CN105956551A - Target detection method and device - Google Patents
Target detection method and device Download PDFInfo
- Publication number
- CN105956551A CN105956551A CN201610280638.4A CN201610280638A CN105956551A CN 105956551 A CN105956551 A CN 105956551A CN 201610280638 A CN201610280638 A CN 201610280638A CN 105956551 A CN105956551 A CN 105956551A
- Authority
- CN
- China
- Prior art keywords
- targeted customer
- photo
- identified
- hidden markov
- observed
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 50
- 238000000034 method Methods 0.000 claims abstract description 27
- 230000001143 conditioned effect Effects 0.000 claims description 14
- 239000000284 extract Substances 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 description 8
- 230000009466 transformation Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000006243 chemical reaction Methods 0.000 description 5
- 239000004744 fabric Substances 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 208000010086 Hypertelorism Diseases 0.000 description 2
- 206010020771 Hypertelorism of orbit Diseases 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 239000007795 chemical reaction product Substances 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a target detection method. The method comprises the steps of: obtaining a photograph to be identified, extracting a vector sequence to be observed according to an RGB pixel value of the photograph to be identified, calculating the similarity between the vector sequence to be observed and a hidden Markov model corresponding to a preset target user, and when the similarity reaches a preset condition, determining that the target user is detected, and determining that a user to be identified in the photograph to be identified is the target user corresponding to the hidden Markov model when the similarity reaches the preset condition. The invention further discloses a target detection device. According to the invention, the RGB pixel value is utilized in the detection process, face information is not relied on, the target user can be identified directly according to the RGB pixel value, the tracked user does not need to face a camera of a terminal in the identification and tracking process, and the tracked user can walk randomly, so that the terminal is more intelligent and convenient in the target identification process.
Description
Technical field
The present invention relates to technical field of image processing, particularly relate to a kind of object detection method and device.
Background technology
At present, a lot of end products have identification function, such as robot, can complete the knowledge to face
Other function.Prior art is in face recognition process, and terminal must photograph the face letter of user to be identified
Breath, it is thus achieved that after the human face photo of user to be identified, could be according to the human face photo obtained to user to be identified
It is identified.If user is walking about or back to terminal, cause terminal cannot photograph user's to be identified
Face information, then can not complete the identification to user to be identified.The target recognition of existing terminal compares limitation
Property.
Summary of the invention
Present invention is primarily targeted at a kind of object detection method of offer and device, it is intended to solve existing end
The more circumscribed technical problem of target recognition of end.
The object detection method that the present invention provides comprises the following steps:
Obtain photo to be identified, and extract vector to be observed according to the rgb pixel value of described photo to be identified
Sequence；
Calculate the phase of the described sequence vector to be observed hidden Markov model corresponding with the targeted customer preset
Like degree；
When described similarity reaches pre-conditioned, determine and targeted customer detected, and by described to be identified
User to be identified in photo is defined as the described hidden Markov model correspondence that similarity reaches pre-conditioned
Described targeted customer.
Preferably, described acquisition photo to be identified, and carry according to the rgb pixel value of described photo to be identified
Before taking the step of sequence vector to be observed, also include:
Obtain the reference photo of targeted customer；
The observation vector sequence that described targeted customer is corresponding is extracted according to the described rgb pixel value with reference to photo
Row；
The hidden Markov model corresponding with described targeted customer is set up according to described observation vector sequence.
Preferably, described determine step targeted customer being detected after, also include:
The hidden Ma Erke corresponding with the described targeted customer detected is updated according to described sequence vector to be observed
Husband's model.
Preferably, described determine step targeted customer being detected after, also include:
According to the hidden Markov mould that described sequence vector to be observed is corresponding with the described targeted customer detected
Type computation model parameter；
Moving direction and the distance of terminal is determined according to described model parameter；
Move according to terminal described in the moving direction determined and distance controlling, with to the described target detected
User is tracked.
Preferably, the step of the described reference photo obtaining targeted customer includes:
Gather the targeted customer's multiple reference photos in rotation process；Wherein, turn described targeted customer
During Dong, shoot a described reference photo every prefixed time interval.
Additionally, the object detecting device that the present invention provides includes:
Acquisition module, is used for obtaining photo to be identified；
Extraction module, extracts sequence vector to be observed for the rgb pixel value according to described photo to be identified；
Computing module, similar to described hidden Markov model for calculating described sequence vector to be observed
Degree；
Determine module, for when described similarity reaches pre-conditioned, determine and targeted customer detected,
And the user to be identified in described photo to be identified is defined as the described hidden horse that similarity reaches pre-conditioned
The described targeted customer that Er Kefu model is corresponding.
Preferably, described acquisition module is additionally operable to obtain the reference photo of targeted customer；
Described extraction module is additionally operable to extract described targeted customer according to the described rgb pixel value with reference to photo
Corresponding observation vector sequence；
Described object detecting device also includes setting up module, described set up module for according to described observation to
Amount sequence sets up the hidden Markov model corresponding with described targeted customer, with according to described hidden Markov
Model carries out target detection..
