CN110765982A - Video smoke detection method based on change accumulation graph and cascaded depth network - Google Patents
Video smoke detection method based on change accumulation graph and cascaded depth network Download PDFInfo
- Publication number
- CN110765982A CN110765982A CN201911078837.7A CN201911078837A CN110765982A CN 110765982 A CN110765982 A CN 110765982A CN 201911078837 A CN201911078837 A CN 201911078837A CN 110765982 A CN110765982 A CN 110765982A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- smoke
- images
- change
- 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
Images
Classifications
-
- 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
- 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
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/254—Analysis of motion involving subtraction of images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a video smoke detection method based on a change accumulation graph and a cascaded depth network. The concept of a change accumulation graph is provided, the YUV color space of a multi-frame video image is converted into the change accumulation graph, and the motion characteristic and the color change characteristic of smoke are described; then designing a cascade deep network, and cascading two layers of convolution layers on each convolution layer of the VGG16 network to increase the network depth and enhance the characteristic identification capability; through cascading a VGG16 network and a Resnet50 network model, more detailed characteristics of the smoke image are extracted, and the smoke characteristic identification capability is improved.
Description
Technical Field
The invention mainly relates to a video smoke detection method based on a change accumulation graph and a cascaded depth network.
Background
Smoke is one of the main signals for detecting fires. It is well known that fire is extremely harmful to human beings, which not only causes property loss and resource destruction, but also may endanger life. The fire hazard is detected and found as early as possible, and the hazard of the fire is reduced as far as possible. The traditional smoke sensor can detect smoke after the smoke enters a detector to reach a certain concentration, and the application effect is poor in an outdoor open environment. In recent years, video surveillance technology is widely applied in various industries, and provides support for vision-based smoke detection technology. The smoke detection based on the computer vision technology can realize long-distance and large-range detection, and is very suitable for application occasions requiring large-range fire detection, such as forests, scenic spots and the like.
The basic idea of detecting fire based on computer vision technology is as follows: the method comprises the steps of extracting personalized features of the smoke, classifying the features, and distinguishing smoke targets from non-smoke targets. The extraction of smoke features is a key step of smoke detection, common smoke features include color features, texture features, shape features, motion features and the like, and some technologies improve the performance of smoke detection by integrating the features. For the smoke target, the smoke belongs to a non-rigid target, the shape changes at multiple ends, the color is not obvious, the texture layering is not strong, and the motion is not obvious, so the smoke detection based on computer vision is very difficult. In recent years, the deep learning technology has a good application effect, and a plurality of new ideas are brought to smoke detection. The document 'connected neural network for video fire and smoke detection' adopts a Convolutional neural network to detect smoke, can automatically extract smoke features with strong identification capability, and improves the accuracy of smoke detection. The document 'deep normalization and conditional Neural Network for Image Smoke Detection' proposes a DNCNN deep Network model for Smoke Detection, and adopts a normalization method to accelerate the training process of a deep Network and improve the Detection precision of a Smoke target. However, these methods do not combine the motion characteristics of smoke, and there is also a phenomenon of misclassification when distinguishing the fog-like targets such as clouds and fog.
Disclosure of Invention
In order to solve the problems, the invention particularly provides a video smoke detection method based on a change accumulation graph and a cascade depth network, which is mainly characterized in that the concept of the change accumulation graph is provided, on a YUV color model, adjacent frame images in a Y space are subjected to change detection, and the change amount is accumulated to obtain a change accumulation image which is used for reflecting the motion diffusion characteristic of smoke; and carrying out mean filtering processing on the accumulated images of the U space and the V space to obtain an accumulated image for reflecting the color change characteristic of the smoke. And for the transformed video frame image data, providing a cascaded deep network model to extract features and classify the features, wherein the cascaded deep network model comprises two cascaded deep network models of VGG16 and ResNet50, and a cascaded convolutional layer is introduced into a convolutional layer of VGG16, so that the depth of the network model is enhanced, and the smoke feature discrimination capability is improved.
