CN110490058A - Training method, device, system and the computer-readable medium of pedestrian detection model - Google Patents
Training method, device, system and the computer-readable medium of pedestrian detection model Download PDFInfo
- Publication number
- CN110490058A CN110490058A CN201910615436.4A CN201910615436A CN110490058A CN 110490058 A CN110490058 A CN 110490058A CN 201910615436 A CN201910615436 A CN 201910615436A CN 110490058 A CN110490058 A CN 110490058A
- Authority
- CN
- China
- Prior art keywords
- detection block
- training
- error
- pedestrian
- classification
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Abstract
The present invention provides training method, device, system and the computer-readable mediums of a kind of pedestrian detection model, the training method includes: that training image is input to neural network, to generate the predictive information about the target object in the training image, the predictive information includes detection block position, detection block weight and detection block score, wherein, the detection block weight indicates the similitude of target object and background in the detection block, and similitude is higher, then detection block weight is lower；The first error in classification between the detection block score and score true value is calculated, and weighting classification error is calculated according to the detection block weight and first error in classification；Network parameter at least based on neural network described in the weighting classification error update.The detection block weight that the present invention reduces pedestrian sample similar with background automatically in the training process of pedestrian's detection model greatly improves the precision of pedestrian detection to reduce the adverse effect for easily obscuring sample to network parameter.
Description
Technical field
The present invention relates to pedestrian detection technology fields, relate more specifically to training method, the dress of a kind of pedestrian detection model
It sets, system and computer-readable medium.
Background technique
Pedestrian detection has a wide range of applications in the fields such as security protection and automatic Pilot, and the purpose is to the handles from image or video
The position of pedestrian, which searches out, to be come.Pedestrian detection is the basis of a lot of other visual tasks, for example, pedestrian identifies again, pedestrian tracking and
Pedestrian's action recognition etc..Due to there is many and quite similar background foreign matter of pedestrian's appearance in pedestrian's scene, pedestrian is caused to examine
Examining system will appear the situation of error detection in these scenes, reduce the accuracy of pedestrian detection.In security protection or automatically
In Driving Scene, the testing result of mistake will lead to serious consequence, thus need more accurate detection system, with reduce with
The similar foreign matter of pedestrian is to interference caused by pedestrian detection.
Summary of the invention
To solve the above-mentioned problems, the invention proposes a kind of training sides of pedestrian detection model based on weight self-regulation
Case.The training program proposed by the present invention about pedestrian detection model is briefly described below, more details will be attached in subsequent combination
Figure is described in a specific embodiment.
According to embodiments of the present invention on the one hand, a kind of training method of pedestrian detection model is provided, which comprises
Training image is input to neural network, to generate the prediction letter about the target object in the training image
Breath, the predictive information includes detection block position, detection block weight and detection block score, wherein the detection block weight indicates
The similitude of target object and background in the detection block, the similitude is higher, then the detection block weight is lower；
Calculate the first error in classification between the detection block score and score true value, and according to the detection block weight and
First error in classification calculates weighting classification error；
Network parameter at least based on neural network described in the weighting classification error update.
In one embodiment, the weighting classification error is multiplying for the detection block weight and first error in classification
Product.
In one embodiment, the method also includes: calculate second between the detection block weight and weight true value
Error in classification；And the network parameter is updated based on second error in classification.
In one embodiment, the method also includes: calculate the position between the detection block position and position true value
Error；And the network parameter is updated based on the location error.
In one embodiment, the predictive information for generating the training objective in the training image includes: based on institute
It states neural network and feature extraction is carried out to the training image, to generate the characteristic pattern of the training image；According to the feature
Figure generates the predictive information.
According to embodiments of the present invention on the other hand, a kind of training device of pedestrian detection model, pedestrian's inspection are provided
Survey model training device include:
Prediction module, for training image to be input to neural network, to generate about the target in the training image
The predictive information of object, the predictive information include detection block position, detection block weight and detection block score, wherein the inspection
Surveying frame weight indicates the similitude of target object and background in the detection block, and the similitude is higher, then the detection block power
It is again lower；
Error calculating module, for calculating the first error in classification between the detection block score and score true value, and root
Weighting classification error is calculated according to the detection block weight and first error in classification；And
Training module, for being at least based on the weighting classification error update network parameter.
In one embodiment, the weighting classification error is multiplying for the detection block weight and first error in classification
Product.
In one embodiment, described device further include: characteristic extracting module, for being based on the neural network to described
Training image carries out feature extraction, to generate the characteristic pattern of the training image；Also, the prediction module is according to the feature
Figure generates the predictive information.
According to embodiments of the present invention in another aspect, providing a kind of training system of pedestrian detection model, pedestrian's inspection
The training system for surveying model includes storage device and processor, and the meter run by the processor is stored on the storage device
Calculation machine program, the computer program execute pedestrian detection model described in any of the above embodiments when being run by the processor
Training method.
Another aspect according to embodiments of the present invention, provides a kind of computer-readable medium, the computer-readable medium
On be stored with computer program, the computer program executes the instruction of pedestrian detection model described in any of the above embodiments at runtime
Practice method.
