CN114049585A - Mobile phone action detection method based on motion foreground extraction - Google Patents
Mobile phone action detection method based on motion foreground extraction Download PDFInfo
- Publication number
- CN114049585A CN114049585A CN202111187354.8A CN202111187354A CN114049585A CN 114049585 A CN114049585 A CN 114049585A CN 202111187354 A CN202111187354 A CN 202111187354A CN 114049585 A CN114049585 A CN 114049585A
- Authority
- CN
- China
- Prior art keywords
- mobile phone
- motion
- module
- layer
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 65
- 230000009471 action Effects 0.000 title claims abstract description 49
- 238000000605 extraction Methods 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 10
- 238000004458 analytical method Methods 0.000 claims abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 39
- 238000012549 training Methods 0.000 claims description 33
- 238000009826 distribution Methods 0.000 claims description 21
- 238000011176 pooling Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 7
- 239000000284 extract Substances 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 6
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 abstract description 6
- 230000000694 effects Effects 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a mobile phone action detection method based on motion foreground extraction, which utilizes background modeling and background comparison analysis to extract motion foreground from a video sequence, segments the video sequence to obtain a small-size image containing a motion area, and then utilizes a convolutional neural network to detect a mobile phone target in the motion area image, thereby realizing the detection of the action of using a mobile phone. The invention fully utilizes the space-time information provided by the video, realizes the detection process from coarse to fine, has simple steps and high practicability, utilizes the monitoring camera which is installed and fixed in the places such as a laboratory, a conference room, a classroom and the like, can detect the condition that personnel use the mobile phone, and improves the monitoring effect.
Description
Technical Field
The invention relates to a motion detection method, in particular to a mobile phone motion detection method based on motion foreground extraction.
Background
With the rapid development of computer vision and the gradual improvement of computing power, intelligent video monitoring technology gradually appears in the public vision. The technology selects image processing, pattern recognition and other methods to effectively analyze the video collected by the monitoring camera, so that specific targets or abnormal conditions in the video image are automatically recognized, and early warning is timely given out. The application and popularization of the intelligent video monitoring technology greatly promote the improvement of social security, and have important significance in the aspects of improving the quality of life, defending disasters and the like. However, limited by the detection and identification algorithm and the hardware platform, some existing deployed intelligent video monitoring systems have the problems of low identification accuracy, poor real-time performance and the like, and a mature detection method which can be universally applied to all application scenes and application requirements is still lacking, so that a motion detection method which is good in performance and simple to implement needs to be provided for different scenes.
At present, in a fixed indoor scene such as a laboratory, a conference room, a classroom, and the like, a detection method using a mobile phone action mainly processes and analyzes a single-frame image in a video, and performs object detection with the mobile phone as a target, which is used as a basis for judging whether the mobile phone action is used. The method adopts a typical target detection algorithm based on deep learning to detect the mobile phone object, utilizes an image sample marked out of a mobile phone frame to train a detection model, selects image data of a single frame from a plurality of frames in a video as input during application, and detects the mobile phone target through the trained detection model, so that the mobile phone action detection can be realized, and the action of using the mobile phone is considered to exist when the mobile phone is detected. However, in the surveillance video, the mobile phone has a small size, an unobvious feature, a high similarity to other objects such as a notebook, and is easily influenced by factors such as the view field and angle of the surveillance camera to generate changes in shape and size, and when a user holds the mobile phone by hand, the mobile phone is easily blocked, and the mobile phone target is not clear in the image, so that problems such as false detection and missing detection are easily caused when the mobile phone is used as a basis for motion detection. In addition, the detection method is based on a single-frame image, only the spatial domain characteristics of the image are utilized, namely, whether the mobile phone action detection exists is judged by detecting the mobile phone target in a single time space.
Disclosure of Invention
The invention aims to provide a mobile phone action detection method based on motion foreground extraction, which solves the problems of false detection and missing detection existing when a single-frame image is used for mobile phone target detection at present.
A method for detecting actions of a mobile phone based on motion foreground extraction specifically comprises the following steps:
firstly, a mobile phone action detection system based on motion foreground extraction is built
Use cell-phone action detecting system based on motion prospect draws includes: the device comprises a background model building module, a motion foreground extracting module, an off-line training module and a mobile phone action detecting module.
