CN110163081A - Regional invasion real-time detection method, system and storage medium based on SSD - Google Patents
Regional invasion real-time detection method, system and storage medium based on SSD Download PDFInfo
- Publication number
- CN110163081A CN110163081A CN201910260758.1A CN201910260758A CN110163081A CN 110163081 A CN110163081 A CN 110163081A CN 201910260758 A CN201910260758 A CN 201910260758A CN 110163081 A CN110163081 A CN 110163081A
- Authority
- CN
- China
- Prior art keywords
- ssd
- detected
- training
- frame
- model
- 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.)
- Pending
Links
Classifications
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses regional invasion real-time detection method, system and storage medium based on SSD, method includes: that the training set training after being expanded according to data obtains SSD model;Compression processing is carried out to SSD model, and the SSD model after compression processing is deployed to end side equipment;It according to the identification classification after screening, is detected by image to be detected that SSD model opposite end side apparatus obtains, generates intrusion detection result;Cloud is updated to when by intrusion detection fructufy.Present invention reduces the service requirements of model, reduce detection time, and real-time is high, can be widely applied to machine learning techniques field.
Description
Technical field
The present invention relates to machine learning techniques field, be based especially on the regional invasion real-time detection method of SSD, system and
Storage medium.
Background technique
Object detection and recognition is always one of problem for computer vision field, recently as artificial intelligence,
The development of neural network and computer vision technique, numerous scholars propose a series of neural network deep layer frameworks, due to these
Work, the very big progress that object detection and recognition technology obtains, while having bigger application prospect.Conducive to target detection with
It is the application scheme that it is applied that identification technology, which carries out intelligent video monitoring, and in actual life, some special regions need
The protection and monitoring of intelligence, such as airport, around the important region in the ground such as railway or critical facility or wired home, wisdom city
The foundation in city requires to carry out real-time region intrusion detection.The introducing real-time region intrusion detection of neural network framework is being protected
While having demonstrate,proved algorithm stability, target category can be accurately identified.
Traditional target detection mainly uses optical flow method, frame difference method and background subtraction division, these methods can in real time into
Row region intrusion detection, but this serial of methods is illuminated by the light, complex degree of background, the influences of many factors such as weather conditions compared with
Greatly.As to real-time, stability, accuracy and the requirement for adapting to complex scene are higher and higher, pass through neural network framework
Object detection and recognition paid attention to by more and more scholars, and have lot of research.R-CNN algorithm, YOLO (You
Only Look Once) algorithm, SSD (Single Shot MultiBox Detector) algorithm, is all mesh proposed in recent years
Mark detection and recognizer, these algorithms have outstanding performance in stability, accuracy.But these algorithms are identifying
When need to occupy a large amount of computing resource, real-time is low, significantly limit object detection and recognition region invasion on answering
With with development.
Summary of the invention
In view of this, the embodiment of the present invention provide a kind of high regional invasion real-time detection method based on SSD of real-time,
System and storage medium.
On the one hand, the embodiment of the invention provides a kind of regional invasion real-time detection methods based on SSD, including following step
It is rapid:
Training set training after being expanded according to data obtains SSD model;
Compression processing is carried out to SSD model, and the SSD model after compression processing is deployed to end side equipment;
According to the identification classification after screening, detected by image to be detected that SSD model opposite end side apparatus obtains, it is raw
At intrusion detection result;
Cloud is updated to when by intrusion detection fructufy.
Further, it is described expanded according to data after training set training the step for obtaining SSD model, including following step
It is rapid:
Determine the priori frame in training picture;
IOU value is calculated according to priori frame;
According to the IOU value and preset threshold value being calculated, the positive sample and negative sample in priori frame are determined;
Data amplification is carried out to training picture by data amplification technique, the data amplification technique includes at flip horizontal
Reason, random cropping processing and color distortion processing;
Determine training parameter, the training picture and negative sample training after expanding by data obtain SSD model.
Further, it is described expanded according to data after training set training the step for obtaining SSD model, further include following step
It is rapid:
The IOU value of negative sample is ranked up;
Negative sample is extracted according to ranking results;
Determine the loss function of SSD model.