Training method, device, system and the computer-readable medium of the pedestrian detection model of the embodiment of the present invention are in pedestrian
Automatically the detection block weight of pedestrian sample similar with background is reduced in the training process of detection model, so as to avoid easily obscuring
Sample greatly improves the precision of pedestrian detection to the adverse effect of network parameter.
Detailed description of the invention
The embodiment of the present invention is described in more detail in conjunction with the accompanying drawings, the above and other purposes of the present invention,
Feature and advantage will be apparent.Attached drawing is used to provide to further understand the embodiment of the present invention, and constitutes explanation
A part of book, is used to explain the present invention together with the embodiment of the present invention, is not construed as limiting the invention.In the accompanying drawings,
Identical reference label typically represents same parts or step.
Fig. 1 shows for realizing training method, device, system and the meter of pedestrian detection model according to an embodiment of the present invention
The schematic block diagram of the exemplary electronic device of calculation machine readable medium；
Fig. 2 shows the schematic flow charts of the training method of pedestrian detection model according to an embodiment of the present invention；
Fig. 3 shows the frame diagram of the neural network of the training method of pedestrian detection model according to an embodiment of the present invention；
Fig. 4 shows the schematic block diagram of the training device of pedestrian detection model according to an embodiment of the present invention；And
Fig. 5 shows the schematic block diagram of the training system of pedestrian detection model according to an embodiment of the present invention.
Specific embodiment
In order to enable the object, technical solutions and advantages of the present invention become apparent, root is described in detail below with reference to accompanying drawings
According to example embodiments of the present invention.Obviously, described embodiment is only a part of the embodiments of the present invention, rather than this hair
Bright whole embodiments, it should be appreciated that the present invention is not limited by example embodiment described herein.Based on described in the present invention
The embodiment of the present invention, those skilled in the art's obtained all other embodiment in the case where not making the creative labor
It should all fall under the scope of the present invention.
Firstly, referring to Fig.1 come describe the training method of the pedestrian detection model for realizing the embodiment of the present invention, device,
System and the exemplary electronic device of computer-readable medium 100.
As shown in Figure 1, electronic equipment 100 include one or more processors 102, it is one or more storage device 104, defeated
Enter device 106, output device 108 and imaging sensor 110, these components pass through bus system 112 and/or other forms
The interconnection of bindiny mechanism's (not shown).It should be noted that the component and structure of electronic equipment 100 shown in FIG. 1 are only exemplary, and
Unrestricted, as needed, the electronic equipment also can have other assemblies and structure.
The processor 102 can be central processing unit (CPU) or have data-handling capacity and/or instruction execution
The processing unit of the other forms of ability, and the other components that can control in the electronic equipment 100 are desired to execute
Function.
The storage device 104 may include one or more computer program products, and the computer program product can
To include various forms of computer readable storage mediums, such as volatile memory and/or nonvolatile memory.It is described easy
The property lost memory for example may include random access memory (RAM) and/or cache memory (cache) etc..It is described non-
Volatile memory for example may include read-only memory (ROM), hard disk, flash memory etc..In the computer readable storage medium
On can store one or more computer program instructions, processor 102 can run described program instruction, to realize hereafter institute
The client functionality (realized by processor) in the embodiment of the present invention stated and/or other desired functions.In the meter
Can also store various application programs and various data in calculation machine readable storage medium storing program for executing, for example, the application program use and/or
The various data etc. generated.
The input unit 106 can be the device that user is used to input instruction, and may include keyboard, mouse, wheat
One or more of gram wind and touch screen etc..
The output device 108 can export various information (such as image or sound) to external (such as user), and
It may include one or more of display, loudspeaker etc..
Described image sensor 110 can be shot the desired image of user (such as photo, video etc.), and will be captured
Image be stored in the storage device 104 for other components use.
When note that the component and structure of electronic equipment shown in FIG. 1 100 are only exemplary, although electronics shown in fig. 1
Equipment 100 include multiple and different devices, but as needed, some of which device can not be it is necessary, therein one
The quantity of a little devices can be more etc., and the present invention does not limit this.
Illustratively, the training method for realizing pedestrian detection model according to an embodiment of the present invention, recognition methods, dress
It sets and the exemplary electronic device of processing equipment may be implemented as the intelligent terminals such as smart phone, tablet computer, computer.
In the following, reference Fig. 2 to be described to the training method 200 of pedestrian detection model according to an embodiment of the present invention.
As shown in Fig. 2, training image is input to neural network in step S210, the instruction in the training image is obtained
Practice the predictive information of target, the predictive information includes detection block position, detection block weight and detection block score, wherein described
Detection block weight indicates the similitude of target object and background in the detection block, and the similitude is higher, then the detection block
Weight is lower.Conversely, similitude is lower, then the detection block weight is higher.
Wherein, the training image can be any candid photograph image comprising many pedestrians.Training set can be constructed in advance,
Include several training images in the training set, all includes one or more pedestrians, the position of pedestrian on usual every width training image
It sets and is marked in advance with bounding box, referred to as true frame.Based on the mark carried out in advance, each detection block can be obtained
Position true value, score true value and weight true value.
The neural network includes but is not limited to convolutional neural networks, can be in existing various target detection nerve nets
It is improved on the basis of network, such as the neural networks such as Faster R-CNN, RetinaNet, R-CNN, Fast R-CNN.
The neural network specifically includes feature extraction network with pedestrian detection network.Wherein, feature extraction network is used for
The feature in original image is extracted, the characteristic pattern of original image is exported.Pedestrian detection network is used to carry out pedestrian based on characteristic pattern
Detection, output test result.
Specifically, training image is inputted into neural network first, feature extraction is carried out, to generate the feature of a variety of scales
Figure.Can by such as HOG (Histogram of Oriented Gradient, histograms of oriented gradients) feature extraction algorithm,
LBP (Local Binary Pattern, local binary patterns) feature extraction algorithm, Haar-like feature extraction algorithm etc. is calculated
Method carries out feature extraction processing to target image, obtains the characteristic pattern of target image.In practice, feature extraction network is not necessarily
Need to rebuild completely, can also directly by some pre-training, the convolutional neural networks that are used for image classification task delete
Part after being eventually used for the full articulamentum of classification output is as feature extraction network.The structure of feature extraction network and specific
Feature extraction mode be not limited herein.
Then, it is based on the characteristic pattern, the predictive information about target object is determined, specifically includes the detection of target object
Frame position, detection block weight and detection block score, wherein if there is pedestrian, detection block position indicates to surround this pedestrian's
Rectangle frame, detection block score represent the probability in the detection block there are pedestrian.In one embodiment, detection block weight can also
To be indicated by detection block score, referring specifically to hereafter.
As shown in figure 3, existing pedestrian detection model only exports the detection block position of each detection block in the training process
With detection block score, and the embodiment of the present invention is while detection block position and the detection block score for exporting each detection block, also
Detection block weight is exported, for identifying the probability in the detection block there are target object, namely indicates the target in the detection block
With the similitude of background.
Wherein it is possible to the predictive information of target object is obtained using various feasible algorithm of target detection, it is finally defeated respectively
Detection block position, detection block weight and detection block score out.Specifically, general detection model is pre- using two classifier difference
Detection block position and detection block score are surveyed, for example, RPN network (Region Proposal Network, Area generation network) can
Can predict object boundary and target fractional simultaneously in each position；In the present embodiment, other than detection block position,
The probability in detection block there are pedestrian also is calculated separately using two the same or similar classifiers, and respectively as detection block point
Several and detection block weight.
Specifically, larger in the presence of a possibility that obscuring if the similitude of target and background in detection block is higher, because
There are the probability of target object is lower in this detection block；On the contrary, if the similitude of target and background in detection block compared with
Low, then there are the probability of target object is higher in the detection block.Therefore, if the pedestrian of this detection block position and back
Scape foreign matter is similar, then network is intended to export low probability, so that the detection block weight of the detection block at the position reduces, in
It is that the subsequent weighting classification error measured is lower, that is, reduces the weight of this detection block when training, easily obscures sample to reduce
This adverse effect to network parameter.
It is understood that detection block weight and detection block score are all used to indicate to deposit at position that the detection block is indicated
In the probability of target object.Therefore in hands-on, similar or identical classifier can be used and export detection block power respectively
Weight and detection block score, that is to say, that the probability one of the two classifiers output is another as detection as detection block score
Frame weight.
In step S220, the first error in classification between the detection block score and score true value is calculated, and according to described
Detection block weight and first error in classification calculate weighting classification error.
As described above, detection block weight represents the probability in the detection block there are pedestrian, the detection block is further indicated
In pedestrian and background similitude, if the pedestrian of this position is similar with background foreign matter, network be intended to export low probability,
So that the detection block weight of the detection block reduces.It include detection block weight and the first error in classification two in weighting classification error
It is smaller to calculate resulting weighting classification error when detection block weight is lower for a factor, easily obscures sample to instruction to reduce
Practice the influence of result.In one embodiment, the weighting classification error is that the detection block weight and first classification miss
The product of difference.
It in addition to this, further include the second error in classification calculated between the detection block weight and weight true value, Yi Jiji
Calculate the location error between the detection block position and position true value, when detection block weight by detection block score to indicate when, institute
Stating weight true value can also be indicated by score true value；Second error in classification is the training error of detection block weight, location error
For the training error of detection block position.
In step S230, it is at least based on the weighting classification error update network parameter.
Specifically, the network parameter of adjustable initial neural network, to minimize the value of weighting classification error as far as possible.Its
In, network parameter may include weight and the number of iterations of each layer of neural network etc..It in addition to this, further include based on institute
It states the second error in classification and location error updates the network parameter.
Due in detection block pedestrian and context similarity it is higher, then detection block weight is lower, and weighting classification error is opposite
Smaller, then influence of the pedestrian sample at this to training result is smaller.
Specifically, can use by backpropagation (BP, back-propagation), stochastic gradient descent (SGD,
Stochastic gradient descent) or gradient passback scheduling algorithm progress end-to-end (end-to-end) training, Lai Youhua
Parameters in model.It is every handled a width training image after, it can be determined that whether meet trained termination condition, if meeting
Condition then terminates to train, and network parameter at this time can be used as the parameter of trained pedestrian detection model.If being unsatisfactory for condition
S210 is then back to continue to train.The condition that training terminates may include that the training image in training set has been used up, lost letter
Number has been restrained etc..
The training method of pedestrian detection model according to an embodiment of the present invention is described above exemplarily.Illustratively,
The training method of pedestrian detection model according to an embodiment of the present invention can with memory and processor unit or
It is realized in person's system.
In addition, the training method of pedestrian detection model according to an embodiment of the present invention is deployed to intelligent hand in which can be convenient
In the mobile devices such as machine, tablet computer, personal computer.Alternatively, the instruction of pedestrian detection model according to an embodiment of the present invention
Server end (or cloud) can also be deployed in by practicing method.Alternatively, the instruction of pedestrian detection model according to an embodiment of the present invention
Practicing method can also be deployed at server end (or cloud) and personal terminal with being distributed.
Based on above description, training method according to an embodiment of the present invention in the training process of pedestrian's detection model from
The dynamic detection block weight for reducing pedestrian sample similar with background, to reduce the unfavorable shadow for easily obscuring sample to network parameter
It rings, greatly improves the precision of pedestrian detection.
Show included by the training method for describing pedestrian detection model according to an embodiment of the present invention above exemplarily
Example property steps flow chart.
The training device of the pedestrian detection model of another aspect of the present invention offer is described below with reference to Fig. 4.Fig. 4 shows root
According to the schematic block diagram of the training device 400 of the pedestrian detection model of the embodiment of the present invention.
As shown in figure 4, the training device 300 of pedestrian detection model according to an embodiment of the present invention include prediction module 410,
Error calculating module 420 and training module 430.The modules can execute pedestrian's inspection above in conjunction with Fig. 2 description respectively
Survey each step/function of the training method of model.
Prediction module 410 is used to training image being input to neural network, obtains the training objective in the training image
Predictive information, the predictive information includes detection block position, detection block weight and detection block score, wherein the detection block
Weight indicates the similitude of target object and background in the detection block, and the similitude is higher, then the detection block weight is got over
It is low.Conversely, similitude is lower, then the detection block weight is higher.
Wherein, the training image can be any candid photograph image comprising many pedestrians.Training set can be constructed in advance,
Include several training images in the training set, all includes one or more pedestrians, the position of pedestrian on usual every width training image
It sets and is marked in advance with bounding box, referred to as true frame.Based on the mark carried out in advance, the position of detection block can be obtained
Set true value, score true value and weight true value.
The neural network includes but is not limited to convolutional neural networks, can be in existing various target detection nerve nets
It is improved on the basis of network, such as the neural networks such as Faster R-CNN, RetinaNet, R-CNN, Fast R-CNN.
The neural network specifically includes feature extraction network with pedestrian detection network.Wherein, feature extraction network is used for
The feature in original image is extracted, the characteristic pattern of original image is exported.Pedestrian detection network is used to carry out pedestrian based on characteristic pattern
Detection, output test result.
Specifically, training image is inputted into neural network first, feature extraction is carried out, to generate the feature of a variety of scales
Figure.Can by such as HOG (Histogram of Oriented Gradient, histograms of oriented gradients) feature extraction algorithm,
LBP (Local Binary Pattern, local binary patterns) feature extraction algorithm, Haar-like feature extraction algorithm etc. mentions
It takes algorithm to carry out feature extraction processing to target image, obtains the characteristic pattern of target image.In practice, feature extraction network is not
Centainly need to rebuild completely, can also directly by some pre-training, be used for the convolutional neural networks of image classification task
It deletes and is eventually used for the part after the full articulamentum of classification output as feature extraction network.The structure of feature extraction network and
Specific feature extraction mode is not limited herein.
Then, it is based on the characteristic pattern, the predictive information about target object is determined, specifically includes the detection of target object
Frame position, detection block weight and detection block score, wherein if there is pedestrian, detection block position indicates to surround this pedestrian's
Rectangle frame, detection block score represent the probability in the detection block there are pedestrian.In one embodiment, detection block weight can also
To be indicated by detection block score, referring specifically to hereafter.
As shown in figure 3, existing pedestrian detection model only exports the detection block position of each detection block in the training process
With detection block score, and the embodiment of the present invention is while detection block position and the detection block score for exporting each detection block, also
Detection block weight is exported, for identifying the probability in the detection block there are target object, namely indicates the target in the detection block
With the similitude of background.
Wherein it is possible to the predictive information of target object is obtained using various feasible algorithm of target detection, it is finally defeated respectively
Detection block position, detection block weight and detection block score out.Specifically, general detection model is pre- using two classifier difference
Detection block position and detection block score are surveyed, for example, RPN network (Region Proposal Network, Area generation network) can
Can predict object boundary and target fractional simultaneously in each position.In the present embodiment, other than detection block position,
The probability in detection block there are pedestrian also is calculated separately using two the same or similar classifiers, and respectively as detection block point
Several and detection block weight.
Specifically, larger in the presence of a possibility that obscuring if the similitude of target and background in detection block is higher, because
There are the probability of target object is lower in this detection block；On the contrary, if the similitude of target and background in detection block compared with
Low, then there are the probability of target object is higher in the detection block.Therefore, if the pedestrian of this detection block position and back
Scape foreign matter is similar, then network is intended to export low probability, so that the detection block weight of the detection block at the position reduces, in
It is that the subsequent weighting classification error measured is lower, that is, reduces the weight of this detection block when training, easily obscures sample to reduce
This adverse effect to network parameter.
It is understood that detection block weight and detection block score are all used to indicate to deposit at position that the detection block is indicated
In the probability of target object.Therefore in hands-on, similar or identical classifier can be used and export detection block power respectively
Weight and detection block score, that is to say, that the probability one of the two classifiers output is another as detection as detection block score
Frame weight.
Error calculating module 420 is used to calculate the first error in classification between the detection block score and score true value, and
Weighting classification error is calculated according to the detection block weight and first error in classification.
As described above, detection block weight represents the probability in the detection block there are pedestrian, the detection block is further indicated
In pedestrian and background similitude, if the pedestrian of this position is similar with background foreign matter, network be intended to export low probability,
So that the detection block weight of the detection block reduces.It include detection block weight and the first error in classification two in weighting classification error
It is smaller to calculate resulting weighting classification error when detection block weight is lower for a factor, easily obscures sample to instruction to reduce
Practice the influence of result.In one embodiment, the weighting classification error is that the detection block weight and first classification miss
The product of difference.
It in addition to this, further include the second error in classification calculated between the detection block weight and weight true value, Yi Jiji
Calculate the location error between the detection block position and position true value, when detection block weight by detection block score to indicate when, institute
Stating weight true value can also be indicated by score true value；Second error in classification is the training error of detection block weight, location error
For the training error of detection block position.
Training module 430 is for being at least based on the weighting classification error update network parameter.
Specifically, the network parameter of adjustable initial neural network, to minimize the value of weighting classification error as far as possible.Its
In, network parameter may include weight and the number of iterations of each layer of neural network etc..It in addition to this, further include based on institute
It states the second error in classification and location error updates the network parameter.
Due in detection block pedestrian and context similarity it is higher, then detection block weight is lower, and weighting classification error is opposite
Smaller, then influence of the pedestrian sample at this to training result is smaller.
Specifically, can use by backpropagation (BP, back-propagation), stochastic gradient descent (SGD,
Stochastic gradient descent) or gradient passback scheduling algorithm progress end-to-end (end-to-end) training, Lai Youhua
Parameters in model.It is every handled a width training image after, it can be determined that whether meet trained termination condition, if meeting
Condition then terminates to train, and network parameter at this time can be used as the parameter of trained pedestrian detection model.If being unsatisfactory for condition
Then continue to train.The condition that training terminates may include that the training image in training set has been used up, loss function has been restrained
Etc..
The training device of pedestrian detection model according to an embodiment of the present invention is described above exemplarily.Illustratively,
The training device of pedestrian detection model according to an embodiment of the present invention can with memory and processor unit or
It is realized in person's system.
In addition, the training device of pedestrian detection model according to an embodiment of the present invention is deployed to intelligent hand in which can be convenient
In the mobile devices such as machine, tablet computer, personal computer.Alternatively, the instruction of pedestrian detection model according to an embodiment of the present invention
Server end (or cloud) can also be deployed in by practicing device.Alternatively, the instruction of pedestrian detection model according to an embodiment of the present invention
Practicing device can also be deployed at server end (or cloud) and personal terminal with being distributed.
Based on above description, training device according to an embodiment of the present invention in the training process of pedestrian's detection model from
The dynamic detection block weight for reducing pedestrian sample similar with background, to reduce the unfavorable shadow for easily obscuring sample to network parameter
It rings, greatly improves the precision of pedestrian detection.
Fig. 5 shows the schematic block diagram of the training system 500 of pedestrian detection model according to an embodiment of the present invention.Pedestrian
The training system 500 of detection model includes storage device 510 and processor 520.
Wherein, the training method that storage device 510 stores for realizing pedestrian detection model according to an embodiment of the present invention
In corresponding steps program code.Program code of the processor 520 for being stored in Running storage device 510, to execute root
According to the corresponding steps of the training method of the pedestrian detection model of the embodiment of the present invention, and for realizing according to embodiments of the present invention
Pedestrian detection model training device in corresponding module.
In one embodiment, the training of pedestrian detection model is made when said program code is run by processor 520
System 500 executes following steps:
Training image is input to neural network, to generate the prediction letter about the target object in the training image
Breath, the predictive information includes detection block position, detection block weight and detection block score, wherein the detection block weight indicates
The similitude of target object and background in the detection block, the similitude is higher, then the detection block weight is lower；
Calculate the first error in classification between the detection block score and score true value, and according to the detection block weight and
First error in classification calculates weighting classification error；
Network parameter at least based on neural network described in the weighting classification error update.
In one embodiment, the weighting classification error is multiplying for the detection block weight and first error in classification
Product.
In one embodiment, the instruction of pedestrian detection model is also made when said program code is run by processor 520
Practice system 500 to execute: calculating the second error in classification between the detection block weight and weight true value；And it is based on described second
Error in classification updates the network parameter.
In one embodiment, the instruction of pedestrian detection model is also made when said program code is run by processor 520
Practice system 500 to execute: calculating the location error between the detection block position and position true value；And it is based on the location error
Update the network parameter.
In one embodiment, the predictive information for generating the training objective in the training image includes: based on institute
It states neural network and feature extraction is carried out to the training image, to generate the characteristic pattern of the training image；According to the feature
Figure generates the predictive information.
In addition, according to embodiments of the present invention, additionally providing a kind of storage medium, storing program on said storage
Instruction, when described program instruction is run by computer or processor for executing the pedestrian detection model of the embodiment of the present invention
The corresponding steps of training method, and for realizing the phase in the training device of pedestrian detection model according to an embodiment of the present invention
Answer module.The storage medium for example may include the storage card of smart phone, the storage unit of tablet computer, personal computer
Hard disk, read-only memory (ROM), Erasable Programmable Read Only Memory EPROM (EPROM), portable compact disc read-only memory
(CD-ROM), any combination of USB storage or above-mentioned storage medium.The computer readable storage medium can be one
Any combination of a or multiple computer readable storage mediums.
In one embodiment, the computer program instructions may be implemented real according to the present invention when being run by computer
Each functional module of the training device of the pedestrian detection model of example is applied, and/or can be executed according to embodiments of the present invention
Pedestrian detection model training method.
In one embodiment, the computer program instructions make computer or place when being run by computer or processor
It manages device and executes following steps:
Training image is input to neural network, to generate the prediction letter about the target object in the training image
Breath, the predictive information includes detection block position, detection block weight and detection block score, wherein the detection block weight indicates
The similitude of target object and background in the detection block, the similitude is higher, then the detection block weight is lower；
Calculate the first error in classification between the detection block score and score true value, and according to the detection block weight and
First error in classification calculates weighting classification error；
Network parameter at least based on neural network described in the weighting classification error update.
In one embodiment, the weighting classification error is multiplying for the detection block weight and first error in classification
Product.
In one embodiment, the computer program instructions also make when being run by computer or processor computer or
Processor executes: calculating the second error in classification between the detection block weight and weight true value；And it is based on described second point
Network parameter described in class error update.
In one embodiment, the computer program instructions also make when being run by computer or processor computer or
Processor executes: calculating the location error between the detection block position and position true value；And more based on the location error
The new network parameter.
In one embodiment, the predictive information for generating the training objective in the training image includes: based on institute
It states neural network and feature extraction is carried out to the training image, to generate the characteristic pattern of the training image；According to the feature
Figure generates the predictive information.
Training method, device, system and the computer-readable medium of pedestrian detection model according to an embodiment of the present invention exist
Automatically the detection block weight of pedestrian sample similar with background is reduced in the training process of pedestrian detection model, to reduce easily
Sample is obscured to the adverse effect of network parameter, greatly improves the precision of pedestrian detection.
Although describing example embodiment by reference to attached drawing here, it should be understood that above example embodiment are only exemplary
, and be not intended to limit the scope of the invention to this.Those of ordinary skill in the art can carry out various changes wherein
And modification, it is made without departing from the scope of the present invention and spiritual.All such changes and modifications are intended to be included in appended claims
Within required the scope of the present invention.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.For example, apparatus embodiments described above are merely indicative, for example, the division of the unit, only
Only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can be tied
Another equipment is closed or is desirably integrated into, or some features can be ignored or not executed.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that implementation of the invention
Example can be practiced without these specific details.In some instances, well known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the present invention and help to understand one or more of the various inventive aspects, In
To in the description of exemplary embodiment of the present invention, each feature of the invention be grouped together into sometimes single embodiment, figure,
Or in descriptions thereof.However, the method for the invention should not be construed to reflect an intention that i.e. claimed
The present invention claims features more more than feature expressly recited in each claim.More precisely, such as corresponding power
As sharp claim reflects, inventive point is that the spy of all features less than some disclosed single embodiment can be used
Sign is to solve corresponding technical problem.Therefore, it then follows thus claims of specific embodiment are expressly incorporated in this specific
Embodiment, wherein each, the claims themselves are regarded as separate embodiments of the invention.
It will be understood to those skilled in the art that any combination pair can be used other than mutually exclusive between feature
All features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosed any method
Or all process or units of equipment are combined.Unless expressly stated otherwise, this specification (is wanted including adjoint right
Ask, make a summary and attached drawing) disclosed in each feature can be replaced with an alternative feature that provides the same, equivalent, or similar purpose.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments mean it is of the invention
Within the scope of and form different embodiments.For example, in detail in the claims, embodiment claimed it is one of any
Can in any combination mode come using.
Various component embodiments of the invention can be implemented in hardware, or to run on one or more processors
Software module realize, or be implemented in a combination thereof.It will be understood by those of skill in the art that can be used in practice
Microprocessor or other suitable processors realize some or all function of some modules according to an embodiment of the present invention
Energy.The present invention be also implemented as executing method as described herein some or all program of device (for example,
Computer program and computer program product).It is such to realize that program of the invention can store on a computer-readable medium,
Or it may be in the form of one or more signals.Such signal can be downloaded from an internet website to obtain, or
It is provided on the carrier signal, or is provided in any other form.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and ability
Field technique personnel can be designed alternative embodiment without departing from the scope of the appended claims.In the claims,
Any reference symbol between parentheses should not be configured to limitations on claims.Word "comprising" does not exclude the presence of not
Element or step listed in the claims.Word "a" or "an" located in front of the element does not exclude the presence of multiple such
Element.The present invention can be by means of including the hardware of several different elements and being come by means of properly programmed computer real
It is existing.In the unit claims listing several devices, several in these devices can be through the same hardware branch
To embody.The use of word first, second, and third does not indicate any sequence.These words can be explained and be run after fame
Claim.
The above description is merely a specific embodiment or to the explanation of specific embodiment, protection of the invention
Range is not limited thereto, and anyone skilled in the art in the technical scope disclosed by the present invention, can be easily
Expect change or replacement, should be covered by the protection scope of the present invention.Protection scope of the present invention should be with claim
Subject to protection scope.
Claims (10)
1. a kind of training method of pedestrian detection model, which is characterized in that the training method includes:
Training image is input to neural network, to generate the predictive information about the target object in the training image, institute
Stating predictive information includes detection block position, detection block weight and detection block score, wherein the detection block weight indicates the inspection
The similitude of target object and background in frame is surveyed, the similitude is higher, then the detection block weight is lower；
Calculate the first error in classification between the detection block score and score true value, and according to the detection block weight and described
First error in classification calculates weighting classification error；
Network parameter at least based on neural network described in the weighting classification error update.
2. the training method of pedestrian detection model according to claim 1, which is characterized in that the weighting classification error is
The product of the detection block weight and first error in classification.
3. the training method of pedestrian detection model according to claim 1, which is characterized in that further include:
Calculate the second error in classification between the detection block weight and weight true value；And
The network parameter is updated based on second error in classification.
4. the training method of pedestrian detection model according to claim 1, which is characterized in that further include:
Calculate the location error between the detection block position and position true value；And
The network parameter is updated based on the location error.
5. the training method of pedestrian detection model according to claim 1, which is characterized in that described to generate the training figure
The predictive information of training objective as in includes:
Feature extraction is carried out to the training image based on the neural network, to generate the characteristic pattern of the training image；
The predictive information is generated according to the characteristic pattern.
6. a kind of training device of pedestrian detection model, which is characterized in that the training device of the pedestrian detection model includes:
Prediction module, for training image to be input to neural network, to generate about the target object in the training image
Predictive information, the predictive information includes detection block position, detection block weight and detection block score, wherein the detection block
Weight indicates the similitude of target object and background in the detection block, and the similitude is higher, then the detection block weight is got over
It is low；
Error calculating module, for calculating the first error in classification between the detection block score and score true value, and according to institute
It states detection block weight and first error in classification calculates weighting classification error；And
Training module, for the network parameter at least based on neural network described in the weighting classification error update.
7. the training device of pedestrian detection model according to claim 6, which is characterized in that the weighting classification error is
The product of the detection block weight and first error in classification.
8. the training device of pedestrian detection model according to claim 6, which is characterized in that further include:
Characteristic extracting module, for carrying out feature extraction to the training image based on the neural network, to generate the instruction
Practice the characteristic pattern of image；Also,
The prediction module generates the predictive information according to the characteristic pattern.
9. a kind of training system of pedestrian detection model, which is characterized in that the training system of the pedestrian detection model includes depositing
Method for storing and processor are stored with the computer program run by the processor, the computer journey in the storage method
Sequence executes the training side of the pedestrian detection model as described in any one of claim 1-5 when being run by the processor
Method.
10. a kind of computer-readable medium, which is characterized in that be stored with computer program, institute on the computer-readable medium
State the training method for the pedestrian detection model that computer program is executed at runtime as described in any one of claim 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910615436.4A CN110490058B (en) | 2019-07-09 | 2019-07-09 | Training method, device and system of pedestrian detection model and computer readable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910615436.4A CN110490058B (en) | 2019-07-09 | 2019-07-09 | Training method, device and system of pedestrian detection model and computer readable medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110490058A true CN110490058A (en) | 2019-11-22 |
| CN110490058B CN110490058B (en) | 2022-07-26 |
Family
ID=68546867
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910615436.4A Active CN110490058B (en) | 2019-07-09 | 2019-07-09 | Training method, device and system of pedestrian detection model and computer readable medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110490058B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091091A (en) * | 2019-12-16 | 2020-05-01 | 北京迈格威科技有限公司 | Method, device and equipment for extracting target object re-identification features and storage medium |
| CN112863187A (en) * | 2021-01-18 | 2021-05-28 | 北京百度网讯科技有限公司 | Detection method of perception model, electronic equipment, road side equipment and cloud control platform |
| WO2022043741A1 (en) * | 2020-08-25 | 2022-03-03 | 商汤国际私人有限公司 | Network training method and apparatus, person re-identification method and apparatus, storage medium, and computer program |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
| CN107909027A (en) * | 2017-11-14 | 2018-04-13 | 电子科技大学 | It is a kind of that there is the quick human body target detection method for blocking processing |
| CN107944369A (en) * | 2017-11-17 | 2018-04-20 | 大连大学 | A kind of pedestrian detection method based on tandem zones generation network and enhancing random forest |
| CN108038409A (en) * | 2017-10-27 | 2018-05-15 | 江西高创保安服务技术有限公司 | A kind of pedestrian detection method |
| CN108446662A (en) * | 2018-04-02 | 2018-08-24 | 电子科技大学 | A kind of pedestrian detection method based on semantic segmentation information |
| CN108563977A (en) * | 2017-12-18 | 2018-09-21 | 华南理工大学 | A kind of the pedestrian's method for early warning and system of expressway entrance and exit |
| CN108875770A (en) * | 2018-02-06 | 2018-11-23 | 北京迈格威科技有限公司 | Mask method, device, system and the storage medium of pedestrian detection wrong report data |
| CN108875903A (en) * | 2018-01-02 | 2018-11-23 | 北京迈格威科技有限公司 | Method, apparatus, system and the computer storage medium of image detection |
| WO2019071739A1 (en) * | 2017-10-13 | 2019-04-18 | 平安科技（深圳）有限公司 | Face living body detection method and apparatus, readable storage medium and terminal device |
-
2019
- 2019-07-09 CN CN201910615436.4A patent/CN110490058B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
| WO2019071739A1 (en) * | 2017-10-13 | 2019-04-18 | 平安科技（深圳）有限公司 | Face living body detection method and apparatus, readable storage medium and terminal device |
| CN108038409A (en) * | 2017-10-27 | 2018-05-15 | 江西高创保安服务技术有限公司 | A kind of pedestrian detection method |
| CN107909027A (en) * | 2017-11-14 | 2018-04-13 | 电子科技大学 | It is a kind of that there is the quick human body target detection method for blocking processing |
| CN107944369A (en) * | 2017-11-17 | 2018-04-20 | 大连大学 | A kind of pedestrian detection method based on tandem zones generation network and enhancing random forest |
| CN108563977A (en) * | 2017-12-18 | 2018-09-21 | 华南理工大学 | A kind of the pedestrian's method for early warning and system of expressway entrance and exit |
| CN108875903A (en) * | 2018-01-02 | 2018-11-23 | 北京迈格威科技有限公司 | Method, apparatus, system and the computer storage medium of image detection |
| CN108875770A (en) * | 2018-02-06 | 2018-11-23 | 北京迈格威科技有限公司 | Mask method, device, system and the storage medium of pedestrian detection wrong report data |
| CN108446662A (en) * | 2018-04-02 | 2018-08-24 | 电子科技大学 | A kind of pedestrian detection method based on semantic segmentation information |
Non-Patent Citations (2)
| Title |
|---|
| QIWEI PENG, ETC.: "Pedestrian Detection for Transformer Substation Based on Gaussian Mixture Model and YOLO", 《2016 8TH INTERNATIONAL CONFERENCE ON INTELLIGENT HUMAN-MACHINE SYSTEMS AND CYBERNETICS》 * |
| 刘建国等: "基于YOLOv2的行人检测方法研究", 《数字制造科学》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091091A (en) * | 2019-12-16 | 2020-05-01 | 北京迈格威科技有限公司 | Method, device and equipment for extracting target object re-identification features and storage medium |
| WO2022043741A1 (en) * | 2020-08-25 | 2022-03-03 | 商汤国际私人有限公司 | Network training method and apparatus, person re-identification method and apparatus, storage medium, and computer program |
| CN112863187A (en) * | 2021-01-18 | 2021-05-28 | 北京百度网讯科技有限公司 | Detection method of perception model, electronic equipment, road side equipment and cloud control platform |
| CN112863187B (en) * | 2021-01-18 | 2022-04-15 | 阿波罗智联(北京)科技有限公司 | Detection method of perception model, electronic equipment, road side equipment and cloud control platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110490058B (en) | 2022-07-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108197532B (en) | The method, apparatus and computer installation of recognition of face | |
| CN108875932A (en) | Image-recognizing method, device and system and storage medium | |
| CN108875522A (en) | Face cluster methods, devices and systems and storage medium | |
| CN109146830A (en) | For generating the method, apparatus, system and storage medium of training data | |
| CN109670452A (en) | Method for detecting human face, device, electronic equipment and Face datection model | |
| CN108875766A (en) | Method, apparatus, system and the computer storage medium of image procossing | |
| CN109919211A (en) | Commodity recognition method, device, system and computer-readable medium | |
| CN110309706A (en) | Face critical point detection method, apparatus, computer equipment and storage medium | |
| CN108921159A (en) | Method and apparatus for detecting the wear condition of safety cap | |
| CN105938552A (en) | Face recognition method capable of realizing base image automatic update and face recognition device | |
| CN108229341A (en) | Sorting technique and device, electronic equipment, computer storage media, program | |
| CN108875546A (en) | Face auth method, system and storage medium | |
| CN106650662A (en) | Target object occlusion detection method and target object occlusion detection device | |
| CN110490058A (en) | Training method, device, system and the computer-readable medium of pedestrian detection model | |
| CN107918767B (en) | Object detection method, device, electronic equipment and computer-readable medium | |
| CN111522987A (en) | Image auditing method and device and computer readable storage medium | |
| CN110009614A (en) | Method and apparatus for output information | |
| CN108875731A (en) | Target identification method, device, system and storage medium | |
| CN108875767A (en) | Method, apparatus, system and the computer storage medium of image recognition | |
| CN106131405A (en) | Image processing method and device and for its image capturing system | |
| CN108875509A (en) | Biopsy method, device and system and storage medium | |
| CN108876764A (en) | Render image acquiring method, device, system and storage medium | |
| CN108875517A (en) | Method for processing video frequency, device and system and storage medium | |
| CN108229418A (en) | Human body critical point detection method and apparatus, electronic equipment, storage medium and program | |
| CN108875483A (en) | Image detecting method, device, system and computer-readable 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 |