | 2021-08-10 | 云从科技集团股份有限公司 | Human body position detection method and device and computer readable storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105224912A (en) * | 2015-08-31 | 2016-01-06 | 电子科技大学 | Based on the video pedestrian detection and tracking method of movable information and Track association |
| US20170161591A1 (en) * | 2015-12-04 | 2017-06-08 | Pilot Ai Labs, Inc. | System and method for deep-learning based object tracking |
| WO2017120336A2 (en) * | 2016-01-05 | 2017-07-13 | Mobileye Vision Technologies Ltd. | Trained navigational system with imposed constraints |
| CN107274433A (en) * | 2017-06-21 | 2017-10-20 | 吉林大学 | Method for tracking target, device and storage medium based on deep learning |
| CN107330920A (en) * | 2017-06-28 | 2017-11-07 | 华中科技大学 | A kind of monitor video multi-target tracking method based on deep learning |
| CN107563313A (en) * | 2017-08-18 | 2018-01-09 | 北京航空航天大学 | Multiple target pedestrian detection and tracking based on deep learning |
| CN107767405A (en) * | 2017-09-29 | 2018-03-06 | 华中科技大学 | A kind of nuclear phase for merging convolutional neural networks closes filtered target tracking |
| CN108021848A (en) * | 2016-11-03 | 2018-05-11 | 浙江宇视科技有限公司 | Passenger flow volume statistical method and device |
-
2018
- 2018-06-05 CN CN201810578415.5A patent/CN108805907B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105224912A (en) * | 2015-08-31 | 2016-01-06 | 电子科技大学 | Based on the video pedestrian detection and tracking method of movable information and Track association |
| US20170161591A1 (en) * | 2015-12-04 | 2017-06-08 | Pilot Ai Labs, Inc. | System and method for deep-learning based object tracking |
| WO2017120336A2 (en) * | 2016-01-05 | 2017-07-13 | Mobileye Vision Technologies Ltd. | Trained navigational system with imposed constraints |
| CN108021848A (en) * | 2016-11-03 | 2018-05-11 | 浙江宇视科技有限公司 | Passenger flow volume statistical method and device |
| CN107274433A (en) * | 2017-06-21 | 2017-10-20 | 吉林大学 | Method for tracking target, device and storage medium based on deep learning |
| CN107330920A (en) * | 2017-06-28 | 2017-11-07 | 华中科技大学 | A kind of monitor video multi-target tracking method based on deep learning |
| CN107563313A (en) * | 2017-08-18 | 2018-01-09 | 北京航空航天大学 | Multiple target pedestrian detection and tracking based on deep learning |
| CN107767405A (en) * | 2017-09-29 | 2018-03-06 | 华中科技大学 | A kind of nuclear phase for merging convolutional neural networks closes filtered target tracking |
Non-Patent Citations (5)
| Title |
|---|
| CHRISTOPH FEICHTENHOFER 等: "Detect to Track and Track to Detect", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| ZHENGHAO SHI 等: "Object Location and Track in Image Sequences by Means of Neural Networks", 《INTERNATIONAL JOURNAL OF COMPUTATIONAL SCIENCE》 * |
| 唐军祥 等: "一种无刷同步发电机旋转整流器故障快速识别方法", 《计算机月现代化》 * |
| 崔东文 等: "基于ALO－ENN算法的洪灾评估模型及应用", 《人民珠江》 * |
| 程朋 等: "基于卷积神经网络的实时跟踪算法", 《传感器与微系统》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712160A (en) * | 2018-12-26 | 2019-05-03 | 桂林电子科技大学 | Improved lion group algorithm is combined to realize carrying out image threshold segmentation method based on generalized entropy |
| CN109712160B (en) * | 2018-12-26 | 2023-05-23 | 桂林电子科技大学 | Method for realizing image threshold segmentation based on generalized entropy combined improved lion group algorithm |
| CN110288606A (en) * | 2019-06-28 | 2019-09-27 | 中北大学 | A kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization |
| CN110288606B (en) * | 2019-06-28 | 2024-04-09 | 中北大学 | Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization |
| CN112836721A (en) * | 2020-12-17 | 2021-05-25 | 北京仿真中心 | Image identification method and device, computer equipment and readable storage medium |
| CN112836721B (en) * | 2020-12-17 | 2024-03-22 | 北京仿真中心 | Image recognition method and device, computer equipment and readable storage medium |
| CN113239900A (en) * | 2021-06-17 | 2021-08-10 | 云从科技集团股份有限公司 | Human body position detection method and device and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108805907B (en) | 2022-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104063719B (en) | Pedestrian detection method and device based on depth convolutional network | |
| CN108805907A (en) | A kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method | |
| CN104598890B (en) | A kind of Human bodys' response method based on RGB D videos | |
| CN105069472B (en) | A kind of vehicle checking method adaptive based on convolutional neural networks | |
| CN107016357B (en) | Video pedestrian detection method based on time domain convolutional neural network | |
| CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
| CN103473539B (en) | Gait recognition method and device | |
| CN110135314A (en) | A kind of multi-object tracking method based on depth Trajectory prediction | |
| CN108764050A (en) | Skeleton Activity recognition method, system and equipment based on angle independence | |
| CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
| CN110399856A (en) | Feature extraction network training method, image processing method, device and its equipment | |
| CN108898093A (en) | A kind of face identification method and the electronic health record login system using this method | |
| Yu et al. | An object-based visual attention model for robotic applications | |
| CN108009482A (en) | One kind improves recognition of face efficiency method | |
| CN110008867A (en) | A kind of method for early warning based on personage's abnormal behaviour, device and storage medium | |
| CN106156765A (en) | safety detection method based on computer vision | |
| CN109241829A (en) | The Activity recognition method and device of convolutional neural networks is paid attention to based on space-time | |
| CN107967442A (en) | A kind of finger vein identification method and system based on unsupervised learning and deep layer network | |
| CN110532850A (en) | A kind of fall detection method based on video artis and hybrid classifer | |
| Kumar et al. | An improved method of segmentation using fuzzy-neuro logic | |
| CN108537181A (en) | A kind of gait recognition method based on the study of big spacing depth measure | |
| CN109241814A (en) | Pedestrian detection method based on YOLO neural network | |
| CN106874825A (en) | The training method of Face datection, detection method and device | |
| CN106127161A (en) | Fast target detection method based on cascade multilayer detector | |
| CN110363238A (en) | Intelligent vehicle damage identification method, system, electronic equipment and storage medium |
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 |