Further, the identification classification according to after screening, the image to be detected obtained by SSD model opposite end side apparatus
The step for being detected, generating intrusion detection result, comprising the following steps:
Obtain the original predictive frame of image to be detected;
Classification confidence level is determined according to SSD model;
The Low threshold prediction block in original predictive frame is filtered with background forecast frame according to classification confidence level, obtains mesh
Mark prediction block;
Target prediction frame is decoded, location parameter is obtained;
The detected region in image to be detected is demarcated according to location parameter;
Detected region is detected by SSD model, obtains intrusion detection result.
Further, the identification classification according to after screening, the image to be detected obtained by SSD model opposite end side apparatus
The step for being detected, generating intrusion detection result, further comprising the steps of:
It is ranked up according to the confidence level of target prediction frame;
According to ranking results, is extracted from target prediction frame and obtain final prediction block.
Further, the end side equipment is raspberry pie.
On the other hand, the real-time region intruding detection system based on SSD that the embodiment of the invention also provides a kind of, comprising:
Training module obtains SSD model for the training set training after expanding according to data;
SSD model after compression processing for carrying out compression processing to SSD model, and is deployed to end side by deployment module
Equipment;
Detection module, for passing through the mapping to be checked of SSD model opposite end side apparatus acquisition according to the identification classification after screening
As being detected, intrusion detection result is generated;
Release module is updated to cloud when for by intrusion detection fructufy.
Further, the detection module includes:
Acquiring unit, for obtaining the original predictive frame of image to be detected;
Confidence level determination unit, for determining classification confidence level according to SSD model;
Prediction block filter element, for according to classification confidence level in original predictive frame Low threshold prediction block and background it is pre-
It surveys frame to be filtered, obtains target prediction frame;
Decoding unit obtains location parameter for being decoded to target prediction frame;
Unit is demarcated, for demarcating the detected region in image to be detected according to location parameter;
Detection unit obtains intrusion detection result for detecting by SSD model to detected region.
On the other hand, the real-time region intruding detection system based on SSD that the embodiment of the invention also provides a kind of, comprising:
At least one processor;
At least one processor, for storing at least one program;
When at least one described program is executed by least one described processor, so that at least one described processor is realized
The regional invasion real-time detection method based on SSD.
On the other hand, the embodiment of the invention also provides a kind of storage mediums, wherein being stored with the executable finger of processor
It enables, the executable instruction of the processor is when executed by the processor for executing the real-time region invasion based on SSD
Detection method.
One or more technical solutions in the embodiments of the present invention have the advantages that the embodiment of the present invention passes through
The identification classification of SSD model is screened, the service requirement of model is reduced, reduces detection time, moreover it is possible to by instruction
Practice collection and carries out data amplification to guarantee the stability and accuracy of model;SSD model is deployed to end side by the embodiment of the present invention
Equipment is distributed to cloud by will test result after the completion detection of end side equipment, and real-time is high.
Detailed description of the invention
Fig. 1 is the overall step flow chart of the embodiment of the present invention;
Fig. 2 is the model training flow chart of the embodiment of the present invention;
Fig. 3 is the intrusion detection flow chart of the embodiment of the present invention.
Specific embodiment
The present invention is further explained and is illustrated with specific embodiment with reference to the accompanying drawings of the specification.For of the invention real
The step number in example is applied, is arranged only for the purposes of illustrating explanation, any restriction is not done to the sequence between step, is implemented
The execution sequence of each step in example can be adaptively adjusted according to the understanding of those skilled in the art.
The present invention proposes a kind of regional invasion real-time detection method based on SSD, and this method is transported in the intrusion detection of region
While guaranteeing algorithm stability and accuracy, model is reduced by reducing the scale of model with improved SSD algorithm
Service requirement, reduce detection time, realize the real-time of intrusion detection.Model on GPU to public data collection and
The data set of acquisition is configured and is trained, and to model measurement and exports deployment in raspberry pie, when mode input is invasion
The detected region for acquiring video information and first frame image specification, since end side computing resource is limited, the present embodiment is exported not
The invader for being video or picture but being identified identifies after successfully uploading cloud, sends invasion information to user terminal.
Referring to Fig.1, the embodiment of the invention provides a kind of regional invasion real-time detection method based on SSD, including it is following
Step:
Training set training after being expanded according to data obtains SSD model;
Compression processing is carried out to SSD model, and the SSD model after compression processing is deployed to end side equipment;
According to the identification classification after screening, detected by image to be detected that SSD model opposite end side apparatus obtains, it is raw
At intrusion detection result;
Cloud is updated to when by intrusion detection fructufy.
Wherein, when label is with using training data, it is person that the present embodiment, which modifies the training classification in VOC2007,
This five kinds common invasion classifications of car, bus, cat, dog.
The classification of traditional training dataset VOC2007 shares 20 classes, and the composition of data set includes class label and true
Target area etc., because intrusion detection, to the particularity of class requirement, some classifications will not be involved.Then the present embodiment uses
Identification class method for distinguishing is reduced, i.e., in model training stage, only defines and uses five classifications in data set, respectively
Person, car, bus, cat, dog.
Be further used as preferred embodiment, it is described expanded according to data after training set training obtain SSD model this
One step, comprising the following steps:
Determine the priori frame in training picture;
IOU value is calculated according to priori frame;
According to the IOU value and preset threshold value being calculated, the positive sample and negative sample in priori frame are determined;
Data amplification is carried out to training picture by data amplification technique, the data amplification technique includes at flip horizontal
Reason, random cropping processing and color distortion processing;
Determine training parameter, the training picture and negative sample training after expanding by data obtain SSD model.
Be further used as preferred embodiment, it is described expanded according to data after training set training obtain SSD model this
One step, further comprising the steps of:
The IOU value of negative sample is ranked up;
Negative sample is extracted according to ranking results;
Determine the loss function of SSD model.
It is further used as preferred embodiment, the identification classification according to after screening sets end side by SSD model
The step for standby image to be detected obtained is detected, generates intrusion detection result, comprising the following steps:
Obtain the original predictive frame of image to be detected;
Classification confidence level is determined according to SSD model;
The Low threshold prediction block in original predictive frame is filtered with background forecast frame according to classification confidence level, obtains mesh
Mark prediction block;
Target prediction frame is decoded, location parameter is obtained;
The detected region in image to be detected is demarcated according to location parameter;
Detected region is detected by SSD model, obtains intrusion detection result.
It is further used as preferred embodiment, the identification classification according to after screening sets end side by SSD model
The step for standby image to be detected obtained is detected, generates intrusion detection result, further comprising the steps of:
It is ranked up according to the confidence level of target prediction frame;
According to ranking results, is extracted from target prediction frame and obtain final prediction block.
It is further used as preferred embodiment, the end side equipment is raspberry pie.
It is corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of real-time region intrusion detection system based on SSD
System, comprising:
Training module obtains SSD model for the training set training after expanding according to data;
SSD model after compression processing for carrying out compression processing to SSD model, and is deployed to end side by deployment module
Equipment;
Detection module, for passing through the mapping to be checked of SSD model opposite end side apparatus acquisition according to the identification classification after screening
As being detected, intrusion detection result is generated;
Release module is updated to cloud when for by intrusion detection fructufy.
It is further used as preferred embodiment, the detection module includes:
Acquiring unit, for obtaining the original predictive frame of image to be detected;
Confidence level determination unit, for determining classification confidence level according to SSD model;
Prediction block filter element, for according to classification confidence level in original predictive frame Low threshold prediction block and background it is pre-
It surveys frame to be filtered, obtains target prediction frame;
Decoding unit obtains location parameter for being decoded to target prediction frame;
Unit is demarcated, for demarcating the detected region in image to be detected according to location parameter;
Detection unit obtains intrusion detection result for detecting by SSD model to detected region.
It is corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of real-time region intrusion detection system based on SSD
System, comprising:
At least one processor;
At least one processor, for storing at least one program;
When at least one described program is executed by least one described processor, so that at least one described processor is realized
The regional invasion real-time detection method based on SSD.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of storage mediums, wherein being stored with processor can
The instruction of execution, the executable instruction of the processor are described based on the real-time of SSD for executing when executed by the processor
Region intrusion detection method.
A kind of whole implementation step of regional invasion real-time detection method based on SSD of the embodiment of the present invention is described below in detail
It is rapid:
S1, selection training set, setting training parameter carry out model training;
Specifically, the network structure of the model of the present embodiment is used based on VGG16 using classical SSD model, SSD
Then model has increased convolutional layer newly on the basis of VGG16 to obtain more characteristic patterns so that for detecting, SSD utilizes more rulers
Characteristic pattern detection is spent, the present embodiment has also done intensive sampling to training picture.
As shown in Fig. 2, the model training process of the present embodiment specifically includes the following steps:
S11, the priori frame trained in picture is determined
Specifically, it is determined that being responsible for predicting to train picture with this with the matched priori frame of ground truth in training picture;
S12, IOU value is calculated according to priori frame;
Specifically, the present embodiment carries out the matching of priori frame by IOU threshold decision, while separating positive negative sample;
Preferably, the present embodiment be calculated IOU threshold value process it is as follows:
By handing over and than (Intersection-Over-Union, IOU), a concept used in target detection is to produce
The overlapping rate of raw candidate frame (candidate bound) and former indicia framing (ground truth bound), i.e. their friendship
The ratio of collection and union, most ideally completely overlapped, i.e., ratio is 1.Friendship can be calculated using following formula and is compared.According to warp
It tests threshold value and sets its threshold value as 0.5.
Wherein, area (C) represents candidate frame;Area (G) represents former indicia framing.
The IOU value and preset threshold value that S13, basis are calculated, determine the positive sample and negative sample in priori frame;
Specifically, if IOU > 0.5, it is determined as positive sample;
If IOU={ max }, is also determined as positive sample;
Remaining situation is determined as background i.e. negative sample;
S14, it samples to negative sample and extracts the small sample of confidence level, the quantity of positive negative sample is balanced with this;
Specifically, the process that the small sample of confidence level is extracted in the present embodiment is as follows: thinking that IOU value is less than when model training
0.5 priori frame is background information namely negative sample, to guarantee that positive negative sample does not interfere with each other, in negative sample selection, first
Choosing the minimum sample of IOU value is negative sample, is carrying out ascending order arrangement to sample, until positive and negative sample proportion is close to 1:3.
S15, loss function is determined;
Preferably, the loss function of the present embodiment is made of the loss of confidence level and the loss of position.
Wherein, the confidence level in the present embodiment loses formula are as follows:
Wherein, Lconf(x, c) is the loss of multi-class softmax loss;
Indicate the coefficient that j-th of Ground Truth Box of i-th of default frame and classification p match;
Expression is not matched to suitable Ground Truth Box;
Indicate the softmax probability that prediction default frame i is classification p;
It is expressed as the probability of background.
S16, increased by data and expand lift scheme performance;
Preferably, the present embodiment uses flip horizontal, random cropping, color distortion and stochastical sampling in the training process
The data such as block domain increase expansion technology to carry out data amplification processing.
S17, adjusting training parameter simultaneously carry out model training.
Preferably, the present embodiment regularization attenuation coefficient in training is 0.0005, learning rate 0.00001, learning rate
Decay factor is 0.94, and training classification is 5 classes.
S2, test model simultaneously export;
S3, in end side deployment model;
Specifically, the prediction model of the present embodiment is deployed in raspberry pie after generating, and intrusion detection is broadly divided into two
Stage: object detection and identification.
S4, pretreatment and detected regional assignment are carried out to the video of input;
S5, detected region is performed intrusion detection;
As shown in figure 3, the step of intrusion detection of the present embodiment, includes:
S51, image to be detected information prediction frame is obtained;
S52, the judgement of classification confidence level is carried out according to the model of generation;
S53, Low threshold prediction block and background forecast frame are filtered according to classification confidence level;
S54, the remaining prediction block of decoding and the acquisition for carrying out location parameter;
Specifically, decoding prediction block is by priori frame from the predicted value of bounding box to obtain the actual position of bounding box
Process.Priori frame position d=(dcx, dcy, dw, dh) indicate, correspond to bounding box b=(bcx, bcy, bw, bh) indicate, that
The predicted value l of bounding box is conversion value of the b relative to d in fact:
This above-mentioned process is the cataloged procedure of bounding box, this reversed process, as decoding process when prediction, from prediction
The actual position b of bounding box is obtained in value l:
bcx=dwlcx+dcx, bcy=dhlcy+dcy
bw=dw exp(lw), bh=dh exp(lh)
Wherein, in c generation, refers to center, refers to the centre coordinate of target;Cx represents central point abscissa;It is vertical that cy represents central point
Coordinate;Cx, cy are respectively with the x that occurs in the present embodiment, and y is corresponding, and x represents abscissa, and y represents ordinate.
The location parameter of the target of the present embodiment, that is, bounding box parameter, i.e. b=(bx, by, bw, bh), wherein bx, by
For the centre coordinate of target, width and height that w and h are bounding box.
S55, confidence level is judged again and according to the big minispread of confidence level, extract the big prediction block of confidence level;
S56, according to the detected region of final prediction block and calibration, perform intrusion detection.
S6, the information that will test reach user terminal.
Specifically, after detecting invasion information, upload information to Cloud Server is retransmited to user's mobile device, most
The real-time intrusion detection in region is realized eventually.
The present invention is realized using deep learning frame tensorflow, and accelerates operation, instruction using graphics processor (GPU)
Practicing platform is Ubuntu16.04 and Tesla-P100 video card, and test platform is raspberry pie 3B+, and is built in raspberry pie
Tensorflow frame is completed.
In order to realize the intrusion detection of end side, the present invention carries out threshold decision to training data first, and positive and negative sample is balanced,
Increasing expansion, the performance of lift scheme are carried out to data again, adjusting parameter generates model after being trained;On the other hand, in end side
Tensorflow learning framework is built in raspberry pie, model is deployed in raspberry pie after configuration successful, and detection process passes through reading
It takes detected area video to start, starts reading model and detected region in video, by classification Confidence, filtering is not
Reliable prediction frame, obtain location information, the location information of final goal is extracted again according to confidence level, by with area to be detected
Intrusion detection is realized in the comparison of location information domain;When detect invasion information after, upload information to Cloud Server, retransmit to
Family mobile device.It is final to realize the real-time intrusion detection in region.
Due to illumination, environment, the influence of the various situations such as complex conditions, traditional optical flow method or background subtraction intrusion detection
Method limitation is larger;The method universal model of deep learning is larger, cannot complete the task of intrusion detection in end side.Therefore,
The present invention realizes a kind of real-time region intruding detection system based on SSD, and mainly there are four modules to form for the system, will compress
SSD model after training is deployed in end side and realizes intrusion detection, and the information that will test uploads cloud and passes to user again,
While the precision and stability that ensure that calculating, real-time detection is realized.Present invention could apply to airport, railway, families
Around front yard, area video information is obtained by camera, region intrusion detection is realized in raspberry pie, system cost is low, by ring
Border influence is small, and accuracy is high, and real-time is good, application easy to spread.
In conclusion a kind of regional invasion real-time detection method based on SSD of the present invention, system and storage medium, mainly
Including 4 modules: 1) model training module is matched using priori frame and separates positive negative sample with IOU threshold decision, to positive negative sample
Learning parameter is arranged after balanced proportions to be trained, generates model;2) end side deployment module, the deployment model in raspberry pie are
Guarantee detection real-time, identifies classification by reducing, and increase to image data information and expand to compact model, pass through invasion classification
Output complete intrusion detection;3) intrusion detection module passes through the judgement of classification confidence level and threshold calculations to prediction block is obtained
It filters extra detection block, after decoded positions parameter, final detection block is extracted again according to position confidence level, according to final detection block
Information and intrusion detection detected area information complete intrusion detection;4) invade information transfer module, end side complete into
When invading detection and obtaining invader classification, end side uploads invasion information to cloud, and cloud server retransmits intrusion detection information
To user's mobile device, the transmission of invasion information is realized.
It is to be illustrated to preferable implementation of the invention, but the present invention is not limited to the embodiment above, it is ripe
Various equivalent deformation or replacement can also be made on the premise of without prejudice to spirit of the invention by knowing those skilled in the art, this
Equivalent deformation or replacement are all included in the scope defined by the claims of the present application a bit.
Claims (10)
1. the regional invasion real-time detection method based on SSD, it is characterised in that: the following steps are included:
Training set training after being expanded according to data obtains SSD model;
Compression processing is carried out to SSD model, and the SSD model after compression processing is deployed to end side equipment;
According to the identification classification after screening, detected by image to be detected that SSD model opposite end side apparatus obtains, generate into
Invade testing result;
Cloud is updated to when by intrusion detection fructufy.
2. the regional invasion real-time detection method according to claim 1 based on SSD, it is characterised in that: described according to number
The step for obtaining SSD model according to the training set training after amplification, comprising the following steps:
Determine the priori frame in training picture;
IOU value is calculated according to priori frame;
According to the IOU value and preset threshold value being calculated, the positive sample and negative sample in priori frame are determined;
By data amplification technique to training picture carry out data amplification, the data amplification technique include flip horizontal processing,
Random cropping processing and color distortion processing;
Determine training parameter, the training picture and negative sample training after expanding by data obtain SSD model.
3. the regional invasion real-time detection method according to claim 2 based on SSD, it is characterised in that: described according to number
The step for obtaining SSD model according to the training set training after amplification, further comprising the steps of:
The IOU value of negative sample is ranked up;
Negative sample is extracted according to ranking results;
Determine the loss function of SSD model.
4. the regional invasion real-time detection method according to claim 1 based on SSD, it is characterised in that: described according to sieve
Identification classification after choosing is detected by image to be detected that SSD model opposite end side apparatus obtains, and generates intrusion detection result
The step for, comprising the following steps:
Obtain the original predictive frame of image to be detected;
Classification confidence level is determined according to SSD model;
The Low threshold prediction block in original predictive frame is filtered with background forecast frame according to classification confidence level, it is pre- to obtain target
Survey frame;
Target prediction frame is decoded, location parameter is obtained;
The detected region in image to be detected is demarcated according to location parameter;
Detected region is detected by SSD model, obtains intrusion detection result.
5. the regional invasion real-time detection method according to claim 4 based on SSD, it is characterised in that: described according to sieve
Identification classification after choosing is detected by image to be detected that SSD model opposite end side apparatus obtains, and generates intrusion detection result
The step for, it is further comprising the steps of:
It is ranked up according to the confidence level of target prediction frame;
According to ranking results, is extracted from target prediction frame and obtain final prediction block.
6. the regional invasion real-time detection method according to claim 1 based on SSD, it is characterised in that: the end side is set
Standby is raspberry pie.
7. the real-time region intruding detection system based on SSD, it is characterised in that: include:
Training module obtains SSD model for the training set training after expanding according to data;
SSD model after compression processing for carrying out compression processing to SSD model, and is deployed to end side equipment by deployment module;
Detection module, for according to the identification classification after screening, image to be detected for being obtained by SSD model opposite end side apparatus into
Row detection, generates intrusion detection result;
Release module is updated to cloud when for by intrusion detection fructufy.
8. the real-time region intruding detection system according to claim 7 based on SSD, it is characterised in that: the detection mould
Block includes:
Acquiring unit, for obtaining the original predictive frame of image to be detected;
Confidence level determination unit, for determining classification confidence level according to SSD model;
Prediction block filter element, for according to classification confidence level in original predictive frame Low threshold prediction block and background forecast frame
It is filtered, obtains target prediction frame;
Decoding unit obtains location parameter for being decoded to target prediction frame;
Unit is demarcated, for demarcating the detected region in image to be detected according to location parameter;
Detection unit obtains intrusion detection result for detecting by SSD model to detected region.
9. the real-time region intruding detection system based on SSD, it is characterised in that: include:
At least one processor;
At least one processor, for storing at least one program;
When at least one described program is executed by least one described processor, so that at least one described processor is realized as weighed
Benefit requires the regional invasion real-time detection method described in any one of 1-6 based on SSD.
10. a kind of storage medium, wherein being stored with the executable instruction of processor, it is characterised in that: the processor is executable
Instruction enter when executed by the processor for executing the real-time region of any of claims 1-6 based on SSD such as
Invade detection method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910260758.1A CN110163081A (en) | 2019-04-02 | 2019-04-02 | Regional invasion real-time detection method, system and storage medium based on SSD |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910260758.1A CN110163081A (en) | 2019-04-02 | 2019-04-02 | Regional invasion real-time detection method, system and storage medium based on SSD |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110163081A true CN110163081A (en) | 2019-08-23 |
Family
ID=67638851
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910260758.1A Pending CN110163081A (en) | 2019-04-02 | 2019-04-02 | Regional invasion real-time detection method, system and storage medium based on SSD |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110163081A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781964A (en) * | 2019-10-28 | 2020-02-11 | 兰州交通大学 | Human body target detection method and system based on video image |
CN112464687A (en) * | 2020-11-19 | 2021-03-09 | 苏州摩比信通智能系统有限公司 | Graphic code processing method and device and terminal equipment |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180137642A1 (en) * | 2016-11-15 | 2018-05-17 | Magic Leap, Inc. | Deep learning system for cuboid detection |
CN108596053A (en) * | 2018-04-09 | 2018-09-28 | 华中科技大学 | A kind of vehicle checking method and system based on SSD and vehicle attitude classification |
CN108647655A (en) * | 2018-05-16 | 2018-10-12 | 北京工业大学 | Low latitude aerial images power line foreign matter detecting method based on light-duty convolutional neural networks |
CN108764228A (en) * | 2018-05-28 | 2018-11-06 | 嘉兴善索智能科技有限公司 | Word object detection method in a kind of image |
CN109446922A (en) * | 2018-10-10 | 2019-03-08 | 中山大学 | A kind of method for detecting human face of real-time robust |
CN109446925A (en) * | 2018-10-08 | 2019-03-08 | 中山大学 | A kind of electric device maintenance algorithm based on convolutional neural networks |
-
2019
- 2019-04-02 CN CN201910260758.1A patent/CN110163081A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180137642A1 (en) * | 2016-11-15 | 2018-05-17 | Magic Leap, Inc. | Deep learning system for cuboid detection |
CN108596053A (en) * | 2018-04-09 | 2018-09-28 | 华中科技大学 | A kind of vehicle checking method and system based on SSD and vehicle attitude classification |
CN108647655A (en) * | 2018-05-16 | 2018-10-12 | 北京工业大学 | Low latitude aerial images power line foreign matter detecting method based on light-duty convolutional neural networks |
CN108764228A (en) * | 2018-05-28 | 2018-11-06 | 嘉兴善索智能科技有限公司 | Word object detection method in a kind of image |
CN109446925A (en) * | 2018-10-08 | 2019-03-08 | 中山大学 | A kind of electric device maintenance algorithm based on convolutional neural networks |
CN109446922A (en) * | 2018-10-10 | 2019-03-08 | 中山大学 | A kind of method for detecting human face of real-time robust |
Non-Patent Citations (1)
Title |
---|
WEI LIU ET AL.: "Single Shot MultiBox Detector", 《ARXIV》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781964A (en) * | 2019-10-28 | 2020-02-11 | 兰州交通大学 | Human body target detection method and system based on video image |
CN112464687A (en) * | 2020-11-19 | 2021-03-09 | 苏州摩比信通智能系统有限公司 | Graphic code processing method and device and terminal equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111563557B (en) | Method for detecting target in power cable tunnel | |
CN111444821B (en) | Automatic identification method for urban road signs | |
CN111754498B (en) | Conveyor belt carrier roller detection method based on YOLOv3 | |
CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
CN111784685A (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
CN108509954A (en) | A kind of more car plate dynamic identifying methods of real-time traffic scene | |
CN112233073A (en) | Real-time detection method for infrared thermal imaging abnormity of power transformation equipment | |
CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
CN111914634B (en) | Automatic detection method and system for well lid class resisting complex scene interference | |
CN112070135A (en) | Power equipment image detection method and device, power equipment and storage medium | |
CN112257799A (en) | Method, system and device for detecting household garbage target | |
CN113869162A (en) | Violation identification method and system based on artificial intelligence | |
CN114841972A (en) | Power transmission line defect identification method based on saliency map and semantic embedded feature pyramid | |
CN113435407B (en) | Small target identification method and device for power transmission system | |
CN112232371A (en) | American license plate recognition method based on YOLOv3 and text recognition | |
CN114049572A (en) | Detection method for identifying small target | |
CN112668375B (en) | Tourist distribution analysis system and method in scenic spot | |
CN113221956A (en) | Target identification method and device based on improved multi-scale depth model | |
CN110163081A (en) | Regional invasion real-time detection method, system and storage medium based on SSD | |
CN111160100A (en) | Lightweight depth model aerial photography vehicle detection method based on sample generation | |
CN110765900B (en) | Automatic detection illegal building method and system based on DSSD | |
CN116342894B (en) | GIS infrared feature recognition system and method based on improved YOLOv5 | |
CN115830302B (en) | Multi-scale feature extraction fusion power distribution network equipment positioning identification method | |
CN111897993A (en) | Efficient target person track generation method based on pedestrian re-recognition | |
CN116597411A (en) | Method and system for identifying traffic sign by unmanned vehicle in extreme weather |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190823 |
|
RJ01 | Rejection of invention patent application after publication |