Preferably, described object detecting device also includes more new module, for according to described vector to be observed
Sequence updates the hidden Markov model corresponding with the described targeted customer detected.
Preferably, described computing module is additionally operable to according to described sequence vector to be observed with described in detecting
The hidden Markov model computation model parameter that targeted customer is corresponding；
Described determine that module is additionally operable to determine moving direction and the distance of terminal according to described model parameter；
Described object detecting device also includes tracking module, for according to the moving direction determined and distance control
Make described terminal to move, so that the described targeted customer detected to be tracked.
Preferably, described acquisition module is additionally operable to the multiple reference photographs gathering targeted customer in rotation process
Sheet；Wherein, in described targeted customer's rotation process, shoot a described ginseng every prefixed time interval
Examine photo.
The object detection method of present invention offer and device, by obtaining the reference photo of targeted customer, root
The observation vector sequence that described targeted customer is corresponding, and root is extracted according to the described rgb pixel value with reference to photo
The hidden Markov model corresponding with described targeted customer is set up, with according to institute according to described observation vector sequence
State hidden Markov model and carry out target detection, owing to have employed rgb pixel value in modeling process, and
Need not rely on face information, directly can i.e. may recognize that targeted customer according to rgb pixel value, knowing
Not and need not the photographic head of tracked user terminaloriented always during tracking, tracked user is permissible
Arbitrarily walking about, therefore terminal is more intelligent and convenient during target recognition.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of object detection method first embodiment of the present invention；
Fig. 2 is the high-level schematic functional block diagram of object detecting device first embodiment of the present invention；
Fig. 3 is the high-level schematic functional block diagram of object detecting device the second embodiment of the present invention；
Fig. 4 is the high-level schematic functional block diagram of object detecting device the 3rd embodiment of the present invention.
The realization of the object of the invention, functional characteristics and advantage will in conjunction with the embodiments, do referring to the drawings further
Explanation.
Detailed description of the invention
Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not used to limit
Determine the present invention.
The present invention provides a kind of object detection method, can realize based on terminal, optionally, and can be based on
Robot realizes.In the present embodiment and each embodiment following, say as a example by being applied to robot
Bright.With reference to the schematic flow sheet that Fig. 1, Fig. 1 are object detection method first embodiment of the present invention, the present invention
The object detection method proposed comprises the following steps:
Step S10, obtains photo to be identified, and extracts according to the rgb pixel value of described photo to be identified
Sequence vector to be observed；
In the present embodiment, for a fabric width be W, the photo to be identified of a height of H, can be from top to bottom
Order extracts the subimage block that some height are L, is used for generating sequence vector to be observed.It is to say,
Definition one W × L sample window, sample window sequential sampling from top to bottom, move down every time L away from
From, obtain some subimage blocks.Can directly using the rgb pixel value of subimage block as observed value.Also
First the rgb pixel value of subimage block can be done Karhunen-Loeve transformation, take the coefficient after its conversion as observed value.
Each subimage block of sampling from a photo to be identified is carried out Karhunen-Loeve transformation, has just obtained this and treated
Identify the sequence vector to be observed that photo is corresponding.
Step S20, calculates the hidden Markov that described sequence vector to be observed is corresponding with the targeted customer preset
The similarity of model；
In the present embodiment, the hidden Markov model that default targeted customer is corresponding can use with lower section
Formula is set up, and i.e. before step S10, also includes:
Obtain the reference photo of targeted customer；
The observation vector sequence that described targeted customer is corresponding is extracted according to the described rgb pixel value with reference to photo
Row；
The hidden Markov model corresponding with described targeted customer is set up according to described observation vector sequence.
In the present embodiment, reference photo can be the body photo of targeted customer, it is also possible to shine for face
Sheet.Targeted customer can be one, two or more.Reference photo corresponding to each targeted customer is also
Can be one, two or multiple.Optionally, in the present embodiment, the reference that each targeted customer is corresponding
Photo is multiple.Assume that the targeted customer pre-set in robot is two, respectively targeted customer A
With targeted customer B, each target correspondence obtains 50 with reference to photo.
Optionally, in order to improve the comprehensive of reference photo, the accuracy of target detection, step S10 are improved
Including: gather the targeted customer's multiple reference photos in rotation process；Wherein, described targeted customer
In rotation process, shoot a described reference photo every prefixed time interval and learn.
Optionally, can be with preset reference photo acquisition control, when user triggers with reference to photo acquisition control,
Then robot starts to gather with reference to photo.Optionally, receive user in robot to adopt based on reference to photo
During the acquisition instructions that collection control inputs, first export information.Can be with voice, word or the side of video
Formula exports, to point out user to rotate slowly in the front original place of photographic head of robot.Robot turns user
Dynamic period, a reference photo can be shot every prefixed time interval, for example, it is possible to every 0.1 second beats
Take the photograph one with reference to photo.
