CN111476314A - Fuzzy video detection method integrating optical flow algorithm and deep learning - Google Patents
Fuzzy video detection method integrating optical flow algorithm and deep learning Download PDFInfo
- Publication number
- CN111476314A CN111476314A CN202010342615.8A CN202010342615A CN111476314A CN 111476314 A CN111476314 A CN 111476314A CN 202010342615 A CN202010342615 A CN 202010342615A CN 111476314 A CN111476314 A CN 111476314A
- Authority
- CN
- China
- Prior art keywords
- video
- frame
- frames
- detection
- optical flow
- 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
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a fuzzy video detection method integrating an optical flow algorithm and deep learning, which comprises the following steps: preprocessing a training video sample; obtaining a fuzzy video detection model, constructing a fuzzy video time sequence training model, and obtaining a characteristic diagram of a video frame through a deep learning algorithm; aggregating twenty feature graphs of the first ten frames and the last ten frames of the frames to be detected to one feature graph by a light stream algorithm according to the weight which is from 0 to 1 and conforms to normal distribution; determining the weight according to a normal distribution algorithm; detecting a frame feature map, and detecting the feature map; and locating and marking the specific position of the target in the video frame. The method not only considers the characteristics of the video frames, but also considers the video time sequence, and the related factors such as space, geographic position, weather and the like, and carries out optical flow fusion on each frame and the frames before and after the frame by using an optical flow method; the fuzzy video detection and identification capability under the complex application condition is improved, and the detection rate of the target in the fuzzy video is improved.
Description
Technical Field
The invention relates to the technical field of video identification, in particular to a fuzzy video detection method integrating an optical flow algorithm and deep learning.
Background
How to improve the detection rate of the blurred video is a difficult problem, and because the factors of video defocusing, partial shielding and motion blurring exist in the blurred video, even if the blurred video is a high-definition video, the intercepted frame is not as clear as a photo shot by a camera; in video monitoring, due to dim light at night, too long shooting distance and other reasons, the shot video is often blurred.
At present, detection work of fuzzy videos including monitored videos is mainly completed by professional personnel, but under the condition that the detection background is complex, the personnel are influenced by factors such as knowledge level and the like, and the accuracy of the fuzzy videos is difficult to guarantee by means of naked eyes. Meanwhile, the monitoring video in the natural environment is greatly influenced by weather conditions, such as heavy fog, wind and snow, heavy rain and the like, and is influenced by doped illumination, shadow and the like, so that the traditional fuzzy video detection method based on deep learning has low efficiency and unsatisfactory robustness. In addition, most of the existing detection methods focus on the feature extraction of video frames, and the consideration of relevant condition factors such as time sequence information of videos is neglected, so that the automatic identification of the fuzzy video can only exist in an experimental stage. How to improve the accuracy of the detection of the blurred video becomes a technical problem which needs to be solved urgently.
Disclosure of Invention
The invention aims to provide a fuzzy video detection method integrating an optical flow algorithm and deep learning, which can improve the detection and recognition capability of a fuzzy video under a complex application condition and improve the detection rate of a target in the fuzzy video.
In order to achieve the purpose, the invention adopts the following technical scheme: a fuzzy video detection method integrating an optical flow algorithm and deep learning comprises the following steps in sequence:
(1) preprocessing a training video sample: collecting a plurality of fuzzy videos and time and geographical position information of the corresponding targets as training data, manually marking detection targets in the fuzzy videos, and capturing frames of all marked videos to obtain a plurality of types of targets, wherein each type of target is provided with a plurality of training samples;
(2) training the frames obtained in the step (1) to obtain a fuzzy video detection model based on a deep learning algorithm, constructing a fuzzy video time sequence training model, introducing different fuzzy video shooting time and geographic positions as feature data, and training a rice disease detection model based on a deep learning fusion optical flow algorithm; obtaining a feature map of the video frame through a deep learning algorithm;
(3) aggregating twenty feature graphs of the first ten frames and the last ten frames of the frames to be detected to one feature graph by a light stream algorithm according to the weight which is from 0 to 1 and conforms to normal distribution;
(4) determining the weight in the step (3) according to a normal distribution algorithm;
(5) constructing a frame feature map detection model based on an image detection algorithm of deep learning, detecting a frame feature map generated by combining calculation of an optical flow algorithm, and detecting the feature map;
(6) and (3) combining the mark of the specific position of the target in the fuzzy video, inputting the space, geographic position and time information of the video to be detected into the trained frame characteristic diagram detection model, identifying and detecting the fuzzy video, positioning and marking the specific position of the target in the video frame by the computer.
In step (1), the pre-processing of the training video sample comprises the following steps:
(1a) collecting a plurality of videos which are fuzzily shot due to rain, snow, fog and night, and classifying the videos according to the time and geographical position information when the corresponding target occurs;
(1b) marking a video by using a video marking tool, wherein the video marking tool marks the video frame by frame, and the marked content is the category of an object in the video;
(1c) and (2) intercepting frames of the videos classified according to the time and the geographic position information when the corresponding target occurs in the step (1a) by using an algorithm, and storing the frames according to the time and the geographic position when the corresponding target occurs in a classified manner for training a detection model.
In the step (2), the training is performed by using the obtained frames based on the deep learning algorithm to obtain the fuzzy video detection model, so as to obtain the frame feature map, and the method specifically comprises the following steps:
(2a) training networks for obtaining the frame feature maps respectively by adopting ResNet-50, ResNet-101 and Goog L eNet for diversified frame formation and detection in the subsequent steps;
(2b) the network structure of ResNet-50 is that 49 convolution layers, 1 average pooling layer, wherein the convolution layer is divided into 16 blocks, each block has 1 shortcut connection, and finally, a softmax layer is used for generating a classification prediction confidence coefficient;
(2c) when the frame passes through the three types of networks respectively, the feature map of the frame is output before the last softmax layer.
The step (3) specifically comprises the following steps:
(3a) providing different information of the target object example according to the frame feature map;
(3b) by using an optical flow algorithm, the feature map of a specific frame and the feature maps of five frames before and after the specific frame are fused together, wherein the formula is as follows:
wherein f isiIs a feature map of a specific frame, ∑ denotes an optical flow aggregate, wiRepresenting the aggregation of adjacent characteristic maps by different weights, fjA characteristic diagram after polymerization;
wherein wiIs determined by the following formula:
and z is the distance of a particular frame from an adjacent frame, defining: z is | i-j |, μ is the mean of the normal distribution, σ is the variance of the normal distribution, and μ is taken to be 0, and σ is taken to be 1;
the optical flow algorithm adopts an optical flow algorithm in a computer vision library, and specifically comprises the following steps:
let pixel I (x, y, t), x, y denote coordinates, t denotes time, moved by a distance (dx, dy) to the next frame, with dt times, assuming this pixel is unchanged for a small time, i.e.:
I(x,y,t)=I(x+dx,y+dy,t+dt)
expanding the above formula to obtain:
wherein represents a second order infinite element, comparing the above two equations, we can obtain:
by removing dt from the above formula, we can obtain:
The step (4) comprises the following steps:
(4a) numbering eleven adjacent frames of the frame to be detected;
(4b) according to the weight calculation formulaAnd calculating the weight of each frame of eleven frames in total, wherein the value range of the weight is between 0 and 1.
The image detection algorithm based on deep learning in the step (5) is used for constructing a frame feature map detection model and detecting a frame feature map, and the method comprises the following steps:
(5a) using an R-FCN network as a network for detecting the frame characteristic diagram, wherein the R-FCN network comprises an RPN network and an R-FCN network;
(5b) the RPN network uses 9 anchor boxes, each graph generates 300 suggestion boxes, the position sensitive graph in the R-FCN network is 7 × 7 pixels;
(5c) training the R-FCN network by using a training sample frame to obtain a frame detection model related to the target category;
(5d) and (5) inputting the aggregated feature map into the frame detection model in the step (5c) to obtain a detection result.
The step (6) specifically comprises the following steps:
(6a) respectively training the training sample frames classified to obtain detection models of the fuzzy frames of the respective classes of the intercepted fuzzy video;
(6b) inputting the video frames of the respective types into respective detection models, and detecting to obtain detection results of the video frames of the respective types containing spatial, geographic position and time information;
(6c) the detection result is marked in the output result;
(6d) and counting the detection results of the training sample frames of each category, wherein the detection results comprise time and geographical position information when the target occurs.
According to the technical scheme, the beneficial effects of the invention are as follows: compared with the prior art, the method not only considers the characteristics of the video frames, but also considers the video time sequence, and the related factors such as space, geographic position, weather and the like, and carries out optical flow fusion on each frame and the frames before and after the frame by using an optical flow method; the method and the device respectively establish models for factors such as different geographic positions, weather and the like, and respectively detect the factors, thereby improving the detection and identification capabilities of the fuzzy video under the complex application condition and improving the detection rate of the target in the fuzzy video.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a diagram illustrating feature maps of an optical flow aggregation video frame according to the present invention.
Detailed Description
As shown in fig. 1, a blurred video detection method combining an optical flow algorithm and deep learning includes the following steps:
(1) preprocessing a training video sample: collecting a plurality of fuzzy videos and time and geographical position information of the corresponding targets as training data, manually marking detection targets in the fuzzy videos, and capturing frames of all marked videos to obtain a plurality of types of targets, wherein each type of target is provided with a plurality of training samples;
(2) training the frames obtained in the step (1) to obtain a fuzzy video detection model based on a deep learning algorithm, constructing a fuzzy video time sequence training model, introducing different fuzzy video shooting time and geographic positions as feature data, and training a rice disease detection model based on a deep learning fusion optical flow algorithm; obtaining a feature map of the video frame through a deep learning algorithm;
(3) aggregating twenty feature graphs of the first ten frames and the last ten frames of the frames to be detected to one feature graph by a light stream algorithm according to the weight which is from 0 to 1 and conforms to normal distribution;
(4) determining the weight in the step (3) according to a normal distribution algorithm;
(5) constructing a frame feature map detection model based on an image detection algorithm of deep learning, detecting a frame feature map generated by combining calculation of an optical flow algorithm, and detecting the feature map;
(6) and (3) combining the mark of the specific position of the target in the fuzzy video, inputting the space, geographic position and time information of the video to be detected into the trained frame characteristic diagram detection model, identifying and detecting the fuzzy video, positioning and marking the specific position of the target in the video frame by the computer.
In step (1), the pre-processing of the training video sample comprises the following steps:
(1a) collecting a plurality of videos which are fuzzily shot due to rain, snow, fog and night, and classifying the videos according to the time and geographical position information when the corresponding target occurs;
(1b) marking a video by using a video marking tool, wherein the video marking tool marks the video frame by frame, and the marked content is the category of an object in the video;
(1c) and (2) intercepting frames of the videos classified according to the time and the geographic position information when the corresponding target occurs in the step (1a) by using an algorithm, and storing the frames according to the time and the geographic position when the corresponding target occurs in a classified manner for training a detection model. The frames in step (1c) should have several classes, and there should be several training sample frames in each class.
Collecting a plurality of fuzzy videos and time, geographical position and weather information of shooting corresponding videos as training data, manually marking out targets in the fuzzy videos to obtain a plurality of types of videos, wherein each type of video has a plurality of video training samples. Here, not only a sample of the video but also information of time, geographical location, weather, etc. at the time of video capture is obtained, by which the robustness of the blurred video recognition is further increased.
In the step (2), the training is performed by using the obtained frames based on the deep learning algorithm to obtain the fuzzy video detection model, so as to obtain the frame feature map, and the method specifically comprises the following steps:
(2a) training networks for obtaining the frame feature maps respectively by adopting ResNet-50, ResNet-101 and Goog L eNet for diversified frame formation and detection in the subsequent steps;
(2b) the network structure of ResNet-50 is that 49 convolution layers, 1 average pooling layer, wherein the convolution layer is divided into 16 blocks, each block has 1 shortcut connection, and finally, a softmax layer is used for generating a classification prediction confidence coefficient;
(2c) when the frame passes through the three types of networks respectively, the feature map of the frame is output before the last softmax layer.
The step (3) specifically comprises the following steps:
(3a) providing different information of the target object example according to the frame feature map;
(3b) by using an optical flow algorithm, the feature map of a specific frame and the feature maps of five frames before and after the specific frame are fused together, wherein the formula is as follows:
wherein f isiIs a feature map of a specific frame, ∑ denotes an optical flow aggregate, wiRepresenting the aggregation of adjacent characteristic maps by different weights, fjA characteristic diagram after polymerization;
wherein wiIs determined by the following formula:
and z is the distance of a particular frame from an adjacent frame, defining: z is | i-j |, μ is the mean of the normal distribution, σ is the variance of the normal distribution, and should be adjusted according to different application ranges, and generally, μ is 0, and σ is 1;
the optical flow algorithm adopts an optical flow algorithm in a computer vision library, and specifically comprises the following steps:
let pixel I (x, y, t), x, y denote coordinates, t denotes time, moved by a distance (dx, dy) to the next frame, with dt times, assuming this pixel is unchanged for a small time, i.e.:
I(x,y,t)=I(x+dx,y+dy,t+dt)
expanding the above formula to obtain:
wherein represents a second order infinite element, comparing the above two equations, we can obtain:
by removing dt from the above formula, we can obtain:
The above equation indicates that the closer to the specific frame, the larger the weight value, and the farther from the specific frame, the smaller the weight value.
The step (4) comprises the following steps:
(4a) numbering eleven adjacent frames of the frame to be detected;
(4b) according to the weight calculation formulaAnd calculating the weight of each frame of eleven frames in total, wherein the value range of the weight is between 0 and 1.
The image detection algorithm based on deep learning in the step (5) is used for constructing a frame feature map detection model and detecting a frame feature map, and the method comprises the following steps:
(5a) using an R-FCN network as a network for detecting the frame characteristic diagram, wherein the R-FCN network comprises an RPN network and an R-FCN network;
(5b) the RPN network uses 9 anchor boxes, each graph generates 300 suggestion boxes, the position sensitive graph in the R-FCN network is 7 × 7 pixels;
(5c) training the R-FCN network by using a training sample frame to obtain a frame detection model related to the target category;
(5d) and (5) inputting the aggregated feature map into the frame detection model in the step (5c) to obtain a detection result.
The step (6) specifically comprises the following steps:
(6a) respectively training the training sample frames classified to obtain detection models of the fuzzy frames of the respective classes of the intercepted fuzzy video;
(6b) inputting the video frames of the respective types into respective detection models, and detecting to obtain detection results of the video frames of the respective types containing spatial, geographic position and time information;
(6c) the detection result has labels in the output result, such as the position of the target in the frame, the target type, the target confidence coefficient, etc.;
(6d) and counting the detection results of the training sample frames of each category, wherein the detection results comprise time and geographical position information when the target occurs.
In summary, compared with the prior art, the method not only considers the characteristics of the video frames, but also considers the video time sequence, and the related factors such as space, geographic position, weather and the like, and carries out optical flow fusion on each frame and the frames before and after the frame by using an optical flow method; the method and the device respectively establish models for factors such as different geographic positions, weather and the like, and respectively detect the factors, thereby improving the detection and identification capabilities of the fuzzy video under the complex application condition and improving the detection rate of the target in the fuzzy video.
Claims (7)
1. A fuzzy video detection method integrating an optical flow algorithm and deep learning is characterized in that: the method comprises the following steps in sequence:
(1) preprocessing a training video sample: collecting a plurality of fuzzy videos and time and geographical position information of the corresponding targets as training data, manually marking detection targets in the fuzzy videos, and capturing frames of all marked videos to obtain a plurality of types of targets, wherein each type of target is provided with a plurality of training samples;
(2) training the frames obtained in the step (1) to obtain a fuzzy video detection model based on a deep learning algorithm, constructing a fuzzy video time sequence training model, introducing different fuzzy video shooting time and geographic positions as feature data, and training a rice disease detection model based on a deep learning fusion optical flow algorithm; obtaining a feature map of the video frame through a deep learning algorithm;
(3) aggregating twenty feature graphs of the first ten frames and the last ten frames of the frames to be detected to one feature graph by a light stream algorithm according to the weight which is from 0 to 1 and conforms to normal distribution;
(4) determining the weight in the step (3) according to a normal distribution algorithm;
(5) constructing a frame feature map detection model based on an image detection algorithm of deep learning, detecting a frame feature map generated by combining calculation of an optical flow algorithm, and detecting the feature map;
(6) and (3) combining the mark of the specific position of the target in the fuzzy video, inputting the space, geographic position and time information of the video to be detected into the trained frame characteristic diagram detection model, identifying and detecting the fuzzy video, positioning and marking the specific position of the target in the video frame by the computer.
2. The method of detecting blurred video with integrated optical flow algorithm and deep learning according to claim 1, wherein: in step (1), the pre-processing of the training video sample comprises the following steps:
(1a) collecting a plurality of videos which are fuzzily shot due to rain, snow, fog and night, and classifying the videos according to the time and geographical position information when the corresponding target occurs;
(1b) marking a video by using a video marking tool, wherein the video marking tool marks the video frame by frame, and the marked content is the category of an object in the video;
(1c) and (2) intercepting frames of the videos classified according to the time and the geographic position information when the corresponding target occurs in the step (1a) by using an algorithm, and storing the frames according to the time and the geographic position when the corresponding target occurs in a classified manner for training a detection model.
3. The method of detecting blurred video with integrated optical flow algorithm and deep learning according to claim 1, wherein: in the step (2), the training is performed by using the obtained frames based on the deep learning algorithm to obtain the fuzzy video detection model, so as to obtain the frame feature map, and the method specifically comprises the following steps:
(2a) training networks for obtaining the frame feature maps respectively by adopting ResNet-50, ResNet-101 and Goog L eNet for diversified frame formation and detection in the subsequent steps;
(2b) the network structure of ResNet-50 is that 49 convolution layers, 1 average pooling layer, wherein the convolution layer is divided into 16 blocks, each block has 1 shortcut connection, and finally, a softmax layer is used for generating a classification prediction confidence coefficient;
(2c) when the frame passes through the three types of networks respectively, the feature map of the frame is output before the last softmax layer.
4. The method of detecting blurred video with integrated optical flow algorithm and deep learning according to claim 1, wherein: the step (3) specifically comprises the following steps:
(3a) providing different information of the target object example according to the frame feature map;
(3b) by using an optical flow algorithm, the feature map of a specific frame and the feature maps of five frames before and after the specific frame are fused together, wherein the formula is as follows:
wherein f isiIs a feature map of a specific frame, ∑ denotes an optical flow aggregate, wiRepresenting the aggregation of adjacent characteristic maps by different weights, fjA characteristic diagram after polymerization;
wherein wiIs determined by the following formula:
and z is the distance of a particular frame from an adjacent frame, defining: z is | i-j |, μ is the mean of the normal distribution, σ is the variance of the normal distribution, and μ is taken to be 0, and σ is taken to be 1;
the optical flow algorithm adopts an optical flow algorithm in a computer vision library, and specifically comprises the following steps:
let pixel I (x, y, t), x, y denote coordinates, t denotes time, moved by a distance (dx, dy) to the next frame, with dt times, assuming this pixel is unchanged for a small time, i.e.:
I(x,y,t)=I(x+dx,y+dy,t+dt)
expanding the above formula to obtain:
wherein represents a second order infinite element, comparing the above two equations, we can obtain:
by removing dt from the above formula, we can obtain:
5. The method of detecting blurred video with integrated optical flow algorithm and deep learning according to claim 1, wherein: the step (4) comprises the following steps:
(4a) numbering eleven adjacent frames of the frame to be detected;
6. The video detection method with fusion of optical flow algorithm and deep learning model according to claim 1, characterized in that: the image detection algorithm based on deep learning in the step (5) is used for constructing a frame feature map detection model and detecting a frame feature map, and the method comprises the following steps:
(5a) using an R-FCN network as a network for detecting the frame characteristic diagram, wherein the R-FCN network comprises an RPN network and an R-FCN network;
(5b) the RPN network uses 9 anchor boxes, each graph generates 300 suggestion boxes, the position sensitive graph in the R-FCN network is 7 × 7 pixels;
(5c) training the R-FCN network by using a training sample frame to obtain a frame detection model related to the target category;
(5d) and (5) inputting the aggregated feature map into the frame detection model in the step (5c) to obtain a detection result.
7. The video detection method with fusion of optical flow algorithm and deep learning model according to claim 1, characterized in that: the step (6) specifically comprises the following steps:
(6a) respectively training the training sample frames classified to obtain detection models of the fuzzy frames of the respective classes of the intercepted fuzzy video;
(6b) inputting the video frames of the respective types into respective detection models, and detecting to obtain detection results of the video frames of the respective types containing spatial, geographic position and time information;
(6c) the detection result is marked in the output result;
(6d) and counting the detection results of the training sample frames of each category, wherein the detection results comprise time and geographical position information when the target occurs.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010342615.8A CN111476314B (en) | 2020-04-27 | 2020-04-27 | Fuzzy video detection method integrating optical flow algorithm and deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010342615.8A CN111476314B (en) | 2020-04-27 | 2020-04-27 | Fuzzy video detection method integrating optical flow algorithm and deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111476314A true CN111476314A (en) | 2020-07-31 |
CN111476314B CN111476314B (en) | 2023-03-07 |
Family
ID=71762850
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010342615.8A Active CN111476314B (en) | 2020-04-27 | 2020-04-27 | Fuzzy video detection method integrating optical flow algorithm and deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111476314B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112053327A (en) * | 2020-08-18 | 2020-12-08 | 南京理工大学 | Video target detection method and system, storage medium and server |
CN113781353A (en) * | 2021-07-30 | 2021-12-10 | 杭州当虹科技股份有限公司 | Video rain and snow removing method based on explicit optical flow alignment fusion |
CN114419517A (en) * | 2022-01-27 | 2022-04-29 | 腾讯科技(深圳)有限公司 | Video frame processing method and device, computer equipment and storage medium |
CN115631478A (en) * | 2022-12-02 | 2023-01-20 | 广汽埃安新能源汽车股份有限公司 | Road image detection method, device, equipment and computer readable medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10318842B1 (en) * | 2018-09-05 | 2019-06-11 | StradVision, Inc. | Learning method, learning device for optimizing parameters of CNN by using multiple video frames and testing method, testing device using the same |
CN109993095A (en) * | 2019-03-26 | 2019-07-09 | 东北大学 | A kind of other characteristic aggregation method of frame level towards video object detection |
WO2019144575A1 (en) * | 2018-01-24 | 2019-08-01 | 中山大学 | Fast pedestrian detection method and device |
CN110427839A (en) * | 2018-12-26 | 2019-11-08 | 西安电子科技大学 | Video object detection method based on multilayer feature fusion |
CN110458756A (en) * | 2019-06-25 | 2019-11-15 | 中南大学 | Fuzzy video super-resolution method and system based on deep learning |
-
2020
- 2020-04-27 CN CN202010342615.8A patent/CN111476314B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019144575A1 (en) * | 2018-01-24 | 2019-08-01 | 中山大学 | Fast pedestrian detection method and device |
US10318842B1 (en) * | 2018-09-05 | 2019-06-11 | StradVision, Inc. | Learning method, learning device for optimizing parameters of CNN by using multiple video frames and testing method, testing device using the same |
CN110427839A (en) * | 2018-12-26 | 2019-11-08 | 西安电子科技大学 | Video object detection method based on multilayer feature fusion |
CN109993095A (en) * | 2019-03-26 | 2019-07-09 | 东北大学 | A kind of other characteristic aggregation method of frame level towards video object detection |
CN110458756A (en) * | 2019-06-25 | 2019-11-15 | 中南大学 | Fuzzy video super-resolution method and system based on deep learning |
Non-Patent Citations (2)
Title |
---|
李森等: "基于时空建模的视频帧预测模型", 《物联网技术》 * |
邓志新等: "基于时空双流全卷积网络的视频目标分割算法研究及改进", 《工业控制计算机》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112053327A (en) * | 2020-08-18 | 2020-12-08 | 南京理工大学 | Video target detection method and system, storage medium and server |
CN112053327B (en) * | 2020-08-18 | 2022-08-23 | 南京理工大学 | Video target detection method and system, storage medium and server |
CN113781353A (en) * | 2021-07-30 | 2021-12-10 | 杭州当虹科技股份有限公司 | Video rain and snow removing method based on explicit optical flow alignment fusion |
CN114419517A (en) * | 2022-01-27 | 2022-04-29 | 腾讯科技(深圳)有限公司 | Video frame processing method and device, computer equipment and storage medium |
CN115631478A (en) * | 2022-12-02 | 2023-01-20 | 广汽埃安新能源汽车股份有限公司 | Road image detection method, device, equipment and computer readable medium |
Also Published As
Publication number | Publication date |
---|---|
CN111476314B (en) | 2023-03-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111476314B (en) | Fuzzy video detection method integrating optical flow algorithm and deep learning | |
CN111310862B (en) | Image enhancement-based deep neural network license plate positioning method in complex environment | |
CN105745687B (en) | Context aware Moving target detection | |
CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
CN104700099A (en) | Method and device for recognizing traffic signs | |
CN108197604A (en) | Fast face positioning and tracing method based on embedded device | |
CN107622502A (en) | The path extraction of robot vision leading system and recognition methods under the conditions of complex illumination | |
CN108804992B (en) | Crowd counting method based on deep learning | |
CN109801297B (en) | Image panorama segmentation prediction optimization method based on convolution | |
CN103353941B (en) | Natural marker registration method based on viewpoint classification | |
CN112541403B (en) | Indoor personnel falling detection method by utilizing infrared camera | |
CN111476160A (en) | Loss function optimization method, model training method, target detection method, and medium | |
CN113435407B (en) | Small target identification method and device for power transmission system | |
CN113436229A (en) | Multi-target cross-camera pedestrian trajectory path generation method | |
CN111695373A (en) | Zebra crossing positioning method, system, medium and device | |
CN111160100A (en) | Lightweight depth model aerial photography vehicle detection method based on sample generation | |
CN111260687A (en) | Aerial video target tracking method based on semantic perception network and related filtering | |
CN112560623A (en) | Unmanned aerial vehicle-based rapid mangrove plant species identification method | |
CN113435452A (en) | Electrical equipment nameplate text detection method based on improved CTPN algorithm | |
CN115841633A (en) | Power tower and power line associated correction power tower and power line detection method | |
CN115619719A (en) | Pine wood nematode infected wood detection method based on improved Yolo v3 network model | |
CN115100497A (en) | Robot-based method, device, equipment and medium for routing inspection of abnormal objects in channel | |
CN107992799A (en) | Towards the preprocess method of Smoke Detection application | |
CN112418262A (en) | Vehicle re-identification method, client and system | |
CN110738229B (en) | Fine-grained image classification method and device and electronic equipment |
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 |