CN117496224A - Camera disturbance recognition method and device based on integrated algorithm - Google Patents

Camera disturbance recognition method and device based on integrated algorithm Download PDF

Info

Publication number
CN117496224A
CN117496224A CN202311341923.9A CN202311341923A CN117496224A CN 117496224 A CN117496224 A CN 117496224A CN 202311341923 A CN202311341923 A CN 202311341923A CN 117496224 A CN117496224 A CN 117496224A
Authority
CN
China
Prior art keywords
disturbance
camera
identifier
level feature
level
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311341923.9A
Other languages
Chinese (zh)
Inventor
闫军
项炎平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Smart Intercommunication Technology Co ltd
Original Assignee
Smart Intercommunication Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Smart Intercommunication Technology Co ltd filed Critical Smart Intercommunication Technology Co ltd
Priority to CN202311341923.9A priority Critical patent/CN117496224A/en
Publication of CN117496224A publication Critical patent/CN117496224A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/54—Extraction of image or video features relating to texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • 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
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Biophysics (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a camera disturbance recognition method and device based on an integrated algorithm, which relate to the technical field of disturbance recognition, and the method comprises the following steps: collecting an image data set containing camera disturbance, and recording camera motion parameters of the image data set; acquiring a standard image set; carrying out multi-level feature extraction on the standard image set to obtain a plurality of feature sets; sequentially training and acquiring a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier; fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier; acquiring an image acquisition result; according to the camera disturbance recognition result of the target camera obtained according to the output information, the problem that in the prior art, the recognition effect is inaccurate due to insufficient rigor and insufficient completeness of camera disturbance recognition work is solved, and the accuracy of camera disturbance recognition is improved.