In the present embodiment, for a fabric width be W, the reference photo of a height of H, can be the most suitable
Sequence extracts the subimage block that some height are L, is used for generating observation vector sequence.It is to say, it is fixed
The sample window of one W × L of justice, sample window sequential sampling from top to bottom, move down the distance of L every time,
Obtain some subimage blocks.Can directly using the rgb pixel value of subimage block as observed value.Can also
First the rgb pixel value of subimage block is done Karhunen-Loeve transformation, take the coefficient after its conversion as observed value.Right
From a reference photo, each subimage block of sampling carries out Karhunen-Loeve transformation, has just obtained this reference and has shone
The observation vector sequence that sheet is corresponding.
In the present embodiment, each hidden Markov model can with the single width of same targeted customer or several
Image is trained.Training sequentially includes the following steps:
A general hidden Ma Erke is set up according to the observation vector sequence that the reference photo that will train is corresponding
Husband model λ=(A, B, π), determines the status number of model, the state transfer of permission and observation vector sequence
Size.By training data even partition, corresponding with N number of state, calculate the initial of hidden Markov model
Parameter, the transition probability matrix A between init state.Setting state i can only return to itself or transfer
To j=i+1 state.Initial state probabilities is distributed, it is assumed that hidden Markov model is from first state
Start.For observing probability matrix B, it is assumed that
Therefore, a hidden Markov model λ=(A, B, π) is initially dissolved.
After initial model determines, utilize Baum-Welch revaluation algorithm that initial hidden Markov model is carried out
Recalculate.The parameters of hidden Markov model will reappraise, and obtain a new model:
λ=(A', B', π ')
Then Forward-backward algorithm or Viterbi algorithm is utilized to calculate observed value sequence O at this model
Under P (O | λ ').In order to estimate the model closest to observed value sequence O, set threshold value C, order
P (O | λ) converge on C, obtain the hidden Markov model trained.Therefore, targeted customer has i.e. been obtained
Corresponding hidden Markov model.Can utilize this hidden Markov model targeted customer is identified and
Follow the tracks of.Such as, for above-mentioned targeted customer A and targeted customer B, it will set up respectively and targeted customer
The HMM that A is corresponding with targeted customer B, and be pre-stored within robot.
In this example, it is assumed that it is corresponding with targeted customer B to have prestored targeted customer A in robot
HMM.Target detection control can also be set in robot, trigger this target user
During detection control, then robot initially enters target detection mode.User can also specify user to be identified
For targeted customer A or targeted customer is B.Such as, user is triggering after target detection control, robot
The targeted customer A information corresponding with targeted customer B can be shown, select for user current to be identified
User.Such as, user can trigger the control that targeted customer A is corresponding, then by user setup to be identified be
Targeted customer A, therefore, robot will detect current use to be identified during follow-up identification and/or tracking
Whether family is targeted customer A.
After entering target detection mode, then robot begins through camera collection photo to be identified.Can
With 1 photo of middle collection per second, or gather 5 photos each second, specifically can be according to actual needs
It is configured.
In the present embodiment, can calculate described to be observed by Forward-backward algorithm or Viterbi algorithm
Sequence vector and the similarity of described hidden Markov model.Similarity reflect sequence vector to be observed with
The similarity degree of the hidden Markov model in robot.
As a example by user to be identified is for targeted customer A, the most described similarity reflects sequence vector to be observed
With the similarity degree between corresponding for the targeted customer A hidden Markov model prestored in robot.
Step S30, when described similarity reaches pre-conditioned, determines and targeted customer detected, and by institute
State the user to be identified in photo to be identified and be defined as the described hidden Markov that similarity reaches pre-conditioned
The described targeted customer that model is corresponding.
In the present embodiment, when similarity is sufficiently high, such as, when similarity is higher than preset value, then recognize
Reach pre-conditioned for similarity, then it is assumed that targeted customer A detected, the most then think described photograph to be identified
Sheet detects targeted customer A.
The present embodiment is during target detection, owing to have employed rgb pixel value, directly during detection
Connect the rgb pixel value according to photo to be identified the most recognizable with the similarity of the hidden Markov model preset
Go out targeted customer.And need not rely on face information, can be directly the most recognizable according to rgb pixel value
Go out targeted customer, during identifying and following the tracks of, need not the photographic head of tracked user terminaloriented always,
Tracked user can arbitrarily walk about, and therefore terminal is more intelligent and convenient during target recognition.
Further, first embodiment based on object detection method of the present invention, the invention allows for mesh
Mark detection method the second embodiment, described determine step targeted customer being detected after, described target
Detection method also includes: update and the described targeted customer couple detected according to described sequence vector to be observed
The hidden Markov model answered.
In the present embodiment, it is referred in above-mentioned object detection method first embodiment train hidden Ma Erke
The method of husband's model, sets up hidden Markov model according to sequence vector to be observed, does not repeats them here.
In the present embodiment, during follow-up tracking targeted customer, if having recognized targeted customer,
Then the hidden Markov model of this targeted customer is updated such that it is able to such that this targeted customer's is hidden
Markov model is more accurate, further increases the accuracy of target detection.