The background model building module has the functions of: and fitting the background image by using a function to obtain a model, and updating the background model by combining the actual scene change of the video.
The motion foreground extraction module has the functions of: and comparing the video sequence with the background model, extracting the motion foreground, and segmenting a motion area through connectivity analysis.
The off-line training module has the functions of: determining a detection network model, constructing a motion area image sample library, and performing network offline training by using the sample library.
The function of using the mobile phone action detection module is as follows: and calculating the motion area image by using the network model, and detecting whether the action of using the mobile phone exists or not.
The second step background model construction module completes background modeling and background updating of the use scene
The background model building module accurately quantizes the background by using a Gaussian probability density function, fits each pixel point by adopting K Gaussian distributions, builds a background model aiming at a use scene and is expressed by a formula (1):
in the formula (1), the first and second groups,at time t, a certain pixel point (X, y) takes the value of Xt，wi,tIs the weight of the ith Gaussian distribution, eta (X)t,μi,t,∑i,t)、μi,tSum Σi,tRespectively, the ith gaussian probability density function, the mean and the covariance matrix, and n is the dimension of gaussian distribution.
Updating the background model in real time according to the change in the scene, and expressing by formula (2) to formula (4):
wi,t＝(1-α)wi,t-1+α (2)
μi,t＝(1-ρ)μi,t-1+ρXt (3)
∑i,t＝(1-ρ)∑i,t-1+ρ[(Xt-μi,t)(Xt-μi,t)T] (4)
in the formula (2) to the formula (4),
ρ is the update rate of the model. After the model is updated, calculating the pixel point of each pixel point in the image
And (3) sorting the values, selecting the largest B models as background models, namely the number of Gaussian distributions describing the background is B, T is a weight accumulation threshold, and T belongs to (0.5,1), and is expressed by a formula (5):
thirdly, the motion foreground extraction module extracts the motion foreground and divides the motion area to finish the crude extraction
The motion foreground extraction module compares a current frame image of the video sequence with the background image model for calculation, extracts the motion foreground, and divides a target area containing human motion from the current frame image according to the motion foreground.
Starting from the detection time t, the data is inputComparing the frame image with the background model, and calculating pixel values X one by onetAnd matching relation with the obtained B Gaussian distributions, wherein when the pixel value is matched with one of the previous B Gaussian distributions, the pixel point is a background point, otherwise, the pixel point is divided into a motion foreground. And calculating the pixel points in the frame image one by one according to the matching relation, and determining whether the pixel points can be matched with Gaussian distribution to obtain a binary image. The matching relationship is expressed by equation (6):
in the formula (6), the point with the gray value of 0 is a background point, and the point with the gray value of 1 is a moving foreground point.
And after the motion foreground is extracted, performing connectivity analysis on the motion foreground, and segmenting a target area image containing human motion from the current frame image to obtain a small-size image with the size of w x h, thereby completing coarse extraction.
The fourth step is that the off-line training module completes the determination and training of the detection of the mobile phone network
The off-line training module marks the mobile phone in the motion area image obtained by the motion foreground extraction module, completes construction of a training sample library, determines and constructs a deep convolutional neural network model for detecting the mobile phone from the image containing the human motion area, determines the number of network layers, definition of each layer, the number of convolutional surfaces of each layer, the size of a convolutional kernel, the size of a pooling layer, a computation function of the pooling layer, an activation function and a loss function, and then performs off-line learning training on unknown parameters of each convolutional kernel of the deep convolutional neural network by using the constructed sample library.
The convolutional layer elementary operation of the network is expressed by formula (7):
Xa,b+1＝f(∑Xb·Wa,b+ba,b) (7)
in the formula (7), f is an activation function, Wa,bAnd ba,bThe convolution kernel and the offset value X of the a-th convolution surface in the b-th layer of the network respectivelybRepresenting inputs to channels of layer b of the network, Xa,b+1Representing the output of the a-th volume area of the b-th layer of the network.
The basic operation of the pooling layer of the network is represented by equation (8):
Xa,b+1＝p(Xa,b) (8)
in the formula (8), Xa,bRepresenting the input, X, of the b-th channel of the networka,b+1Representing the output of the a channel at the b layer of the network, p is the pooling layer calculation function.