Description

Camera disturbance recognition method and device based on integrated algorithm
Technical Field
The invention relates to the technical field of disturbance recognition, in particular to a camera disturbance recognition method and device based on an integrated algorithm.
Background
Camera perturbation recognition refers to the recognition of distortion, blurring, or deformation caused by camera perturbation by analyzing image data. Such disturbances may result from various factors such as camera movements, changes in light conditions, and changes in the pose of the target object. These disturbances may lead to degradation of image quality and even impact subsequent image processing and machine learning tasks. Therefore, how to effectively identify and process the camera disturbance is one of the problems to be solved currently.
The problem that the camera disturbance recognition work in the prior art is inaccurate in recognition effect due to insufficient rigor and insufficient completeness, so that the recognition accuracy cannot be improved finally about camera disturbance recognition.
Disclosure of Invention
The application provides a camera disturbance recognition method and device based on an integrated algorithm, which solve the problem that in the prior art, the recognition effect is inaccurate due to insufficient rigor and insufficient completeness of camera disturbance recognition work, and realize improvement on the accuracy of camera disturbance recognition.
In view of the above, the present application provides a camera disturbance recognition method based on an integrated algorithm.
In a first aspect, the present application provides a camera disturbance recognition method based on an integration algorithm, the method comprising: collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, and marking the disturbance level; carrying out standardization processing on the marked image data set to obtain a standard image set; carrying out multi-level feature extraction on the standard image set to obtain a plurality of feature sets, wherein the feature sets comprise a low-level feature set, a middle-level feature subset and a high-level feature set; based on the low-level feature set, the middle-level feature set and the high-level feature set, training in sequence to obtain a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier; fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier; image acquisition is carried out on the random targets by using a target camera, and an image acquisition result is obtained, wherein the image acquisition result comprises camera disturbance; and inputting the image acquisition result into the camera disturbance intelligent identifier, and acquiring a camera disturbance identification result of the target camera according to output information.
In a second aspect, the present application provides an integrated algorithm-based camera disturbance recognition apparatus, the apparatus comprising: and a data acquisition module: collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, and marking the disturbance level; standardized processing module: carrying out standardization processing on the marked image data set to obtain a standard image set; and the feature extraction module is used for: carrying out multi-level feature extraction on the standard image set to obtain a plurality of feature sets, wherein the feature sets comprise a low-level feature set, a middle-level feature subset and a high-level feature set; the recognizer training module: based on the low-level feature set, the middle-level feature set and the high-level feature set, training in sequence to obtain a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier; the trainer fusion module: fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier; and an image acquisition module: image acquisition is carried out on the random targets by using a target camera, and an image acquisition result is obtained, wherein the image acquisition result comprises camera disturbance; a disturbance identification module: and inputting the image acquisition result into the camera disturbance intelligent identifier, and acquiring a camera disturbance identification result of the target camera according to output information.
One or more technical solutions provided in the present application have at least the following technical effects or advantages:
according to the camera disturbance recognition method and device based on the integrated algorithm, through collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, marking disturbance level for standardization processing, obtaining a standard image set, carrying out multi-level feature extraction on the standard image set, obtaining a plurality of feature sets including a low-level feature set, a middle-level feature subset and a high-level feature set, training in sequence to obtain a support vector machine recognizer, a decision tree recognizer and a convolutional neural network recognizer, fusing to obtain a camera disturbance intelligent recognizer, then carrying out image collection on a random target by using a target camera, obtaining an image collection result, finally inputting the image collection result into the camera disturbance intelligent recognizer, obtaining a camera disturbance recognition result of the target camera according to output information, solving the problem that in the prior art, the camera disturbance recognition work is inaccurate in recognition effect due to insufficient rigor and complete, and improving the camera disturbance recognition accuracy is achieved.
Drawings
FIG. 1 is a schematic flow chart of a camera disturbance recognition method based on an integration algorithm;
fig. 2 is a schematic structural diagram of a camera disturbance recognition device based on an integrated algorithm.
Reference numerals illustrate: the device comprises a data acquisition module 11, a standardization processing module 12, a feature extraction module 13, a recognizer training module 14, a trainer fusion module 15, an image acquisition module 16 and a disturbance recognition module 17.
Detailed Description