Further, first embodiment based on object detection method of the present invention, the invention allows for mesh
Mark detection method the 3rd embodiment, described determine step targeted customer being detected after, described target
Detection method also includes:
According to the hidden Markov mould that described sequence vector to be observed is corresponding with the described targeted customer detected
Type computation model parameter；
Moving direction and the distance of terminal is determined according to described model parameter；
Move according to terminal described in the moving direction determined and distance controlling, with to the described target detected
User is tracked.
In the present embodiment, model parameter can represent with above-mentioned λ, can pass through formula λ=(A, B, π) and enter
Row calculates, and does not repeats them here.It can be assumed for instance that λ is the biggest, then the targeted customer detected current with
Distance between robot is the nearest；If λ is the least, then the targeted customer detected is current and between robot
Distance is the most remote.Can be when λ value be less than the first predetermined threshold value, then it is assumed that targeted customer is current and robot it
Between hypertelorism, it is therefore desirable to control robot and move near the direction of targeted customer, the most permissible
Control robot forwards to move.Or can also be according to the targeted customer in the photo to be identified detected
The moving direction of position control robot, such as can move to dead ahead, left front or right front,
So that robot moves towards the physical location of targeted customer, and then near targeted customer.Or also may be used
With when λ value is more than the second predetermined threshold value, then it is assumed that targeted customer's distance mistake currently and between robot
Closely, it is therefore desirable to control robot and move to the direction of wide user, such as, can control robot
Rearward move.Or can also be according to the position control of the targeted customer in the photo to be identified detected
The moving direction of robot, such as can to dead astern, left back or right back move, so that machine
People moves to the physical location deviating from targeted customer, and then wide user such that it is able to more accurately
Targeted customer is tracked.The first corresponding for λ predetermined threshold value and the second predetermined threshold value can be according to subscriber frames
Depending on the scope of fixed target object, the range size of the target object confined can affect threshold values district corresponding for λ
Between scope.
The present embodiment is during being tracked targeted customer, it is not necessary to rely on the face of targeted customer
Information is tracked such that it is able to convenient and accurately corresponding targeted customer be tracked.
The present invention further provides a kind of object detecting device.Can realize based on terminal, optionally, can
To realize based on robot.In the present embodiment and each embodiment following, as a example by being applied to robot
Illustrate.
With reference to the high-level schematic functional block diagram that Fig. 2, Fig. 2 are object detecting device first embodiment of the present invention, this
The object detecting device that invention provides includes:
Acquisition module 10, is used for obtaining photo to be identified；
Extraction module 20, extracts vector sequence to be observed for the rgb pixel value according to described photo to be identified
Row；
In the present embodiment, for a fabric width be W, the photo to be identified of a height of H, can be from top to bottom
Order extracts the subimage block that some height are L, is used for generating sequence vector to be observed.It is to say,
Definition one W × L sample window, sample window sequential sampling from top to bottom, move down every time L away from
From, obtain some subimage blocks.Can directly using the rgb pixel value of subimage block as observed value.Also
First the rgb pixel value of subimage block can be done Karhunen-Loeve transformation, take the coefficient after its conversion as observed value.
Each subimage block of sampling from a photo to be identified is carried out Karhunen-Loeve transformation, has just obtained this and treated
Identify the sequence vector to be observed that photo is corresponding.
Computing module 30, for calculating the described sequence vector to be observed phase with described hidden Markov model
Like degree；
In the present embodiment, the hidden Markov model that default targeted customer is corresponding can use with lower section
Formula is set up, it may be assumed that
Described acquisition module 10 is additionally operable to obtain the reference photo of targeted customer；
Described extraction module 20 is additionally operable to extract described target according to the described rgb pixel value with reference to photo
The observation vector sequence that user is corresponding；
Described object detecting device also includes setting up module, described set up module for according to described observation to
Amount sequence sets up the hidden Markov model corresponding with described targeted customer.
In the present embodiment, reference photo can be the body photo of targeted customer, it is also possible to shine for face
Sheet.Targeted customer can be one, two or more.Reference photo corresponding to each targeted customer is also
Can be one, two or multiple.Optionally, in the present embodiment, the reference that each targeted customer is corresponding
Photo is multiple.Assume that the targeted customer pre-set in robot is two, respectively targeted customer A
With targeted customer B, each target correspondence obtains 50 with reference to photo.
Optionally, in order to improve the comprehensive of reference photo, the accuracy of target detection, step S10 are improved
Including: gather the targeted customer's multiple reference photos in rotation process；Wherein, described targeted customer
In rotation process, shoot a described reference photo every prefixed time interval and learn.
Optionally, can be with preset reference photo acquisition control, when user triggers with reference to photo acquisition control,
Then robot starts to gather with reference to photo.Optionally, receive user in robot to adopt based on reference to photo
During the acquisition instructions that collection control inputs, first export information.Can be with voice, word or the side of video
Formula exports, to point out user to rotate slowly in the front original place of photographic head of robot.Robot turns user
Dynamic period, a reference photo can be shot every prefixed time interval, for example, it is possible to every 0.1 second beats
Take the photograph one with reference to photo.