In order to improve the detection accuracy of the smoke target in the video, the invention firstly provides a concept of a change accumulation graph, converts the YUV color space of a multi-frame video image into the change accumulation graph and describes the motion characteristic and the color change characteristic of the smoke; then designing a cascade deep network, and cascading two layers of convolution layers on each convolution layer of the VGG16 network to increase the network depth and enhance the characteristic identification capability; through cascading a VGG16 network and a Resnet50 network model, more detailed characteristics of the smoke image are extracted, and the smoke characteristic identification capability is improved.
The detection method comprises the following steps:
(1) cumulative graph of changes
The cumulative graph of changes is: on a YUV color model, change detection is carried out on adjacent frame images in a Y space, and the change quantity is accumulated to obtain a change accumulation image which is used for reflecting the motion diffusion characteristic of smoke; carrying out mean value filtering processing on the accumulated images of the U space and the V space to obtain an accumulated image for reflecting the color change characteristic of the smoke;
on Y space, calculating a binary image of a changed image by adopting an inter-frame difference method, specifically, calculating a Y space vector image Y of a k frame imagekBinary image of change image
Is shown as
Wherein T is a fixed threshold, and the empirical value T is 10;
then, the adjacent N pieces of variation image binary images are subjected to accumulation summation to obtain variation accumulation images
Is shown as
N is less than 255, change the cumulative image FkThe image is regarded as a gray image;
in the U space, the U space vector images of the adjacent N frames of images are accumulated and summed and subjected to average operation to obtain an accumulated image
Is shown as
Similarly, in the V space, the accumulated summation is carried out on the V space vector images of the adjacent N frames of images, and the average operation is carried out to obtain an accumulated image
Is shown as
The YUV image of the k-th frame image is converted into a change accumulation image whose Y-space component reflects the motion diffusion characteristic of smoke, U, V-space component reflects the color change characteristic of smoke,
(2) cascaded deep network model
Firstly, the VGG16 network is improved, a cascade convolutional layer is introduced into the convolutional layer, secondly, a ResNet50 network is introduced,
(2.1) cascaded deep networks
The VGG16 network comprises 13 convolutional layers and 3 full-connection layers, a ResNet50 network comprises 49 convolutional layers and 1 full-connection layer, and the VGG16 network and the ResNet50 network are cascaded, wherein a VGG16 network feature extractor is a convolutional layer part of VGG16, a ResNet50 network feature extractor is a residual block part of a ResNet50 network, and the specific implementation method comprises the following steps: firstly, freezing all network layers before a full-connection layer in a deep convolutional neural network, so that parameters in the frozen network layers are not subjected to gradient updating in the training process of a model, and the optimized parameters are only all parameters of the full-connection layer which is not frozen; then, connecting the features extracted by the feature extractor, and reconstructing a full connection layer for bearing the whole model output classification work; finally, a sigmoid activation function is adopted for output, the output value is greater than or equal to 0.5, and the smoke is considered to be not smoke when the output value is less than 0.5;
(2.2) cascaded convolutional layers
Two convolutional layers are cascaded after each convolutional layer of the VGG16 network, average operation is carried out on output data and input data after the operation of the convolutional layers, then the output data after the average operation is used as the input data of the cascaded convolutional layers, one convolutional layer operation is carried out again, and the like, three convolutional layers are cascaded in total.
The invention has the advantages that the movement characteristics and the color change characteristics of the smoke are comprehensively reflected through the change accumulation graph, the smoke detail characteristics with stronger identification capability are extracted through the cascade deep network model and are subjected to self-adaptive classification, and finally the accuracy of smoke detection is improved.
Drawings
FIG. 1 is a cascaded deep network;
FIG. 2 compares a cascaded convolutional layer to a conventional convolutional layer;
FIG. 3 is a flow chart of the present invention.