The network full-connection layer basic operation is expressed by the formula (9):
yb＝f(∑xb·wb+bb) (9)
in formula (9), wbAnd bbRespectively representing weight and bias, x, of the b-th layer in the full connection layerbRepresenting the input of the b-th layer of the fully connected layer, ybRepresenting the output of the b-th layer in the fully connected layer.
During the training process, the parameters are updated with equation (10):
in the formula (10), η represents the learning rate designed in the training process, and the superscript (m) represents the calculated amount of the mth iteration process.
After iterative computation, the loss function loss is converged to the minimum value, a deep convolution neural network model suitable for detecting the mobile phone is obtained, and the off-line preparation stage is completed.
Fifthly, finishing final detection by using a mobile phone action detection module
And a mobile phone action detection module is used for detecting the mobile phone by utilizing the network model obtained by the offline training module, inputting the motion area image obtained by the motion foreground extraction module into the network model for calculation, and outputting a mobile phone detection result. When the mobile phone is detected in the moving area image by using the mobile phone action detection module, the action of using the mobile phone is considered to exist; when the mobile phone is not detected in the motion area image, it is considered that there is no motion using the mobile phone.
Therefore, mobile phone action detection based on motion foreground extraction is achieved.
The invention realizes the detection of the action of the mobile phone, extracts the motion foreground for coarse detection in the use scene, detects the mobile phone for fine detection in the small-size image of the motion foreground obtained by the coarse detection by utilizing the deep learning network, realizes the detection steps from coarse to fine, fully utilizes the space-time characteristic information and can achieve the effect of improving the detection accuracy.
Detailed Description
A method for detecting actions of a mobile phone based on motion foreground extraction specifically comprises the following steps:
firstly, a mobile phone action detection system based on motion foreground extraction is built
Use cell-phone action detecting system based on motion prospect draws includes: the device comprises a background model building module, a motion foreground extracting module, an off-line training module and a mobile phone action detecting module.
The background model building module has the functions of: and fitting the background image by using a function to obtain a model, and updating the background model by combining the actual scene change of the video.
The motion foreground extraction module has the functions of: and comparing the video sequence with the background model, extracting the motion foreground, and segmenting a motion area through connectivity analysis.
The off-line training module has the functions of: determining a detection network model, constructing a motion area image sample library, and performing network offline training by using the sample library.
The function of using the mobile phone action detection module is as follows: and calculating the motion area image by using the network model, and detecting whether the action of using the mobile phone exists or not.
The second step background model construction module completes background modeling and background updating of the use scene
The background model building module accurately quantizes the background by using a Gaussian probability density function, fits each pixel point by adopting K Gaussian distributions, builds a background model aiming at a use scene and is expressed by a formula (1):
in the formula (1), a certain pixel point (X, y) takes the value of X at the moment tt，wi,tIs the weight of the ith Gaussian distribution, eta (X)t,μi,t,∑i,t)、μi,tSum Σi,tRespectively, the ith gaussian probability density function, the mean and the covariance matrix, and n is the dimension of gaussian distribution.
Updating the background model in real time according to the change in the scene, and expressing by formula (2) to formula (4):
wi,t＝(1-α)wi,t-1+α (2)
μi,t＝(1-ρ)μi,t-1+ρXt (3)
∑i,t＝(1-ρ)∑i,t-1+ρ[(Xt-μi,t)(Xt-μi,t)T] (4)
in the formula (2) to the formula (4),
ρ is the update rate of the model. After the model is updated, calculating the pixel point of each pixel point in the image
And (3) sorting the values, selecting the largest B models as background models, namely the number of Gaussian distributions describing the background is B, T is a weight accumulation threshold, and T belongs to (0.5,1), and is expressed by a formula (5):
thirdly, the motion foreground extraction module extracts the motion foreground and divides the motion area to finish the crude extraction
The motion foreground extraction module compares a current frame image of the video sequence with the background image model for calculation, extracts the motion foreground, and divides a target area containing human motion from the current frame image according to the motion foreground.
Inputting the frame image from the detection time t, comparing the frame image with a background model, and calculating pixel values X one by onetAnd matching relation with the obtained B Gaussian distributions, wherein when the pixel value is matched with one of the previous B Gaussian distributions, the pixel point is a background point, otherwise, the pixel point is divided into a motion foreground. And calculating the pixel points in the frame image one by one according to the matching relation, and determining whether the pixel points can be matched with Gaussian distribution to obtain a binary image. The matching relationship is expressed by equation (6):
in the formula (6), the point with the gray value of 0 is a background point, and the point with the gray value of 1 is a moving foreground point.
And after the motion foreground is extracted, performing connectivity analysis on the motion foreground, and segmenting a target area image containing human motion from the current frame image to obtain a small-size image with the size of w x h, thereby completing coarse extraction.
The fourth step is that the off-line training module completes the determination and training of the detection of the mobile phone network
The off-line training module marks the mobile phone in the motion area image obtained by the motion foreground extraction module, completes construction of a training sample library, determines and constructs a deep convolutional neural network model for detecting the mobile phone from the image containing the human motion area, determines the number of network layers, definition of each layer, the number of convolutional surfaces of each layer, the size of a convolutional kernel, the size of a pooling layer, a computation function of the pooling layer, an activation function and a loss function, and then performs off-line learning training on unknown parameters of each convolutional kernel of the deep convolutional neural network by using the constructed sample library.
The convolutional layer elementary operation of the network is expressed by formula (7):
Xa,b+1＝f(∑Xb·Wa,b+ba,b) (7)
in the formula (7), f is an activation function, Wa,bAnd ba,bThe convolution kernel and the offset value X of the a-th convolution surface in the b-th layer of the network respectivelybRepresenting inputs to channels of layer b of the network, Xa,b+1Representing the output of the a-th volume area of the b-th layer of the network.
The basic operation of the pooling layer of the network is represented by equation (8):
Xa,b+1＝p(Xa,b) (8)
in the formula (8), Xa,bRepresenting the input, X, of the b-th channel of the networka,b+1Representing the output of the a channel at the b layer of the network, p is the pooling layer calculation function.
The network full-connection layer basic operation is expressed by the formula (9):
yb＝f(∑xb·wb+bb) (9)
in formula (9), wbAnd bbRespectively representing weight and bias, x, of the b-th layer in the full connection layerbRepresenting the input of the b-th layer of the fully connected layer, ybRepresenting the output of the b-th layer in the fully connected layer.
During the training process, the parameters are updated with equation (10):
in the formula (10), η represents the learning rate designed in the training process, and the superscript (m) represents the calculated amount of the mth iteration process.
After iterative computation, the loss function loss is converged to the minimum value, a deep convolution neural network model suitable for detecting the mobile phone is obtained, and the off-line preparation stage is completed.
Fifthly, finishing final detection by using a mobile phone action detection module
And a mobile phone action detection module is used for detecting the mobile phone by utilizing the network model obtained by the offline training module, inputting the motion area image obtained by the motion foreground extraction module into the network model for calculation, and outputting a mobile phone detection result. When the mobile phone is detected in the moving area image by using the mobile phone action detection module, the action of using the mobile phone is considered to exist; when the mobile phone is not detected in the motion area image, it is considered that there is no motion using the mobile phone.
Therefore, mobile phone action detection based on motion foreground extraction is achieved.
Claims (5)
1. A mobile phone action detection method based on motion foreground extraction is characterized by comprising the following specific steps:
firstly, a mobile phone action detection system based on motion foreground extraction is built
Use cell-phone action detecting system based on motion prospect draws includes: the device comprises a background model construction module, a motion foreground extraction module, an off-line training module and a mobile phone action detection module;
the second step background model construction module completes background modeling and background updating of the use scene
The background model building module accurately quantizes the background by using a Gaussian probability density function, fits each pixel point by adopting K Gaussian distributions, builds a background model aiming at a use scene and is expressed by a formula (1):
in the formula (1), a certain pixel point (X, y) takes the value of X at the moment tt，wi,tIs the weight of the ith Gaussian distribution, eta (X)t,μi,t,∑i,t)、μi,tSum Σi,tRespectively an ith Gaussian probability density function, a mean value and a covariance matrix, wherein n is the dimensionality of Gaussian distribution;
updating the background model in real time according to the change in the scene, and expressing by formula (2) to formula (4):
wi,t＝(1-α)wi,t-1+α (2)
μi,t＝(1-ρ)μi,t-1+ρXt (3)
∑i,t＝(1-ρ)∑i,t-1+ρ[(Xt-μi,t)(Xt-μi,t)T] (4)
in the formula (2) to the formula (4),
rho is the update rate of the model; after the model is updated, calculating the pixel point of each pixel point in the image
And (3) sorting the values, selecting the largest B models as background models, namely the number of Gaussian distributions describing the background is B, T is a weight accumulation threshold, and T belongs to (0.5,1), and is expressed by a formula (5):