In the present embodiment, for a fabric width be W, the reference photo of a height of H, can be the most suitable
Sequence extracts the subimage block that some height are L, is used for generating observation vector sequence.It is to say, it is fixed
The sample window of one W × L of justice, sample window sequential sampling from top to bottom, move down the distance of L every time,
Obtain some subimage blocks.Can directly using the rgb pixel value of subimage block as observed value.Can also
First the rgb pixel value of subimage block is done Karhunen-Loeve transformation, take the coefficient after its conversion as observed value.Right
From a reference photo, each subimage block of sampling carries out Karhunen-Loeve transformation, has just obtained this reference and has shone
The observation vector sequence that sheet is corresponding.
In the present embodiment, each hidden Markov model can with the single width of same targeted customer or several
Image is trained.Training sequentially includes the following steps:
A general hidden Ma Erke is set up according to the observation vector sequence that the reference photo that will train is corresponding
Husband model λ=(A, B, π), determines the status number of model, the state transfer of permission and observation vector sequence
Size.By training data even partition, corresponding with N number of state, calculate the initial of hidden Markov model
Parameter, the transition probability matrix A between init state.Setting state i can only return to itself or transfer
To j=i+1 state.Initial state probabilities is distributed, it is assumed that hidden Markov model is from first state
Start.For observing probability matrix B, it is assumed that
Therefore, a hidden Markov model λ=(A, B, π) is initially dissolved.
After initial model determines, utilize Baum-Welch revaluation algorithm that initial hidden Markov model is carried out
Recalculate.The parameters of hidden Markov model will reappraise, and obtain a new model:
λ=(A', B', π ')
Then Forward-backward algorithm or Viterbi algorithm is utilized to calculate observed value sequence O at this model
Under P (O | λ ').In order to estimate the model closest to observed value sequence O, set threshold value C, order
P (O | λ) converge on C, obtain the hidden Markov model trained.Therefore, targeted customer has i.e. been obtained
Corresponding hidden Markov model.Can utilize this hidden Markov model targeted customer is identified and
Follow the tracks of.Such as, for above-mentioned targeted customer A and targeted customer B, it will set up respectively and targeted customer
The HMM that A is corresponding with targeted customer B, and be pre-stored within robot.
In this example, it is assumed that it is corresponding with targeted customer B to have prestored targeted customer A in robot
HMM.Target detection control can also be set in robot, trigger this target user
During detection control, then robot initially enters target detection mode.User can also specify user to be identified
For targeted customer A or targeted customer is B.Such as, user is triggering after target detection control, robot
The targeted customer A information corresponding with targeted customer B can be shown, select for user current to be identified
User.Such as, user can trigger the control that targeted customer A is corresponding, then by user setup to be identified be
Targeted customer A, therefore, robot will detect current use to be identified during follow-up identification and/or tracking
Whether family is targeted customer A.
After entering target detection mode, then robot begins through camera collection photo to be identified.Can
With 1 photo of middle collection per second, or gather 5 photos each second, specifically can be according to actual needs
It is configured.
In the present embodiment, can calculate described to be observed by Forward-backward algorithm or Viterbi algorithm
Sequence vector and the similarity of described hidden Markov model.Similarity reflect sequence vector to be observed with
The similarity degree of the hidden Markov model in robot.
As a example by user to be identified is for targeted customer A, the most described similarity reflects sequence vector to be observed
With the similarity degree between corresponding for the targeted customer A hidden Markov model prestored in robot.
Determine module 40, for when described similarity reaches pre-conditioned, determine and targeted customer detected,
And the user to be identified in described photo to be identified is defined as the described hidden horse that similarity reaches pre-conditioned
The described targeted customer that Er Kefu model is corresponding.
In the present embodiment, when similarity is sufficiently high, such as, when similarity is higher than preset value, then recognize
Reach pre-conditioned for similarity, then it is assumed that targeted customer A detected, the most then think described photograph to be identified
Sheet detects targeted customer A.
The present embodiment is during target detection, owing to have employed rgb pixel value, directly during detection
Connect the rgb pixel value according to photo to be identified the most recognizable with the similarity of the hidden Markov model preset
Go out targeted customer.And need not rely on face information, can be directly the most recognizable according to rgb pixel value
Go out targeted customer, during identifying and following the tracks of, need not the photographic head of tracked user terminaloriented always,
Tracked user can arbitrarily walk about, and therefore terminal is more intelligent and convenient during target recognition.
Further, first embodiment based on object detecting device of the present invention, the invention allows for mesh
Second embodiment of mark detection device, is object detecting device the second embodiment of the present invention with reference to Fig. 3, Fig. 3
High-level schematic functional block diagram, described object detecting device also includes more new module 50, for according to described in treat
Observation vector sequence updates the hidden Markov model corresponding with the described targeted customer detected.