Detailed Description
The invention relates to a video smoke detection method based on a change accumulation graph and a cascade depth network, which converts a YUV color space of a plurality of frames of video images into the change accumulation graph to describe the motion characteristic and the color change characteristic of smoke; then designing a cascade deep network, and cascading two layers of convolution layers on each convolution layer of the VGG16 network to increase the network depth and enhance the characteristic identification capability; through cascading a VGG16 network and a Resnet50 network model, more detailed characteristics of the smoke image are extracted, and the smoke characteristic identification capability is improved. The method comprises the following steps:
(1) cumulative graph of changes
By analyzing the characteristics of the smoke, the fact that the brightness of the smoke can be greatly changed along with the change of concentration and components is found, and the smoke can be very bright or very dark; the smoke chromaticity change is small, and generally speaking, the smoke color saturation is small; from the generation and development process of the smoke, the smoke has obvious upward diffusion characteristics and is represented as follows: the smoke will continuously spread upward and around from the ignition point. Based on the characteristics, the invention provides a concept of a change accumulation graph, and the basic idea is as follows: on a YUV color model, change detection is carried out on adjacent frame images in a Y space, and the change quantity is accumulated to obtain a change accumulation image which is used for reflecting the motion diffusion characteristic of smoke; and carrying out mean filtering processing on the accumulated images of the U space and the V space to obtain an accumulated image for reflecting the color change characteristic of the smoke.
In Y space, the invention adopts an interframe difference method to calculate a binary image of the changed image. Specifically, the Y space vector image Y for the k frame imagekBinary image of change image
Can be expressed as
Wherein, T is a fixed threshold, and the invention takes an empirical value T as 10.
Then, the adjacent N pieces of variation image binary images are subjected to accumulation summation to obtain variation accumulation images
Is shown as
In the present invention, N is less than 255, and thus the change accumulation image FkCan be regarded as a gray scale image.
In the U space, the U space vector images of the adjacent N frames of images are accumulated and summed and subjected to average operation to obtain an accumulated image
Is shown as
Similarly, in the V space, the accumulated summation is carried out on the V space vector images of the adjacent N frames of images, and the average operation is carried out to obtain an accumulated image
Is shown as
Thus, the YUV image of our k-th frame image is converted into a change accumulation image, the Y-space component of which may reflect the motion spread characteristics of smoke, and the U, V-space component may reflect the color change characteristics of smoke. Compared with the original YUV image, the change accumulated image can better highlight the change characteristic of the smoke, and therefore, more remarkable and robust smoke features can be extracted based on the change accumulated image.
(2) Cascaded deep network model
The deep network shows good performance in the aspect of image classification, for example, the convolution layer and the pooling layer of the commonly-used VGG16 network both adopt the same kernel function, and the convolution layer and the pooling layer are stacked to form a convolution block structure, so that the deep network structure has the advantages of simple structure, easiness in forming a deep network structure and the like. However, the maximum pooling is adopted among blocks of the VGG16 network, and for the image with abundant details such as smoke, part of important features of the original image may be lost. In order to make up for the defect of feature loss, the invention firstly improves the VGG16 network, introduces the cascade convolution layer in the convolution layer and improves the feature identification capability; and secondly, a ResNet50 network is introduced, and a ResNet50 network adopts jump connection to form a residual block, so that image information is transmitted to a deeper layer of a neural network, and the loss of important characteristics of the smoke image can be avoided. Meanwhile, the problem of under-fitting caused by gradient disappearance can be avoided, so that the expression capability of the model is effectively improved while the network hierarchy is deepened. Compared with single depth network models such as VGG16 and ResBet50, the cascaded depth network model provided by the invention can extract richer smoke image detail characteristics, and enhances the distinguishing capability of the characteristics on smoke images and similar smoke images.
(2.1) cascaded deep networks
The VGG16 network is composed of 13 convolutional layers and 3 full-connection layers, and the biggest characteristic is that features are extracted through combination and stacking of 3 x 3 filters, and the distinguishing capability of the features is strong. The ResNet50 network contains 49 convolutional layers and 1 fully-connected layer. Because the network is added with the identity mapping layer and directly connected with the shallow network and the deep network, the network can be ensured not to be degraded along with the increase of the depth, and the convergence effect is good. The invention cascades a VGG16 network and a ResNet50 network, as shown in FIG. 1, wherein a VGG16 network Feature Extractor (VGG16Feature Extractor) is a convolution layer part of VGG16, and a ResNet50 network Feature Extractor (ResNet50Feature Extractor) is a residual block part of a ResNet50 network. Through the cascade deep network, firstly, the detailed features of the smoke image can be extracted by using a small window filter of the VGG16, and secondly, the problems of loss and under-fitting of the features of the VGG16 network can be solved by means of the ResNet50 network, and the deeper features can be extracted. The specific implementation method comprises the following steps: firstly, freezing all network layers before a full-connection layer in the deep convolutional neural network, and enabling parameters in the frozen network layers not to be subjected to gradient updating in the training process of the model, wherein the parameters capable of being optimized are only all parameters of the full-connection layer which is not frozen. Then, the features extracted by the feature extractor are connected, and a full connection layer which bears the whole model output classification work is reconstructed. And finally, outputting by adopting a sigmoid activation function, and considering the smoke when the output value is greater than or equal to 0.5, and considering the smoke not to be smoke when the output value is less than 0.5.
(2.2) cascaded convolutional layers
Although the VGG16 network increases the extraction capability of detail features through the combination and stacking of 3 × 3 filters, the distinguishing capability of features is not strong enough for smoke-like objects such as smoke images and cloud mist. To further increase the significance of the feature, the present invention concatenates two convolutional layers after each convolutional layer of the VGG16 network. As shown in fig. 2(a), after the conventional convolutional layer performs convolutional layer operation on the input data, the feature dimension of the output data is identical to that of the input data. As shown in fig. 2(b), the present invention performs an averaging operation on the input data and the output data after the convolutional layer operation, and then performs another convolutional layer operation using the output data after the averaging operation as the input data of the cascaded convolutional layers. By analogy, three convolutional layers are cascaded, so that the purpose of enhancing the depth of the network is to improve the identification capability of the features.
(3) Algorithm implementation
The method comprises the following implementation steps:
step1 the size of the input k frame image is scaled to 224 x 224 using a bilinear interpolation method.
Step 2: if k is larger than N, the requirement of calculating the change accumulated image is met, and the next step is carried out; otherwise, buffering the current frame image, wherein k is k +1, and returning to Step 1;
step 3: calculating a transformation accumulated image of the k frame image according to formula (1) and formula (2);
step 4: features are extracted by a VGG16feature extractor and a ResNet50feature extractor respectively, wherein the convolutional layer of VGG16 adopts the cascade convolutional layer structure. After the 7 × 7 × 2048 ═ 100352 dimensional features extracted by the ResNet50 network are placed in the 7 × 7 × 512 ═ 25088 dimensional features extracted by the VGG16 network, a 100352+25088 ═ 125440 dimensional feature is constructed.
Step 5: and (3) regarding each feature as a node of the neuron, connecting the extracted features in a Full Connection (FC) mode, and outputting 1024 neuron nodes.
Step 6: in order to prevent the overfitting phenomenon of the convolutional neural network, a Dropout method is adopted, and neural units are randomly selected according to a certain probability P (in the invention, P is 0.3) and discarded.
Step 7: the remaining neural units after Dropout are still connected in a Full Connection (FC) mode, and 128 neuron nodes are output.
Step 8: still using the Dropout method, neural units are randomly selected with a certain probability P (in the present invention P ═ 0.3) and discarded.
Step 9: and outputting the rest nerve units through a Sigmoid activation function, and judging the nerve units as smoke images if the output value is greater than or equal to 0.5, otherwise judging the nerve units as non-smoke images. The transformed accumulated image of the k-th frame image and the current frame image are buffered regardless of whether an alarm is output, k being k +1, and the process returns to Step 1.