According to the camera disturbance recognition method and device based on the integrated algorithm, through collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, marking disturbance level to perform standardization processing, obtaining a standard image set, performing multi-level feature extraction on the standard image set, obtaining a plurality of feature sets, including a low-level feature set, a middle-level feature set and a high-level feature set, sequentially training and obtaining a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier, fusing to obtain a camera disturbance intelligent identifier, then using a target camera to perform image collection on a random target, obtaining an image collection result, inputting the image collection result into the camera disturbance intelligent identifier, and obtaining a camera disturbance recognition result of the target camera according to output information. The problem that in the prior art, the camera disturbance recognition work is not strict enough and the completeness is insufficient, so that the recognition effect is inaccurate is solved, and the camera disturbance recognition accuracy is improved.
Example 1
As shown in fig. 1, the present application provides a camera disturbance identification method and device based on an integration algorithm, wherein the method includes:
collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, and marking the disturbance level;
camera perturbation refers to image distortion, blurring, or distortion due to factors such as differences in the performance of the camera, changes in the photographic environment, and operational errors. These factors need to be taken into account when acquiring the image dataset and their impact on the acquisition result is minimized. The method comprises the steps of acquiring image data with camera disturbance, acquiring an image data set containing the camera disturbance, acquiring camera motion parameters of the image data set, wherein the camera motion parameters are attached to the image data, can be directly acquired, and comprise displacement, rotation angle, focal length and the like of a camera.
Carrying out standardization processing on the marked image data set to obtain a standard image set;
before the standardized processing is performed on the marked image data set, the image data needs to be preprocessed first. The preprocessing comprises the operations of denoising, enhancing and the like on the image, can improve the image quality, provides a better data base for the subsequent labeling and standardization processing, and processes the labeled image data set through a denoising algorithm and an image enhancing algorithm to obtain a preprocessed image data set. And then the preprocessed image data set is transformed and standardized, the data transformation comprises normalization processing of the size, the shape and the like of the image, so that the size difference and the deformation between different images are eliminated, and the contrast and the readability of the image are improved. The data normalization is to normalize the brightness, color and other features of the image data, so that the image data has the same feature range, and the subsequent analysis and processing are convenient. The data conversion is carried out on the preprocessed image data set to obtain a size normalized image data set, and the data normalization is carried out on the size normalized image data set to obtain a standard image set, so that the readability and the accuracy of the image data can be effectively improved, and a foundation is provided for subsequent image analysis, processing and utilization.
Carrying out multi-level feature extraction on the standard image set to obtain a plurality of feature sets, wherein the feature sets comprise a low-level feature set, a middle-level feature subset and a high-level feature set;
a convolutional neural network is adopted as a feature extractor;
extracting low-level features from the first convolution layer to obtain a low-level feature set, wherein the low-level feature set comprises edge features, texture features and color features;
extracting middle-level features in a second convolution layer to obtain a middle-level feature set, wherein the middle-level feature set comprises corner features, contour features and texture combination features;
and extracting high-level features from the third convolution layer to obtain a high-level feature set, wherein the high-level feature set comprises object shape features and overall structure features.
The multi-level feature extraction refers to feature extraction of a plurality of different standards on the same feature extraction object to obtain feature extraction results of a plurality of different types, and corresponding analysis is performed on features of different types, so that feature extraction and feature analysis can be performed on an extracted object more comprehensively from a plurality of angles. The convolutional neural network is used as a feature extractor, is a feed-forward neural network comprising a depth structure, can carry out translation invariant classification on input information according to a hierarchical structure, and is suitable for constructing a hierarchical feature extractor. The method comprises the steps of setting a feature extractor to be of a three-layer structure, extracting low-level features in a first convolution layer, and obtaining a low-level feature set, wherein the low-level feature set comprises edge features, texture features and color features of each part in an image; extracting middle-level features in the second convolution layer to obtain a middle-level feature set, wherein the middle-level feature set comprises corner features, contour features and texture combination features, and the texture combination features are features obtained after fusion of multiple textures; and extracting high-level features in the third convolution layer, and obtaining a high-level feature set, wherein the high-level feature set comprises object shape features and overall structure features, so as to complete the construction of a feature extractor. The standard image set is subjected to feature extraction through the multi-level feature extractor, so that the extracted target can be subjected to feature extraction and feature analysis more comprehensively from multiple angles, and the accuracy of the disturbance recognition of the follow-up camera is further improved.
Based on the low-level feature set, the middle-level feature set and the high-level feature set, training in sequence to obtain a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier;
acquiring a first identification model, wherein the first identification model is a support vector machine;
dividing the standard image set into a training set and a testing set, and taking the disturbance level as a category label;
and training the support vector machine by adopting the low-level feature set and the corresponding class label in the training set, evaluating the trained support vector machine by adopting the test set, and optimizing the support vector machine according to the evaluation result until the preset requirement is met, so as to obtain the support vector machine identifier.
And performing corresponding recognizer training by using the low-level feature set, the medium-level feature set and the high-level feature set, and performing feature recognition training by a trainer according to the feature set to obtain a recognizer of the corresponding feature, namely a support vector machine recognizer, a decision tree recognizer and a convolutional neural network recognizer. The support vector machine identifier is an identifier for identifying a low-level feature set, and is mainly constructed through a support vector machine, wherein the support vector machine is a machine learning algorithm and is used for tasks such as classification, regression, anomaly detection and the like, and samples of different types are separated by mainly finding an optimal hyperplane in a high-dimensional space, wherein the optimal hyperplane is determined according to a vector space formed by training samples. Dividing a standard image set into a training set and a test set by taking a support vector machine as a first recognition model, taking a disturbance level as a class label, training the support vector machine by adopting a low-level feature set in the training set and a corresponding class label, evaluating the trained support vector machine by adopting the test set, and optimizing the support vector machine according to an evaluation result until a preset requirement is met to obtain a support vector machine identifier;
the decision tree identifier is an identifier for identifying a middle-level feature set, is mainly constructed through a decision tree algorithm, takes a decision tree as a second identification model, takes a training set of a standard image set as input root node data of the decision tree, takes a test set of the standard image set as a plurality of output leaf node supervision data of the decision tree, trains a first decision tree, and obtains leaf node output deviation of the first decision tree, wherein the leaf node output deviation is the difference between a leaf node output value and the leaf node supervision data. When the absolute value of the leaf node output deviation of the first leaf node is larger than or equal to a preset deviation value, the first leaf node is taken as a secondary root node, a second decision tree is trained, wherein the input of the second decision tree is the leaf node output deviation and the leaf node output value of the first leaf node, and the supervision data of the second decision tree is the same as the supervision data of the first leaf node; and (3) performing iterative analysis until the absolute value of the leaf node output deviation of all the leaf nodes is smaller than a preset deviation value, updating training data, and training a second recognition model to obtain the final decision tree recognizer.
The convolutional neural network identifier is an identifier for identifying a high-level feature set, is mainly constructed through a convolutional neural network, and takes the convolutional neural network as a third identification model, the high-level feature set can well describe the features of an input image, and the characteristics are extracted through algorithms such as HOG, SIFT, SURF, so that training of the third identification model is facilitated. The high-level feature set is taken as input, and a convolutional neural network model is constructed by using algorithms including LeNet, alexNet, VGG and the like for training. Evaluating the effect of the identifier: after training is completed, the effect of the recognizer is required to be evaluated, indexes including accuracy, recall, F1 value and the like are selected to judge whether the recognizer achieves the required effect, and if not, training is continued; and outputting a third recognition model when the index reaches a preset index threshold value to obtain a high-level feature set. By constructing corresponding identifiers for various features, the accuracy and efficiency of feature identification can be improved.
Fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier;
acquiring a verification data set, wherein the verification data set comprises a verification standard image and a corresponding verification disturbance level;
inputting the verification standard image into the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier respectively to obtain a first identification result, a second identification result and a third identification result;
respectively carrying out similarity comparison on the first recognition result, the second recognition result and the third recognition result with the verification disturbance level, and acquiring a first weight, a second weight and a third weight according to the similarity comparison result;
and carrying out weighted fusion on the identifier according to the first weight, the second weight and the third weight to obtain the intelligent camera disturbance identifier.
And fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain the integrated intelligent identifier. In different images, the picture components are different, so that the features required to be identified are different, and the weight distribution is required to be carried out on the three identifiers so as to enable the image identification to be more in line with the actual situation. And acquiring a verification data set, wherein the verification data set comprises a verification standard image and a corresponding verification disturbance level, and the verification standard image and the verification disturbance level are comparison standard data proved by verification. And respectively inputting the verification standard image into a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier to obtain a first identification result, a second identification result and a third identification result. And respectively carrying out similarity comparison on the first recognition result, the second recognition result and the third recognition result with the verification disturbance level to obtain a similarity comparison result, and acquiring a first weight, a second weight and a third weight according to the similarity comparison result, wherein the better the similarity comparison result is, the higher the weight is, so that the image is more suitable for using the recognition result. And carrying out weighted fusion on the identifier according to the first weight, the second weight and the third weight to obtain the intelligent camera disturbance identifier. The three recognizers are weighted and fused to obtain the intelligent camera disturbance recognizers, so that the advantages of the recognizers can be integrated, and the recognition result is more accurate.
Image acquisition is carried out on the random targets by using a target camera, and an image acquisition result is obtained, wherein the image acquisition result comprises camera disturbance;
and inputting the image acquisition result into the camera disturbance intelligent identifier, and acquiring a camera disturbance identification result of the target camera according to output information.
The random targets are subjected to image acquisition by using the target camera, for example, the random targets are used as a vehicle data recorder in an automobile to acquire an image acquisition result, and disturbance is generated in the image acquisition process due to factors such as engine shake and uneven road surface of the automobile in the running process, namely, the image acquisition result comprises camera disturbance. The camera disturbance intelligent identifier carries out image recognition of the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier on the image acquisition result, generates a corresponding camera disturbance intelligent identifier according to the recognition result, inputs the image acquisition result into the camera disturbance intelligent identifier, and acquires a camera disturbance recognition result of the target camera according to the feature extraction result and the output information, so that the acquired recognition result can be more accurate.
Further, the method further comprises:
performing difference value calculation on the first weight, the second weight and the third weight, and judging whether the difference value calculation is smaller than a preset threshold value or not;
when the difference value of any group of weights is larger than a preset threshold value, obtaining a weight maximum value;
and taking the identifier corresponding to the maximum weight value as an intelligent camera disturbance identifier.
When abnormality or recognition failure occurs in the first weight, the second weight and the third weight, the similarity is lower, so that the weight value is lower, and if the lower weight value is fused into the camera disturbance intelligent recognizer, the recognition accuracy is reduced, so that the lower weight value is required to be removed. And carrying out two-by-two difference calculation on the first weight, the second weight and the third weight to obtain three difference calculation results, setting a preset threshold value, and judging whether the difference calculation is smaller than the preset threshold value. When the difference value of any group of weights is smaller than a preset threshold value, the fact that one weight value has larger deviation is indicated, the maximum value of the weights is directly obtained, the identifier corresponding to the maximum value of the weights is used as the intelligent camera disturbance identifier, and the influence of the problematic weight value on the intelligent camera disturbance identifier is avoided.
Further, the method further comprises:
obtaining disturbance category and disturbance degree according to a camera disturbance identification result;
acquiring sensor data, wherein the sensor data comprises acceleration data and angular velocity data;
calculating and obtaining anti-shake parameters by combining the sensor data by using disturbance types and disturbance degrees;
and controlling the camera according to the anti-shake parameters.
According to the camera disturbance recognition result, the disturbance type and the disturbance degree corresponding to the camera disturbance recognition result are obtained based on the big data analysis, sensor data are obtained, the sensors are an acceleration sensor and an angular velocity sensor, the sensors are arranged at the same position of the camera, all sensor data of the camera are all data of the sensors, and the sensor data comprise acceleration data and angular velocity data. And (3) calculating anti-shake parameters according to an anti-shake parameter calculation formula by combining disturbance type and disturbance degree with sensor data, wherein the anti-shake parameters are displacement posture parameters opposite to shake, and performing displacement posture compensation according to the anti-shake parameters to finish stable control of the camera and reduce the conditions of unclear images and the like caused by disturbance of the camera.
Further, the method further comprises:
acquiring environmental monitoring data;
performing environmental impact assessment of the target camera based on the environmental monitoring data to obtain an environmental impact coefficient;
and when the environmental impact coefficient reaches a preset environmental impact threshold value, compensating the camera disturbance recognition result based on the environmental impact coefficient.
For some camera devices installed outdoors to perform work, it is necessary to consider the situation of the environment for camera disturbance, such as shaking of the camera when the wind power level is too high, or corresponding shaking of the large truck when the large truck passes through the road or the like to perform work, so that the environment situation needs to be considered. And acquiring environment monitoring data, and evaluating the environmental influence of the target camera based on the environment monitoring data, namely, the influence degree of the environment monitoring data on the target camera, so as to obtain an environmental influence coefficient. Setting a preset environmental impact threshold, when the environmental impact coefficient reaches the preset environmental impact threshold, indicating that the environmental factor can affect the target camera, processing the environmental factor, analyzing the jitter condition caused by the environmental impact coefficient, compensating the camera disturbance recognition result according to the analysis result, and reducing the influence of the environmental factor on the camera disturbance recognition result.
Example two
Based on the same inventive concept as the camera disturbance recognition method based on the integration algorithm in the foregoing embodiment, as shown in fig. 2, the present application provides a camera disturbance recognition device based on the integration algorithm, which includes:
the data acquisition module 11: the data acquisition module 11 is used for acquiring an image data set containing camera disturbance, recording camera motion parameters of the image data set and marking the disturbance level;
standardized processing module 12: the normalization processing module 12 is configured to perform normalization processing on the marked image data set, and obtain a standard image set;
the feature extraction module 13: the feature extraction module 13 is configured to perform multi-level feature extraction on the standard image set to obtain a plurality of feature sets, where the feature sets include a low-level feature set, a middle-level feature subset, and a high-level feature set;
the recognizer training module 14: the identifier training module 14 is configured to sequentially train and acquire a support vector machine identifier, a decision tree identifier, and a convolutional neural network identifier based on the low-level feature set, the medium-level feature set, and the high-level feature set;
trainer fusion module 15: the trainer fusion module 15 is configured to fuse the support vector machine identifier, the decision tree identifier, and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier;
image acquisition module 16: the image acquisition module 16 is configured to acquire an image acquisition result from a random target by using a target camera, where the image acquisition result includes camera disturbance;
disturbance recognition module 17: the disturbance recognition module 17 is configured to input the image acquisition result into the camera disturbance intelligent recognizer, and obtain a camera disturbance recognition result of the target camera according to output information.
Further, the trainer fusion module 15 includes the following steps:
further, the device further comprises:
further, the feature extraction module 13 includes the following execution steps:
a convolutional neural network is adopted as a feature extractor;
extracting low-level features from the first convolution layer to obtain a low-level feature set, wherein the low-level feature set comprises edge features, texture features and color features;
extracting middle-level features in a second convolution layer to obtain a middle-level feature set, wherein the middle-level feature set comprises corner features, contour features and texture combination features;
and extracting high-level features from the third convolution layer to obtain a high-level feature set, wherein the high-level feature set comprises object shape features and overall structure features.
Further, the device further comprises:
acquiring a first identification model, wherein the first identification model is a support vector machine;
dividing the standard image set into a training set and a testing set, and taking the disturbance level as a category label;
and training the support vector machine by adopting the low-level feature set and the corresponding class label in the training set, evaluating the trained support vector machine by adopting the test set, and optimizing the support vector machine according to the evaluation result until the preset requirement is met, so as to obtain the support vector machine identifier.
Further, the trainer fusion module 15 includes the following steps:
acquiring a verification data set, wherein the verification data set comprises a verification standard image and a corresponding verification disturbance level;
inputting the verification standard image into the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier respectively to obtain a first identification result, a second identification result and a third identification result;
respectively carrying out similarity comparison on the first recognition result, the second recognition result and the third recognition result with the verification disturbance level, and acquiring a first weight, a second weight and a third weight according to the similarity comparison result;
and carrying out weighted fusion on the identifier according to the first weight, the second weight and the third weight to obtain the intelligent camera disturbance identifier.
Further, the trainer fusion module 15 includes the following steps:
performing difference value calculation on the first weight, the second weight and the third weight, and judging whether the difference value calculation is smaller than a preset threshold value or not;
when the difference value of any group of weights is larger than a preset threshold value, obtaining a weight maximum value;
and taking the identifier corresponding to the maximum weight value as an intelligent camera disturbance identifier.
Further, the device further comprises:
obtaining disturbance category and disturbance degree according to a camera disturbance identification result;
acquiring sensor data, wherein the sensor data comprises acceleration data and angular velocity data;
calculating and obtaining anti-shake parameters by combining the sensor data by using disturbance types and disturbance degrees;
and controlling the camera according to the anti-shake parameters.
Further, the device further comprises:
acquiring environmental monitoring data;
performing environmental impact assessment of the target camera based on the environmental monitoring data to obtain an environmental impact coefficient;
and when the environmental impact coefficient reaches a preset environmental impact threshold value, compensating the camera disturbance recognition result based on the environmental impact coefficient.
The foregoing detailed description of the integrated algorithm-based camera disturbance recognition method will be clear to those skilled in the art, and the device disclosed in this embodiment is relatively simple in description and relevant places refer to the method part for description, because the device corresponds to the method disclosed in the embodiment.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (8)

1. The camera disturbance recognition method based on the integrated algorithm is characterized by comprising the following steps of:
collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, and marking the disturbance level;
carrying out standardization processing on the marked image data set to obtain a standard image set;
carrying out multi-level feature extraction on the standard image set to obtain a plurality of feature sets, wherein the feature sets comprise a low-level feature set, a middle-level feature subset and a high-level feature set;
based on the low-level feature set, the middle-level feature set and the high-level feature set, training in sequence to obtain a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier;
fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier;
image acquisition is carried out on the random targets by using a target camera, and an image acquisition result is obtained, wherein the image acquisition result comprises camera disturbance;
and inputting the image acquisition result into the camera disturbance intelligent identifier, and acquiring a camera disturbance identification result of the target camera according to output information.
2. The method of claim 1, wherein performing multi-level feature extraction on the standard image set comprises:
a convolutional neural network is adopted as a feature extractor;
extracting low-level features from the first convolution layer to obtain a low-level feature set, wherein the low-level feature set comprises edge features, texture features and color features;
extracting middle-level features in a second convolution layer to obtain a middle-level feature set, wherein the middle-level feature set comprises corner features, contour features and texture combination features;
and extracting high-level features from the third convolution layer to obtain a high-level feature set, wherein the high-level feature set comprises object shape features and overall structure features.
3. The method as recited in claim 1, further comprising:
acquiring a first identification model, wherein the first identification model is a support vector machine;
dividing the standard image set into a training set and a testing set, and taking the disturbance level as a category label;
and training the support vector machine by adopting the low-level feature set and the corresponding class label in the training set, evaluating the trained support vector machine by adopting the test set, and optimizing the support vector machine according to the evaluation result until the preset requirement is met, so as to obtain the support vector machine identifier.
4. The method of claim 1, wherein fusing the support vector machine identifier, the decision tree identifier, the convolutional neural network identifier comprises:
acquiring a verification data set, wherein the verification data set comprises a verification standard image and a corresponding verification disturbance level;
inputting the verification standard image into the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier respectively to obtain a first identification result, a second identification result and a third identification result;
respectively carrying out similarity comparison on the first recognition result, the second recognition result and the third recognition result with the verification disturbance level, and acquiring a first weight, a second weight and a third weight according to the similarity comparison result;
and carrying out weighted fusion on the identifier according to the first weight, the second weight and the third weight to obtain the intelligent camera disturbance identifier.
5. The method as recited in claim 4, further comprising:
performing difference value calculation on the first weight, the second weight and the third weight, and judging whether the difference value calculation is smaller than a preset threshold value or not;
when the difference value of any group of weights is larger than a preset threshold value, obtaining a weight maximum value;
and taking the identifier corresponding to the maximum weight value as an intelligent camera disturbance identifier.
6. The method as recited in claim 1, further comprising:
obtaining disturbance category and disturbance degree according to a camera disturbance identification result;
acquiring sensor data, wherein the sensor data comprises acceleration data and angular velocity data;
calculating and obtaining anti-shake parameters by combining the sensor data by using disturbance types and disturbance degrees;
and controlling the camera according to the anti-shake parameters.
7. The method as recited in claim 1, further comprising:
acquiring environmental monitoring data;
performing environmental impact assessment of the target camera based on the environmental monitoring data to obtain an environmental impact coefficient;
and when the environmental impact coefficient reaches a preset environmental impact threshold value, compensating the camera disturbance recognition result based on the environmental impact coefficient.
8. Camera disturbance recognition device based on an integrated algorithm, characterized in that it comprises:
and a data acquisition module: collecting an image data set containing camera disturbance, recording camera motion parameters of the image data set, and marking the disturbance level;
standardized processing module: carrying out standardization processing on the marked image data set to obtain a standard image set;
and the feature extraction module is used for: carrying out multi-level feature extraction on the standard image set to obtain a plurality of feature sets, wherein the feature sets comprise a low-level feature set, a middle-level feature subset and a high-level feature set;
the recognizer training module: based on the low-level feature set, the middle-level feature set and the high-level feature set, training in sequence to obtain a support vector machine identifier, a decision tree identifier and a convolutional neural network identifier;
the trainer fusion module: fusing the support vector machine identifier, the decision tree identifier and the convolutional neural network identifier to obtain a camera disturbance intelligent identifier;
and an image acquisition module: image acquisition is carried out on the random targets by using a target camera, and an image acquisition result is obtained, wherein the image acquisition result comprises camera disturbance;
a disturbance identification module: and inputting the image acquisition result into the camera disturbance intelligent identifier, and acquiring a camera disturbance identification result of the target camera according to output information.
CN202311341923.9A 2023-10-17 2023-10-17 Camera disturbance recognition method and device based on integrated algorithm Pending CN117496224A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311341923.9A CN117496224A (en) 2023-10-17 2023-10-17 Camera disturbance recognition method and device based on integrated algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311341923.9A CN117496224A (en) 2023-10-17 2023-10-17 Camera disturbance recognition method and device based on integrated algorithm

Publications (1)

Publication Number Publication Date
CN117496224A true CN117496224A (en) 2024-02-02

Family

ID=89668010

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311341923.9A Pending CN117496224A (en) 2023-10-17 2023-10-17 Camera disturbance recognition method and device based on integrated algorithm

Country Status (1)

Country Link
CN (1) CN117496224A (en)

Similar Documents

Publication Publication Date Title
CN111179251B (en) Defect detection system and method based on twin neural network and by utilizing template comparison
US12020415B2 (en) Method for monitoring manufacture of assembly units
CN111563557B (en) Method for detecting target in power cable tunnel
CN117115147B (en) Textile detection method and system based on machine vision
CN111814850A (en) Defect detection model training method, defect detection method and related device
CN113435546B (en) Migratable image recognition method and system based on differentiation confidence level
CN111860106B (en) Unsupervised bridge crack identification method
CN112766110A (en) Training method of object defect recognition model, object defect recognition method and device
CN111882554B (en) SK-YOLOv 3-based intelligent power line fault detection method
CN110852358A (en) Vehicle type distinguishing method based on deep learning
CN114494780A (en) Semi-supervised industrial defect detection method and system based on feature comparison
CN115100497A (en) Robot-based method, device, equipment and medium for routing inspection of abnormal objects in channel
CN113052295A (en) Neural network training method, object detection method, device and equipment
CN112329893A (en) Data-driven heterogeneous multi-target intelligent detection method and system
CN111652048A (en) A deep learning based 1: n face comparison method
Yang et al. YOLOX with CBAM for insulator detection in transmission lines
KR102664206B1 (en) Aircraft inspection automation system based on image and explainable ai
CN116188445A (en) Product surface defect detection and positioning method and device and terminal equipment
CN117496224A (en) Camera disturbance recognition method and device based on integrated algorithm
CN114757941A (en) Transformer substation equipment defect identification method and device, electronic equipment and storage medium
CN110866440B (en) Pointer instrument deep learning network construction method, pointer instrument deep learning network detection method and pointer instrument deep learning network construction system
CN111626175A (en) Axial type identification method based on deep convolutional neural network
CN118505704B (en) General model modeling detection method for detecting defects of panel production line
CN111291624B (en) Excavator target identification method and system
CN111259843B (en) Multimedia navigator testing method based on visual stability feature classification registration

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