In the present embodiment, it is referred in above-mentioned object detecting device first embodiment train hidden Ma Erke
The method of husband's model, sets up hidden Markov model according to sequence vector to be observed, does not repeats them here.
In the present embodiment, during follow-up tracking targeted customer, if having recognized targeted customer,
Then the hidden Markov model of this targeted customer is updated such that it is able to such that this targeted customer's is hidden
Markov model is more accurate, further increases the accuracy of target detection.
Further, first embodiment based on object detecting device of the present invention, the invention allows for mesh
3rd embodiment of mark detection device, is object detecting device the 3rd embodiment of the present invention with reference to Fig. 4, Fig. 4
High-level schematic functional block diagram, described computing module 30 is additionally operable to according to described sequence vector to be observed and detection
The hidden Markov model computation model parameter that the described targeted customer that arrives is corresponding；
Described determine that module 40 is additionally operable to determine moving direction and the distance of terminal according to described model parameter；
Described object detecting device also includes tracking module 60, for according to the moving direction determined and distance
Control described terminal to move, so that the described targeted customer detected to be tracked.
In the present embodiment, model parameter can represent with above-mentioned λ, can pass through formula λ=(A, B, π) and enter
Row calculates, and does not repeats them here.It can be assumed for instance that λ is the biggest, then the targeted customer detected current with
Distance between robot is the nearest；If λ is the least, then the targeted customer detected is current and between robot
Distance is the most remote.Can be when λ value be less than the first predetermined threshold value, then it is assumed that targeted customer is current and robot it
Between hypertelorism, it is therefore desirable to control robot and move near the direction of targeted customer, the most permissible
Control robot forwards to move.Or can also be according to the targeted customer in the photo to be identified detected
The moving direction of position control robot, such as can move to dead ahead, left front or right front,
So that robot moves towards the physical location of targeted customer, and then near targeted customer.Or also may be used
With when λ value is more than the second predetermined threshold value, then it is assumed that targeted customer's distance mistake currently and between robot
Closely, it is therefore desirable to control robot and move to the direction of wide user, such as, can control robot
Rearward move.Or can also be according to the position control of the targeted customer in the photo to be identified detected
The moving direction of robot, such as can to dead astern, left back or right back move, so that machine
People moves to the physical location deviating from targeted customer, and then wide user such that it is able to more accurately
Targeted customer is tracked.The first corresponding for λ predetermined threshold value and the second predetermined threshold value can be according to subscriber frames
Depending on the scope of fixed target object, the range size of the target object confined can affect threshold values district corresponding for λ
Between scope.
The present embodiment is during being tracked targeted customer, it is not necessary to rely on the face of targeted customer
Information is tracked such that it is able to convenient and accurately corresponding targeted customer be tracked.
It should be noted that in this article, term " include ", " comprising " or its any other variant
Be intended to comprising of nonexcludability so that include the process of a series of key element, method, article or
Person's device not only includes those key elements, but also includes other key elements being not expressly set out, or also
Including the key element intrinsic for this process, method, article or device.In the feelings not having more restriction
Under condition, statement " including ... " key element limited, it is not excluded that include this key element process,
Method, article or device there is also other identical element.
The invention described above embodiment sequence number, just to describing, does not represent the quality of embodiment.
Through the above description of the embodiments, those skilled in the art is it can be understood that arrive above-mentioned
Embodiment method can add the mode of required general hardware platform by software and realize, naturally it is also possible to logical
Cross hardware, but a lot of in the case of the former is more preferably embodiment.Based on such understanding, the present invention's
The part that prior art is contributed by technical scheme the most in other words can be with the form body of software product
Revealing to come, this computer software product is stored in a storage medium (such as ROM/RAM, magnetic disc, light
Dish) in, including some instructions with so that a station terminal equipment (can be mobile phone, computer, service
Device, air-conditioner, or the network equipment etc.) perform the method described in each embodiment of the present invention.
It addition, the description relating to " first ", " second " etc. in invention is not only used for describing purpose, and not
It is understood that as instruction or implies its relative importance or the implicit quantity indicating indicated technical characteristic.
Thus, define " first ", the feature of " second " can express or implicitly include at least one should
Feature.It addition, the technical scheme between each embodiment can be combined with each other, but must be with ability
Based on territory those of ordinary skill is capable of, when the combination of technical scheme occurs conflicting or cannot be real
People should think that the combination of this technical scheme does not exists, the most not at the protection domain of application claims now
Within.
These are only the preferred embodiments of the present invention, not thereby limit the scope of the claims of the present invention, every
Utilize equivalent structure or equivalence flow process conversion that description of the invention and accompanying drawing content made, or directly or
Connect and be used in other relevant technical fields, be the most in like manner included in the scope of patent protection of the present invention.
Claims (10)
1. an object detection method, it is characterised in that described object detection method comprises the following steps:
Obtain photo to be identified, and extract vector to be observed according to the rgb pixel value of described photo to be identified
Sequence；
Calculate the phase of the described sequence vector to be observed hidden Markov model corresponding with the targeted customer preset
Like degree；
When described similarity reaches pre-conditioned, determine and targeted customer detected, and by described to be identified
User to be identified in photo is defined as the described hidden Markov model correspondence that similarity reaches pre-conditioned
Described targeted customer.