Claims (3)
1. The video smoke detection method based on the change accumulation graph and the cascade depth network provides a concept of the change accumulation graph, and is characterized in that on a YUV color model, change detection is carried out on adjacent frame images in a Y space, and variation is accumulated to obtain a change accumulation image which is used for reflecting the motion diffusion characteristic of smoke; carrying out mean value filtering processing on the accumulated images of the U space and the V space to obtain an accumulated image for reflecting the color change characteristic of the smoke;
for the transformed video frame image data, providing a cascade deep network model to extract features and classifying, wherein the cascade deep network model comprises two deep network models of cascade VGG16 and ResNet 50;
and a cascade convolutional layer is introduced into the convolutional layer of the VGG16, so that the depth of a network model is enhanced, and the smoke characteristic identification capability is improved.
2. The video smoke detection method based on the change accumulation graph and the cascaded deep network as claimed in claim 1, wherein the detection method comprises the following steps:
(1) cumulative graph of changes
The cumulative graph of changes is: on a YUV color model, change detection is carried out on adjacent frame images in a Y space, and the change quantity is accumulated to obtain a change accumulation image which is used for reflecting the motion diffusion characteristic of smoke; carrying out mean value filtering processing on the accumulated images of the U space and the V space to obtain an accumulated image for reflecting the color change characteristic of the smoke;
on Y space, calculating a binary image of a changed image by adopting an inter-frame difference method, specifically, calculating a Y space vector image Y of a k frame imagekBinary image of change image
Is shown as
Wherein T is a fixed threshold, and the empirical value T is 10;
then, the adjacent N pieces of variation image binary images are subjected to accumulation summation to obtain variation accumulation images
Is shown as
N is less than 255, change the cumulative image FkThe image is regarded as a gray image;
in U spaceAccumulating and summing the U space vector images of the adjacent N frames of images and carrying out average operation to obtain an accumulated image
Is shown as
Similarly, in the V space, the accumulated summation is carried out on the V space vector images of the adjacent N frames of images, and the average operation is carried out to obtain an accumulated image
Is shown as
The YUV image of the k-th frame image is converted into a change accumulation image whose Y-space component reflects the motion diffusion characteristic of smoke, U, V-space component reflects the color change characteristic of smoke,
(2) cascaded deep network model
Firstly, the VGG16 network is improved, a cascade convolutional layer is introduced into the convolutional layer, secondly, a ResNet50 network is introduced,
(2.1) cascaded deep networks
The VGG16 network comprises 13 convolutional layers and 3 full-connection layers, a ResNet50 network comprises 49 convolutional layers and 1 full-connection layer, and the VGG16 network and the ResNet50 network are cascaded, wherein a VGG16 network feature extractor is a convolutional layer part of VGG16, a ResNet50 network feature extractor is a residual block part of a ResNet50 network, and the specific implementation method comprises the following steps: firstly, freezing all network layers before a full-connection layer in a deep convolutional neural network, so that parameters in the frozen network layers are not subjected to gradient updating in the training process of a model, and the optimized parameters are only all parameters of the full-connection layer which is not frozen; then, connecting the features extracted by the feature extractor, and reconstructing a full connection layer for bearing the whole model output classification work; finally, a sigmoid activation function is adopted for output, the output value is greater than or equal to 0.5, and the smoke is considered to be not smoke when the output value is less than 0.5;
(2.2) cascaded convolutional layers
Two convolutional layers are cascaded after each convolutional layer of the VGG16 network, average operation is carried out on output data and input data after the operation of the convolutional layers, then the output data after the average operation is used as the input data of the cascaded convolutional layers, one convolutional layer operation is carried out again, and the like, three convolutional layers are cascaded in total.
3. The video smoke detection method based on the change accumulation graph and the cascaded deep network as claimed in claim 2, wherein the detection method is implemented by the following steps:
step1, adopting bilinear interpolation method to reduce the size of input k frame image to 224X 224;
step 2: if k is larger than N, the requirement of calculating the change accumulated image is met, and the next step is carried out; otherwise, buffering the current frame image, wherein k = k +1, and returning to Step 1;
step 3: calculating a transformation accumulated image of the k frame image according to formula (1) and formula (2);
step 4: extracting features by respectively adopting a VGG16feature extractor and a ResNet50feature extractor, and placing the 7 × 7 × 2048=100352 dimensional features extracted by the ResNet50 network after the 7 × 7 × 512=25088 dimensional features extracted by the VGG16 network to construct 100352+25088=125440 dimensional features;
step 5: each feature is regarded as a node of a neuron, extracted features are connected in a full-connection mode, and 1024 neuron nodes are output;
step 6: in order to prevent the overfitting phenomenon of the convolutional neural network, a Dropout method is adopted, and neural units are randomly selected and discarded according to the probability p, p = 0.3;
step 7: the rest nerve units after Dropout are still connected in a full connection mode, and 128 neuron nodes are output;
step 8: selecting and discarding the nerve units randomly according to the probability p, p =0.3 by using a Dropout method;
step 9: and outputting the rest nerve units through a Sigmoid activation function, judging the nerve units to be smoke images if the output value is greater than or equal to 0.5, otherwise judging the nerve units to be non-smoke images, caching the transformed accumulated images of the k frame images and the current frame images regardless of whether alarms are output, wherein k = k +1, and returning to Step 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911078837.7A CN110765982B (en) | 2019-11-07 | 2019-11-07 | Video smoke detection method based on change accumulation graph and cascaded depth network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911078837.7A CN110765982B (en) | 2019-11-07 | 2019-11-07 | Video smoke detection method based on change accumulation graph and cascaded depth network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110765982A true CN110765982A (en) | 2020-02-07 |
| CN110765982B CN110765982B (en) | 2022-06-14 |
Family
ID=69336552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911078837.7A Active CN110765982B (en) | 2019-11-07 | 2019-11-07 | Video smoke detection method based on change accumulation graph and cascaded depth network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110765982B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112286190A (en) * | 2020-10-26 | 2021-01-29 | 中国人民解放军国防科技大学 | Security patrol early warning method and system |
| CN113724151A (en) * | 2021-07-30 | 2021-11-30 | 荣耀终端有限公司 | Image enhancement method, electronic equipment and computer readable storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105726018A (en) * | 2016-02-06 | 2016-07-06 | 河北大学 | Automatic atrial fibrillation detection method irrelevant to RR interphase |
| CN106971160A (en) * | 2017-03-23 | 2017-07-21 | 西京学院 | Winter jujube disease recognition method based on depth convolutional neural networks and disease geo-radar image |
| CN107346421A (en) * | 2017-06-23 | 2017-11-14 | 南京理工大学 | A kind of video smoke detection method based on color invariance |
| CN107909566A (en) * | 2017-10-28 | 2018-04-13 | 杭州电子科技大学 | A kind of image-recognizing method of the cutaneum carcinoma melanoma based on deep learning |
| CN109948557A (en) * | 2019-03-22 | 2019-06-28 | 中国人民解放军国防科技大学 | Smoke detection method with multi-network model fusion |
| CN109961042A (en) * | 2019-03-22 | 2019-07-02 | 中国人民解放军国防科技大学 | Smoke detection method combining deep convolutional neural network and visual change diagram |
| CN110378252A (en) * | 2019-06-28 | 2019-10-25 | 浙江大学 | A kind of distress in concrete recognition methods based on depth migration study |
-
2019
- 2019-11-07 CN CN201911078837.7A patent/CN110765982B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105726018A (en) * | 2016-02-06 | 2016-07-06 | 河北大学 | Automatic atrial fibrillation detection method irrelevant to RR interphase |
| CN106971160A (en) * | 2017-03-23 | 2017-07-21 | 西京学院 | Winter jujube disease recognition method based on depth convolutional neural networks and disease geo-radar image |
| CN107346421A (en) * | 2017-06-23 | 2017-11-14 | 南京理工大学 | A kind of video smoke detection method based on color invariance |
| CN107909566A (en) * | 2017-10-28 | 2018-04-13 | 杭州电子科技大学 | A kind of image-recognizing method of the cutaneum carcinoma melanoma based on deep learning |
| CN109948557A (en) * | 2019-03-22 | 2019-06-28 | 中国人民解放军国防科技大学 | Smoke detection method with multi-network model fusion |
| CN109961042A (en) * | 2019-03-22 | 2019-07-02 | 中国人民解放军国防科技大学 | Smoke detection method combining deep convolutional neural network and visual change diagram |
| CN110378252A (en) * | 2019-06-28 | 2019-10-25 | 浙江大学 | A kind of distress in concrete recognition methods based on depth migration study |
Non-Patent Citations (4)
| Title |
|---|
| ANTON M ET AL: "real-time smoke detection in video sequences:combined approach", 《SPRING BERLIN HEIDELBERG》 * |
| 刘通等: "facial peculiarity retrieval via deep netural networks fusion", 《IEEE》 * |
| 杨甲甲等: "采用长短期记忆深度学习模型的工业负荷短期预测方法", 《电力建设》 * |
| 王静等: "基于深度学习的飞行器姿态分析研究", 《新技术新工艺》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112286190A (en) * | 2020-10-26 | 2021-01-29 | 中国人民解放军国防科技大学 | Security patrol early warning method and system |
| CN113724151A (en) * | 2021-07-30 | 2021-11-30 | 荣耀终端有限公司 | Image enhancement method, electronic equipment and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110765982B (en) | 2022-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107463949B (en) | Video action classification processing method and device | |
| Xie et al. | Efficient video fire detection exploiting motion-flicker-based dynamic features and deep static features | |
| CN110580472B (en) | Video foreground detection method based on full convolution network and conditional countermeasure network | |
| CN109948557B (en) | Smoke detection method with multi-network model fusion | |
| Dong et al. | Adaptive learning of multi-subspace for foreground detection under illumination changes | |
| CN110765982B (en) | Video smoke detection method based on change accumulation graph and cascaded depth network | |
| CN111582095A (en) | Light-weight rapid detection method for abnormal behaviors of pedestrians | |
| CN114882430B (en) | Transformer-based lightweight early fire detection method | |
| CN109859236A (en) | Mobile object detection method, calculates equipment and storage medium at system | |
| CN114885119A (en) | Intelligent monitoring alarm system and method based on computer vision | |
| Zhang et al. | DeepFireNet: A real-time video fire detection method based on multi-feature fusion | |
| CN114782859A (en) | Method for establishing space-time perception positioning model of target behaviors and application | |
| CN115131229A (en) | Image noise reduction and filtering data processing method and device and computer equipment | |
| Hou et al. | A survey of efficient deep learning models for moving object segmentation | |
| Tsetsos et al. | A forest fire detection system: The Meleager approach | |
| Geng et al. | Real time foreground-background segmentation using two-layer codebook model | |
| Deldjoo et al. | A novel fuzzy-based smoke detection system using dynamic and static smoke features | |
| CN106355566A (en) | Smoke and flame detection method applied to fixed camera dynamic video sequence | |
| CN116468980A (en) | Infrared small target detection method and device for deep fusion of edge details and deep features | |
| Wang et al. | Video Smoke Detection Based on Multi-feature Fusion and Modified Random Forest. | |
| CN114926348A (en) | Device and method for removing low-illumination video noise | |
| Özbek et al. | Smoke detection from foggy environment based on color spaces | |
| Zhu et al. | Improved adaptive mixture of Gaussians model for moving objects detection | |
| Ma et al. | A Moving Object Detection Method Based on 3D Convolution Neural Network | |
| Huang et al. | Real-time Video Fire Detection via Convolutional Neural Networks |
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 |