thirdly, the motion foreground extraction module extracts the motion foreground and divides the motion area to finish the crude extraction
The motion foreground extraction module compares a current frame image of the video sequence with the background image model for calculation, extracts a motion foreground, and divides a target area containing human motion from the current frame image according to the motion foreground;
inputting the frame image from the detection time t, comparing the frame image with a background model, and calculating pixel values X one by onetMatching relation with the obtained B Gaussian distributions, wherein when the pixel value is matched with one of the previous B Gaussian distributions, the pixel point is a background point, otherwise, the pixel point is divided into a motion foreground; calculating pixel points in the frame image one by one according to a matching relation, and determining whether the pixel points can be matched with Gaussian distribution to obtain a binary image; the matching relationship is expressed by equation (6):
in the formula (6), a point with a gray value of 0 is a background point, and a point with a gray value of 1 is a point motion foreground point;
after the motion foreground is extracted, performing connectivity analysis on the motion foreground, and segmenting a target area image containing human motion from the current frame image to obtain a small-size image with the size of w x h, thereby completing coarse extraction;
the fourth step is that the off-line training module completes the determination and training of the detection of the mobile phone network
The off-line training module is used for marking the mobile phone in the motion area image obtained by the motion foreground extraction module, completing construction of a training sample library, determining and constructing a deep convolutional neural network model, detecting the mobile phone from the image containing the human motion area, determining the number of network layers, each layer definition, the number of convolutional surfaces, the size of a convolutional kernel, the size of a pooling layer, a pooling layer calculation function, an activation function and a loss function, and then performing off-line learning training on unknown parameters of each convolutional kernel of the deep convolutional neural network by using the constructed sample library;
the convolutional layer elementary operation of the network is expressed by formula (7):
Xa,b+1＝f(∑Xb·Wa,b+ba,b) (7)
in the formula (7), f is an activation function, Wa,bAnd ba,bThe convolution kernel and the offset value X of the a-th convolution surface in the b-th layer of the network respectivelybRepresenting inputs to channels of layer b of the network, Xa,b+1An output representing a layer b, layer a, volume area of the network;
the basic operation of the pooling layer of the network is represented by equation (8):
Xa,b+1＝p(Xa,b) (8)
in the formula (8), Xa,bRepresenting the input, X, of the b-th channel of the networka,b+1Representing the output of the a channel of the b layer of the network, and p is a pooling layer calculation function;
the network full-connection layer basic operation is expressed by the formula (9):
yb＝f(∑xb·wb+bb) (9)
in formula (9), wbAnd bbRespectively representing weight and bias, x, of the b-th layer in the full connection layerbRepresenting the input of the b-th layer of the fully connected layer, ybRepresents the output of the b-th layer in the fully connected layer;
during the training process, the parameters are updated with equation (10):
in the formula (10), eta represents the learning rate designed in the training process, and superscript (m) represents the calculated amount of the mth step of the iterative process;
after iterative computation, converging the loss function loss to the minimum value to obtain a deep convolution neural network model suitable for detecting the mobile phone, and completing an off-line preparation stage;
fifthly, finishing final detection by using a mobile phone action detection module
Detecting the mobile phone by using the mobile phone action detection module and the network model obtained by the offline training module, inputting the motion area image obtained by the motion foreground extraction module into the network model for calculation, and outputting a mobile phone detection result; when the mobile phone is detected in the moving area image by using the mobile phone action detection module, the action of using the mobile phone is considered to exist; when the mobile phone is not detected in the moving area image, the action of using the mobile phone is not considered to exist;
therefore, mobile phone action detection based on motion foreground extraction is achieved.
2. The method for detecting actions of the mobile phone based on the motion foreground extraction as claimed in claim 1, wherein the background model building module has the functions of: and fitting the background image by using a function to obtain a model, and updating the background model by combining the actual scene change of the video.
3. The method for detecting actions of a mobile phone based on motion foreground extraction as claimed in claim 1, wherein the motion foreground extraction module functions as: and comparing the video sequence with the background model, extracting the motion foreground, and segmenting a motion area through connectivity analysis.
4. The method for detecting actions of a mobile phone based on motion foreground extraction as claimed in claim 1, wherein the function of the off-line training module is: determining a detection network model, constructing a motion area image sample library, and performing network offline training by using the sample library.
5. The method for detecting actions of using mobile phone based on motion foreground extraction as claimed in claim 1, wherein the function of the module for detecting actions of using mobile phone is: and calculating the motion area image by using the network model, and detecting whether the action of using the mobile phone exists or not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111187354.8A CN114049585B (en) | 2021-10-12 | 2021-10-12 | Mobile phone operation detection method based on motion prospect extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111187354.8A CN114049585B (en) | 2021-10-12 | 2021-10-12 | Mobile phone operation detection method based on motion prospect extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114049585A true CN114049585A (en) | 2022-02-15 |
| CN114049585B CN114049585B (en) | 2024-04-02 |
Family
ID=80205355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111187354.8A Active CN114049585B (en) | 2021-10-12 | 2021-10-12 | Mobile phone operation detection method based on motion prospect extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114049585B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107133974A (en) * | 2017-06-02 | 2017-09-05 | 南京大学 | The vehicle type classification method that Gaussian Background modeling is combined with Recognition with Recurrent Neural Network |
| CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
| WO2019237567A1 (en) * | 2018-06-14 | 2019-12-19 | 江南大学 | Convolutional neural network based tumble detection method |
-
2021
- 2021-10-12 CN CN202111187354.8A patent/CN114049585B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107133974A (en) * | 2017-06-02 | 2017-09-05 | 南京大学 | The vehicle type classification method that Gaussian Background modeling is combined with Recognition with Recurrent Neural Network |
| CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
| WO2019237567A1 (en) * | 2018-06-14 | 2019-12-19 | 江南大学 | Convolutional neural network based tumble detection method |
Non-Patent Citations (1)
| Title |
|---|
| 赵宏伟;冯嘉;臧雪柏;宋波涛;: "一种实用的运动目标检测和跟踪算法", 吉林大学学报(工学版), no. 2, 30 September 2009 (2009-09-30), pages 386 - 390 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114049585B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110660082B (en) | Target tracking method based on graph convolution and trajectory convolution network learning | |
| CN108154118B (en) | A kind of target detection system and method based on adaptive combined filter and multistage detection | |
| CN106778595B (en) | Method for detecting abnormal behaviors in crowd based on Gaussian mixture model | |
| US20230289979A1 (en) | A method for video moving object detection based on relative statistical characteristics of image pixels | |
| CN111709311B (en) | Pedestrian re-identification method based on multi-scale convolution feature fusion | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN111476817A (en) | Multi-target pedestrian detection tracking method based on yolov3 | |
| CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
| Feng et al. | Online learning with self-organizing maps for anomaly detection in crowd scenes | |
| Trnovszký et al. | Comparison of background subtraction methods on near infra-red spectrum video sequences | |
| CN110728694A (en) | Long-term visual target tracking method based on continuous learning | |
| CN110728216A (en) | Unsupervised pedestrian re-identification method based on pedestrian attribute adaptive learning | |
| Daramola et al. | Automatic vehicle identification system using license plate | |
| Mahapatra et al. | Human recognition system for outdoor videos using Hidden Markov model | |
| CN114240997A (en) | Intelligent building online cross-camera multi-target tracking method | |
| CN111273288A (en) | Radar unknown target identification method based on long-term and short-term memory network | |
| Kumar et al. | Background subtraction based on threshold detection using modified K-means algorithm | |
| Yang et al. | A method of pedestrians counting based on deep learning | |
| CN109272036B (en) | Random fern target tracking method based on depth residual error network | |
| CN114038011A (en) | Method for detecting abnormal behaviors of human body in indoor scene | |
| CN117193121B (en) | Control system of coating machine die head | |
| Dewan et al. | Detection of object in motion using improvised background subtraction algorithm | |
| Elbaşi | Fuzzy logic-based scenario recognition from video sequences | |
| CN111223126A (en) | Cross-view-angle trajectory model construction method based on transfer learning | |
| CN114049585B (en) | Mobile phone operation detection method based on motion prospect extraction |
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 |