2. object detection method as claimed in claim 1, it is characterised in that described acquisition photograph to be identified
Sheet, and before extract the step of sequence vector to be observed according to the rgb pixel value of described photo to be identified,
Also include:
Obtain the reference photo of targeted customer；
The observation vector sequence that described targeted customer is corresponding is extracted according to the described rgb pixel value with reference to photo
Row；
The hidden Markov model corresponding with described targeted customer is set up according to described observation vector sequence.
3. object detection method as claimed in claim 1, it is characterised in that described determine mesh detected
After the step of mark user, also include:
The hidden Ma Erke corresponding with the described targeted customer detected is updated according to described sequence vector to be observed
Husband's model.
4. object detection method as claimed in claim 1, it is characterised in that described determine mesh detected
After the step of mark user, also include:
According to the hidden Markov mould that described sequence vector to be observed is corresponding with the described targeted customer detected
Type computation model parameter；
Moving direction and the distance of terminal is determined according to described model parameter；
Move according to terminal described in the moving direction determined and distance controlling, with to the described target detected
User is tracked.
5. the object detection method as described in any one of Claims 1-4, it is characterised in that described in obtain
The step of the reference photo taking targeted customer includes:
Gather the targeted customer's multiple reference photos in rotation process；Wherein, turn described targeted customer
During Dong, shoot a described reference photo every prefixed time interval.
6. an object detecting device, it is characterised in that described object detecting device includes:
Acquisition module, is used for obtaining photo to be identified；
Extraction module, extracts sequence vector to be observed for the rgb pixel value according to described photo to be identified；
Computing module, similar to described hidden Markov model for calculating described sequence vector to be observed
Degree；
Determine module, for when described similarity reaches pre-conditioned, determine and targeted customer detected,
And the user to be identified in described photo to be identified is defined as the described hidden horse that similarity reaches pre-conditioned
The described targeted customer that Er Kefu model is corresponding.
7. object detecting device as claimed in claim 6, it is characterised in that described acquisition module is also used
In the reference photo obtaining targeted customer；
Described extraction module is additionally operable to extract described targeted customer according to the described rgb pixel value with reference to photo
Corresponding observation vector sequence；
Described object detecting device also includes setting up module, described set up module for according to described observation to
Amount sequence sets up the hidden Markov model corresponding with described targeted customer, with according to described hidden Markov
Model carries out target detection.
8. object detecting device as claimed in claim 6, it is characterised in that described object detecting device
Also include more new module, use with the described target detected for updating according to described sequence vector to be observed
The hidden Markov model that family is corresponding.
9. object detecting device as claimed in claim 6, it is characterised in that described computing module is also used
In the hidden Markov model corresponding with the described targeted customer detected according to described sequence vector to be observed
Computation model parameter；
Described determine that module is additionally operable to determine moving direction and the distance of terminal according to described model parameter；
Described object detecting device also includes tracking module, for according to the moving direction determined and distance control
Make described terminal to move, so that the described targeted customer detected to be tracked.
10. the object detecting device as described in any one of claim 6 to 9, it is characterised in that described
Acquisition module is additionally operable to the multiple reference photos gathering targeted customer in rotation process；Wherein, described
In targeted customer's rotation process, shoot a described reference photo every prefixed time interval.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610280638.4A CN105956551B (en) | 2016-04-28 | 2016-04-28 | Object detection method and device |
| PCT/CN2017/080833 WO2017186017A1 (en) | 2016-04-28 | 2017-04-18 | Target detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610280638.4A CN105956551B (en) | 2016-04-28 | 2016-04-28 | Object detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105956551A true CN105956551A (en) | 2016-09-21 |
| CN105956551B CN105956551B (en) | 2018-01-30 |
Family
ID=56916909
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610280638.4A Expired - Fee Related CN105956551B (en) | 2016-04-28 | 2016-04-28 | Object detection method and device |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN105956551B (en) |
| WO (1) | WO2017186017A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017186017A1 (en) * | 2016-04-28 | 2017-11-02 | 深圳市鼎盛智能科技有限公司 | Target detection method and device |
| CN109839614A (en) * | 2018-12-29 | 2019-06-04 | 深圳市天彦通信股份有限公司 | The positioning system and method for fixed acquisition equipment |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107718014A (en) * | 2017-11-09 | 2018-02-23 | 深圳市小村机器人智能科技有限公司 | Highly emulated robot head construction and its method of controlling operation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101227506A (en) * | 2007-01-01 | 2008-07-23 | 华为技术有限公司 | Equipment, system and method for identifying subscriber terminal |
| CN103593680A (en) * | 2013-11-19 | 2014-02-19 | 南京大学 | Dynamic hand gesture recognition method based on self incremental learning of hidden Markov model |
| CN103761748A (en) * | 2013-12-31 | 2014-04-30 | 北京邮电大学 | Method and device for detecting abnormal behaviors |
| CN104112122A (en) * | 2014-07-07 | 2014-10-22 | 叶茂 | Vehicle logo automatic identification method based on traffic video |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5886616B2 (en) * | 2011-11-30 | 2016-03-16 | キヤノン株式会社 | Object detection apparatus, method for controlling object detection apparatus, and program |
| CN102592112B (en) * | 2011-12-20 | 2014-01-29 | 四川长虹电器股份有限公司 | Method for determining gesture moving direction based on hidden Markov model |
| CN103489001B (en) * | 2013-09-25 | 2017-01-11 | 杭州智诺科技股份有限公司 | Method and device for tracing picture target |
| CN105956551B (en) * | 2016-04-28 | 2018-01-30 | 深圳市鼎盛智能科技有限公司 | Object detection method and device |
-
2016
- 2016-04-28 CN CN201610280638.4A patent/CN105956551B/en not_active Expired - Fee Related
-
2017
- 2017-04-18 WO PCT/CN2017/080833 patent/WO2017186017A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101227506A (en) * | 2007-01-01 | 2008-07-23 | 华为技术有限公司 | Equipment, system and method for identifying subscriber terminal |
| CN103593680A (en) * | 2013-11-19 | 2014-02-19 | 南京大学 | Dynamic hand gesture recognition method based on self incremental learning of hidden Markov model |
| CN103761748A (en) * | 2013-12-31 | 2014-04-30 | 北京邮电大学 | Method and device for detecting abnormal behaviors |
| CN104112122A (en) * | 2014-07-07 | 2014-10-22 | 叶茂 | Vehicle logo automatic identification method based on traffic video |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017186017A1 (en) * | 2016-04-28 | 2017-11-02 | 深圳市鼎盛智能科技有限公司 | Target detection method and device |
| CN109839614A (en) * | 2018-12-29 | 2019-06-04 | 深圳市天彦通信股份有限公司 | The positioning system and method for fixed acquisition equipment |
| CN109839614B (en) * | 2018-12-29 | 2020-11-06 | 深圳市天彦通信股份有限公司 | Positioning system and method of fixed acquisition equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2017186017A1 (en) | 2017-11-02 |
| CN105956551B (en) | 2018-01-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107330396B (en) | Pedestrian re-identification method based on multi-attribute and multi-strategy fusion learning | |
| CN103971386B (en) | A kind of foreground detection method under dynamic background scene | |
| CN110070029B (en) | Gait recognition method and device | |
| CN105678253B (en) | Semi-supervised face age estimation device and semi-supervised face age estimation method | |
| CN109934127B (en) | Pedestrian identification and tracking method based on video image and wireless signal | |
| CN107862300A (en) | A kind of descending humanized recognition methods of monitoring scene based on convolutional neural networks | |
| CN109657716A (en) | A kind of vehicle appearance damnification recognition method based on deep learning | |
| CN110378219B (en) | Living body detection method, living body detection device, electronic equipment and readable storage medium | |
| CN111241932A (en) | Automobile exhibition room passenger flow detection and analysis system, method and storage medium | |
| KR102225613B1 (en) | Person re-identification apparatus and method | |
| CN110991397B (en) | Travel direction determining method and related equipment | |
| CN111027555B (en) | License plate recognition method and device and electronic equipment | |
| CN103150552B (en) | A kind of driving training management method based on number of people counting | |
| CN108460340A (en) | A kind of gait recognition method based on the dense convolutional neural networks of 3D | |
| CN115713715B (en) | Human behavior recognition method and recognition system based on deep learning | |
| CN110674680B (en) | Living body identification method, living body identification device and storage medium | |
| CN105956551A (en) | Target detection method and device | |
| CN109389156A (en) | A kind of training method, device and the image position method of framing model | |
| CN111582358A (en) | Training method and device for house type recognition model and house type weight judging method and device | |
| CN114241557A (en) | Image recognition method, device and equipment, intelligent door lock and medium | |
| WO2021022795A1 (en) | Method, apparatus, and device for detecting fraudulent behavior during facial recognition process | |
| CN110992500A (en) | Attendance checking method and device, storage medium and server | |
| CN109711232A (en) | Deep learning pedestrian recognition methods again based on multiple objective function | |
| CN115953831A (en) | Abnormal behavior monitoring method and system for interrogation scene based on ST-Transformer network | |
| CN112508135B (en) | Model training method, pedestrian attribute prediction method, device and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Sheng Ge Inventor after: Sheng Zhangmei Inventor before: Xu Yongchang Inventor before: Sheng Ge |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PP01 | Preservation of patent right |
Effective date of registration: 20210316 Granted publication date: 20180130 |
|
| PP01 | Preservation of patent right | ||
| PD01 | Discharge of preservation of patent |
Date of cancellation: 20240316 Granted publication date: 20180130 |
|
| PD01 | Discharge of preservation of patent | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180130 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |