CN114694090A - Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5 - Google Patents
Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5 Download PDFInfo
- Publication number
- CN114694090A CN114694090A CN202210209202.1A CN202210209202A CN114694090A CN 114694090 A CN114694090 A CN 114694090A CN 202210209202 A CN202210209202 A CN 202210209202A CN 114694090 A CN114694090 A CN 114694090A
- Authority
- CN
- China
- Prior art keywords
- background
- algorithm
- pbas
- point
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 50
- 206010000117 Abnormal behaviour Diseases 0.000 title claims abstract description 46
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000000034 method Methods 0.000 claims abstract description 19
- 230000006399 behavior Effects 0.000 claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 239000000284 extract Substances 0.000 claims abstract description 6
- 238000001914 filtration Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 9
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000006872 improvement Effects 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000009432 framing Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- 238000002360 preparation method Methods 0.000 claims description 2
- 230000035945 sensitivity Effects 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims 1
- 230000003068 static effect Effects 0.000 abstract description 5
- 238000005286 illumination Methods 0.000 abstract description 3
- 230000007613 environmental effect Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 7
- 238000012544 monitoring process Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
A campus abnormal behavior detection method based on an improved PBAS algorithm and YOLOv5 comprises the following steps: 1) and (3) selecting a Yolov5 network for model training: collecting a large number of data samples, calibrating the samples, comparing the models set by different parameters to select final training parameters and models, and preparing for subsequent abnormal behavior detection; 2) the improved PBAS algorithm extracts dynamic foreground: the method comprises the steps of utilizing an improved PBAS algorithm to complete dynamic foreground extraction on a video motion area, capturing effective dynamic behaviors, and filtering static and dynamic backgrounds, so that the interference of environmental factors such as illumination, dynamic backgrounds and the like on target detection is shielded; 3) the Yolov5 model detects abnormal behavior: and performing target detection by taking the processed video frame as an input of a YOLOv5 model so as to determine whether the student has dangerous behaviors. The method has the advantages of accurate detection result, relatively low omission factor and false detection rate, and can well complete the detection of abnormal behaviors of the campus.
Description
Technical Field
The invention relates to a campus abnormal behavior detection method.
Background
In recent years, campus security has received unprecedented attention. An intelligent video monitoring system in a computer vision technology is urgently needed to be used for detecting abnormal behaviors in a campus and giving an alarm for the abnormal behaviors. The method has the advantages that important abnormal behaviors are detected and checked in the campus, the work tasks of monitoring personnel can be reduced, and the method has positive effects and practical significance on safety work management in the campus. For abnormal behavior detection, it can be seen as consisting of two steps: the first step is foreground extraction, namely abstracting a series of abnormal behavior features in a scene and extracting valuable behavior features. And the second step is target detection, and a model for detecting abnormal behaviors is established. However, due to the complexity of the scene and the diversity of the abnormal behaviors, how to efficiently extract the descriptive and distinctive features of the abnormal behaviors and effectively express the descriptive and distinctive features becomes a difficult point and a focus.
In the foreground extraction work, the invention adopts a PBAS algorithm. The PBAS algorithm is a Pixel-Based Adaptive background segmentation algorithm (PBAS) proposed by Hofmann et al in 2012, the background model of which is constructed by collecting background samples, adaptively optimizes the threshold and update rate in the model, and reduces the false detection rate of the background. The PBAS algorithm combines the advantages of the SACON algorithm and the VIBE algorithm, is improved on the basis of the advantages of the SACON algorithm and the VIBE algorithm, and is mainly characterized in that a control theory idea and a method for measuring the background complexity are introduced, so that the foreground judgment threshold and the background model updating rate are adaptively changed along with the change of the background complexity and are updated in time, the PBAS algorithm has a good real-time monitoring video processing effect, and has an adaptive background sample set updating strategy, and the PBAS algorithm can be better used in actual situations than a fixed updating sample set strategy of the VIBE algorithm.
For the target detection work, the present invention employs the YOLOv5 neural network. The YOLOv5 algorithm consists of three parts. The first part is an input end, and the input size of the training picture is 608 parts. The second part is the backbone network, which uses the CSPDarknet53 network to extract rich information features from the input image. The third part is a detection layer, the part adopts multi-scale detection, a new path aggregation network structure of a bottom-up path is added behind the characteristic pyramid network structure, fusion of characteristic information of different scales is realized, and then the generated three characteristic graphs are predicted. In addition, YOLOv5 continued to follow the YOLOv4 multiscale detection structure. After the features of the backbone network are extracted, two times of upsampling and three times of convolution are carried out to respectively realize the prediction of large, medium and small target categories and positions on different scales. YOLOv5 uses adaptive anchor box calculation to train data for different data sets, adaptively calculating the best anchor box value in the training set. YOLOv5 is a detection algorithm with accurate detection and high speed, and has good effect on an open source data set, but because the definition of monitoring cameras used by a part of schools is not high, the detection performance on an abnormal behavior recognition task still needs to be improved.
In summary, the existing foreground extraction and target detection technologies are rapidly developed, but considering that target identification is performed in a specific scene such as school, uncertainty often exists in the effect, so that selecting a proper algorithm and model or a specific identification strategy under different scene conditions is especially important for improving identification efficiency. Since the method mainly focuses on abnormal behaviors in the campus, the contents of the method mainly include common dangers occurring in the campus or actions and behaviors which do not meet campus specifications. Therefore, the method mainly selects three main abnormal behaviors for detection, namely falling, putting up and climbing of students.
Disclosure of Invention
The invention provides a campus abnormal behavior detection method based on an improved PBAS algorithm and YOLOv5, which aims to solve the problems that the prior campus abnormal detection technology in the prior art is low in precision and greatly influenced by the environment, and meets the real-time requirement.
The method extracts the dynamic foreground of the image based on the improved PBAS algorithm, then utilizes the trained network model to perform target detection, finally determines the abnormal behavior in the image, and provides a reliable abnormal behavior detection method for the security system.
The invention discloses a campus abnormal behavior detection method based on an improved PBAS algorithm and YOLOv5, which comprises the following steps:
1) and (3) selecting a Yolov5 network for model training: firstly, collecting a large number of data samples according to three types of abnormal behaviors to manufacture a data set, preprocessing a data set picture, then calibrating an abnormal behavior region of the data set picture, finally repeatedly training a model and adjusting parameters for many times, setting network parameters by combining a synchronous training result and a network structure, and finally obtaining an abnormal behavior detection model;
2) the improved PBAS algorithm extracts dynamic foreground: in the dynamic foreground extraction part, a dynamic area of a video frame is effectively extracted by improving a self-adaptive decision threshold updating mode of a PBAS algorithm, so that the interference of illumination, dynamic background and the like on foreground extraction is avoided;
3) the Yolov5 model detects abnormal behavior: and extracting a dynamic foreground from the video frame through an improved PBAS algorithm, and performing abnormal behavior detection as the input of a YOLOv5 model.
The technical conception of the invention is as follows: the technical route mainly aims at abnormal behavior detection in a campus specific scene and mainly comprises the following 3 steps: 1. collecting a large number of data samples, calibrating the samples, comparing the models set by different parameters to select final training parameters and models, and preparing for subsequent abnormal behavior detection. 2. And taking the video frame as input, framing the dynamic foreground in the image by utilizing an improved PBAS algorithm, and filtering out the static background so as to reduce the false detection of the static background. 3. And inputting the obtained dynamic foreground into a well-trained Yolov5 network model, detecting abnormal behaviors and finally framing an abnormal area.
The invention has the following beneficial effects: 1. the detection result is accurate, and the missing detection rate and the false detection rate are relatively low; 2. the campus abnormal behavior detection method can well complete campus abnormal behavior detection, and labor cost is greatly saved.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a graph of the average loss function of the present invention.
Fig 3 is a flow chart of the PBAS algorithm of the present invention.
Fig. 4(a) -4 (c) are comparison graphs of the effect of the PBAS algorithm before and after improvement, wherein fig. 4(a) is the original graph, fig. 4(b) is the effect graph before improvement, and fig. 4(c) is the effect graph after improvement.
Fig. 5 is a diagram illustrating the effect of detecting abnormal campus behaviors according to the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
A campus abnormal behavior detection method based on an improved PBAS algorithm and YOLOv5 comprises the following steps:
1) and (3) selecting a Yolov5 network for model training: collecting a large number of data samples, calibrating the samples, comparing the models set by different parameters to select final training parameters and models, and preparing for subsequent abnormal behavior detection;
2) the improved PBAS algorithm extracts dynamic foreground: the method comprises the steps of utilizing an improved PBAS algorithm to complete dynamic foreground extraction on a video motion area, capturing effective dynamic behaviors, and filtering static and dynamic backgrounds, so that the interference of environmental factors such as illumination, dynamic backgrounds and the like on target detection is shielded;
3) the Yolov5 model detects abnormal behavior: and performing target detection by taking the processed video frame as an input of a YOLOv5 model so as to determine whether the student has dangerous behaviors.
Further, in the step 1), a Yolov5 network is selected for model training: the specific steps are as follows,
1.1) data set preparation: firstly, extracting video frames of the collected three abnormal videos; then, carrying out picture preprocessing on a data set consisting of video frames and pictures, and strictly screening, filtering and denoising, image enhancement and the like on the pictures; and finally, framing the abnormal behavior region in the preprocessed picture by using a LabelImg plotting tool to generate a txt file for model training.
1.2) model training: after the trial of training the model repeatedly and combining the training result and the structure of the YOLOv5 network, the final setting of the relevant parameters is as follows: the initial learning rate is 0.01, the learning rate decay weight is 0.0005, and the number of training iterations is 200. The loss function of YOLOv5 uses GIOULoss as bounding Box, and the value inferred by Box is the mean value of GIoU loss function. The loss function is shown in fig. 2, and it can be seen that the smaller the marking box is, the more accurate the detection result is.
In the step 2), the PBAS algorithm is improved to extract a dynamic foreground: and (4) screening areas suspected of abnormal behaviors of students, and by combining with the graph 3, firstly, completing accurate extraction of dynamic foreground by using an improved PBAS algorithm, and then, performing target detection by taking the framed dynamic area as the input of the step 3), thereby determining whether dangerous behaviors exist in the students. The method comprises the following specific steps:
2.1) inputting a video frame, and determining the classification of the foreground and the background by the background model and the current pixel of the image. Wherein the background model is composed of N historical pixel values observed adjacent to the current video frame:
B(xi)＝B1(xi)，...，Bk(xi)；...，BN(xi) (1)
meanwhile, the PBAS algorithm passes the current frame I (x)i) With background model B (x)i) The comparison determines that the current pixel belongs to the background or the foreground, specifically, the previous N historical pixel values of a pixel point in the background model are compared with the current pixel value, if the distance between the current value and at least # min historical values is smaller than a judgment threshold value R (x)i) If not, the point is determined as a foreground point, otherwise, the point is determined as a background point.
2.2) updating an improved self-adaptive decision threshold, and determining a foreground segmentation mask calculation formula as follows according to a PBAS algorithm and the step 2.1):
where F ═ 0 and F ═ 1 respectively indicate that the pixel point is a background point and a foreground point, dist (I (x) is presenti),Bk(xi) ) represents the distance of the current point from the background model. In order to solve the problem of extracting dynamic backgrounds together by mistake when extracting dynamic foregrounds, a nonlinear adaptive decision threshold updating method is adopted, and a decision threshold R (x) is utilizedi) Complexity of and background
The nonlinear relation improves the sensitivity of small-area dynamic background identification, thereby solving the problem that the dynamic background is extracted by mistake. The specific improvement steps are as follows:
2.2.1) firstly, when the algorithm judges that the pixel points in the motion area are foreground points or background points through the target frame, calculating the area of the target frame to be Starget(xi) Simultaneously with a balancing function Rban(xi) And set to a fixed absolute value.
2.2.2) then mix R (x)i) And
redefined from the original linear relationship to a non-linear relationship, i.e.
Wherein, R is definedban(xi) Is a fixed absolute value. The target frame area due to the dynamic background is usually small, i.e. when Starget(xi) The smaller the pixel xiThe more likely it is a background point, Rban(xi) And Starget(xi) The larger the ratio of (x) to (x) is, the lower the decision threshold R (x) in the original PBAS algorithmi) Improved decision threshold R (x) compared to the valuei) The value is changed more, and when the foreground or background decision is made, if R (x) is greater, the formula (5) can show thati) The more valueThe size of the product is large,
value less than decision threshold R (x)i) If the number of times is more than # min, the pixel point is judged as a background point at the moment, and a region with more complex background change can be well, quickly and effectively inhibited. When S istarget(xi) When a certain value is exceeded, the pixel points in the motion area have a larger probability of being foreground points Rban(xi) And Starget(xi) The ratio of (x) also tends to 0, the decision threshold R (x)i) The complexity of the target frame is still linear with the background, so the area S of the target frametarget(xi) And the judgment of the motion prospect is not influenced if the size is too large. From the above analysis, the decision threshold R (x) can be seti) Redefining the dynamic updating mode as follows:
wherein, B (x)i) Is a state variable, Rban(xi) Is a fixed absolute value.
2.3) updating the background model if the current pixel point xiIf the background point is judged to be the background point, the pixel point in the background sample is randomly and uniformly replaced by the pixel point, Bk(xi) K ∈ 1, …, N. On the basis, the sample point in the current pixel point sample set is 1/T (x)i) Is replaced, and each pixel point corresponds to a probability value 1/T (x)i)。
2.4) judging a threshold value and updating a learning rate, introducing background complexity to enable a background decision threshold value to be updated in a self-adaptive manner, if a pixel point is greatly different from a background model, detecting the pixel point as a foreground point, and updating the background model by the pixel point with relatively low probability. Therefore, the PBAS algorithm also establishes an array D (x) when establishing the background modeli) The minimum distance value of the similarity is recorded, and the background complexity is the average value of the minimum distance of the similarity.
Wherein,
is N0The average of the minimum distance matrix values, i.e., the background complexity. According to the above analysis, when the background changes greatly, the background complexity
The larger the determination threshold R (x) is, the larger the determination threshold R (x) isi) And T (x)i). Therefore, the PBAS algorithm adaptively adjusts the discrimination threshold as follows:
wherein R isindeAnd RscaleAre all preset fixed values. Meanwhile, in order to reduce the influence on the established background model when the pixel point is judged by mistake, the frequency of background updating needs to be reduced. Therefore, the learning rate T (x)i) The update policy of (a) may be expressed as:
wherein, TincAnd TdecIs a fixed value set in advance.
Fig. 4 is a pre-and post-processing comparison graph through the PBAS algorithm. By extracting the dynamic foreground in the video motion area, the dynamic foreground and the static background can be effectively separated, so that the influence of the video background on the abnormal behavior identification is greatly reduced, and the accuracy of the abnormal behavior detection is improved.
In the step 3), the Yolov5 model detects abnormal behaviors: and extracting a dynamic foreground from the video frame through an improved PBAS algorithm, and performing abnormal behavior detection as the input of a YOLOv5 model. The input end of YOLOv5 utilizes a Mosaic data enhancement method, 4 pictures are randomly used and used as training data after being zoomed, cut and spliced, and the picture background is enriched. Meanwhile, self-adaptive anchor frame calculation and image scaling are adopted to reduce the calculation amount and improve the target detection speed. In addition, the YOLOv5 adopts GIoU _ Loss as a Loss function of the bounding box at the output end, and screens the target frame by suppressing NMS through a non-maximum value, thereby effectively solving the problem of misalignment of the frames.
Fig. 5 is a result of the detection by the method, and it can be seen that interference caused by a dynamic background is effectively suppressed by improving the algorithm to update the adaptive decision threshold, so that the dynamic foreground of the video frame image is accurately extracted. And finally, realizing accurate identification of abnormal behaviors.
The embodiments described in this specification are merely illustrative of implementations of the inventive concepts, which are intended for purposes of illustration only. The scope of the present invention should not be construed as being limited to the particular forms set forth in the examples, but rather as being defined by the claims and the equivalents thereof which can occur to those skilled in the art upon consideration of the present inventive concept.
Claims (3)
1. A campus abnormal behavior detection method based on an improved PBAS algorithm and YOLOv5 comprises the following steps:
1) model training was performed using Yolov5 network: collecting a large number of data samples, calibrating the samples, comparing models under different parameter settings, and selecting a final training parameter and a model;
2) the improved PBAS algorithm extracts dynamic foreground: the method comprises the steps of utilizing an improved PBAS algorithm to complete dynamic foreground extraction on a video motion area and capturing effective dynamic behaviors;
3) detecting abnormal behaviors by using a Yolov5 model: and performing target detection by taking the processed video frame as an input of a YOLOv5 model so as to determine whether the student has dangerous behaviors.
2. The method of claim 1, wherein the improved PBAS algorithm and YOLOv 5-based campus abnormal behavior detection method comprises the following steps:
1.1) data set preparation: extracting video frames of the collected three abnormal videos; then, carrying out image preprocessing on the data set, screening the image, filtering and denoising the image, and enhancing the image; finally, framing the abnormal behavior region in the preprocessed picture by using a LabelImg plotting tool to generate a txt file for model training;
1.2) model training: through the trial of repeatedly training the model, and simultaneously combining the training result and the structure of the YOLOv5 network, the related parameters are set as follows: the initial learning rate is 0.01, the learning rate attenuation weight is 0.0005, and the training iteration times are 200; the Loss function of YOLOv5 uses GIOU Loss as a bounding Box, and the value inferred by the Box is the mean value of the GIOU Loss function.
3. The method for detecting abnormal campus behaviors based on the improved PBAS algorithm and YOLOv5 of claim 1 or 2, wherein the step 2) specifically includes:
2.1) inputting a video frame, and determining the classification of a foreground and a background by a background model of an image and a current pixel; wherein the background model is composed of N historical pixel values observed adjacent to the current video frame:
B(xi)＝B1(xi)，...，Bk(xi)，...，BN(xi) (1)
meanwhile, the PBAS algorithm passes the current frame I (x)i) With background model B (x)i) The comparison determines that the current pixel belongs to the background or the foreground, specifically, the previous N historical pixel values of a pixel point in the background model are compared with the current pixel value, if the distance between the current value and at least # min historical values is smaller than a judgment threshold value R (x)i) If the point is determined as a foreground point, otherwise, the point is determined as a background point;
2.2) updating an improved self-adaptive decision threshold, and determining a foreground segmentation mask calculation formula as follows according to a PBAS algorithm and the step 2.1):
where F ═ 0 and F ═ 1 respectively indicate that the pixel point is a background point and a foreground point, dist (I (x) is presenti),Bk(xi) ) represents the distance of the current point from the background model. In order to solve the problem of extracting dynamic backgrounds together by mistake when extracting dynamic foregrounds, a nonlinear adaptive decision threshold updating method is adopted, and a decision threshold R (x) is utilizedi) Complexity of and background
The nonlinear relation improves the sensitivity of small-area dynamic background identification, thereby solving the problem that the dynamic background is extracted by mistake. The specific improvement steps are as follows:
2.2.1) firstly, when the algorithm judges that pixel points in the motion area are foreground points or background points through a target frame, calculating the area of the target frame to be Starget(xi) Simultaneously with a balancing function Rban(xi) And set to a fixed absolute value;
2.2.2) then mix R (x)i) And
redefining the original linear relation as the nonlinear relation. Namely, it is
Wherein R is definedban(xi) Is a fixed absolute value. The area of the target frame due to the dynamic background is usually small, i.e. when Starget(xi) The smaller the pixel xiThe more likely it is a background point, Rban(xi) And Starget(xi) The larger the ratio of (x) to (x) is, the lower the decision threshold R (x) in the original PBAS algorithmi) Improved decision threshold R (x) compared to the valuei) The value is changed more, and when the foreground or background decision is made, if R (x) is greater, the formula (5) can show thati) The larger the value is，
Value less than decision threshold R (x)i) If the number of times is more than # min, the pixel point is judged as a background point at the moment, and a region with more complex background change can be well, quickly and effectively inhibited. When S istarget(xi) When a certain value is exceeded, the pixel points in the motion area have a larger probability of being foreground points Rban(xi) And Starget(xi) The ratio of (x) also tends to 0, the decision threshold R (x)i) The complexity of the target frame is still linear relation with the background, so the area S of the target frametarget(xi) And the judgment of the motion prospect is not influenced if the size is too large. From the above analysis, the decision threshold R (x) can be seti) Redefining the dynamic updating mode as follows:
wherein, B (x)i) Is a state variable, Rban(xi) Is a fixed absolute value.
2.3) updating the background model if the current pixel point xiIf the background point is judged to be the background point, the pixel point in the background sample is randomly and uniformly replaced by the pixel point, Bk(xi) K ∈ 1, …, N; on the basis, the sample point in the current pixel point sample set is 1/T (x)i) Is replaced, and each pixel point corresponds to a probability value 1/T (x)i)；
2.4) judging a threshold value and updating a learning rate, introducing background complexity to enable a background decision threshold value to be updated in a self-adaptive manner, if a pixel point is greatly different from a background model, detecting the pixel point as a foreground point, and updating the background model by using a relatively small probability; therefore, the PBAS algorithm also establishes an array D (x) when establishing the background modeli) The method is used for recording the minimum distance value of the similarity, and the background complexity is the average value of the minimum distance of the similarity;
wherein,
is N0The average of the minimum distance matrix values, i.e., the background complexity. According to the above analysis, when the background changes greatly, the background complexity
The larger the determination threshold R (x) is, the larger the determination threshold R (x) isi) And T (x)i) (ii) a Therefore, the PBAS algorithm adaptively adjusts the discrimination threshold as follows:
wherein R isindeAnd RscaleAll are preset fixed values; meanwhile, in order to reduce the influence on the established background model when the pixel point is judged by mistake, the frequency of background updating needs to be reduced; therefore, the learning rate T (x)i) The update policy of (a) may be expressed as:
wherein, TincAnd TdecIs a fixed value set in advance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210209202.1A CN114694090A (en) | 2022-03-04 | 2022-03-04 | Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210209202.1A CN114694090A (en) | 2022-03-04 | 2022-03-04 | Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114694090A true CN114694090A (en) | 2022-07-01 |
Family
ID=82137305
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210209202.1A Pending CN114694090A (en) | 2022-03-04 | 2022-03-04 | Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114694090A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115294556A (en) * | 2022-09-28 | 2022-11-04 | 西南石油大学 | Improved YOLOv 5-based method for detecting abnormal flow state fluid on closed vibrating screen |
-
2022
- 2022-03-04 CN CN202210209202.1A patent/CN114694090A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115294556A (en) * | 2022-09-28 | 2022-11-04 | 西南石油大学 | Improved YOLOv 5-based method for detecting abnormal flow state fluid on closed vibrating screen |
| CN115294556B (en) * | 2022-09-28 | 2022-12-13 | 西南石油大学 | Improved YOLOv 5-based method for detecting abnormal flow state fluid on closed vibrating screen |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112200043B (en) | Intelligent danger source identification system and method for outdoor construction site | |
| CN108446617B (en) | Side face interference resistant rapid human face detection method | |
| CN105404847B (en) | A kind of residue real-time detection method | |
| CN110084165B (en) | Intelligent identification and early warning method for abnormal events in open scene of power field based on edge calculation | |
| CN111563557B (en) | Method for detecting target in power cable tunnel | |
| WO2019140767A1 (en) | Recognition system for security check and control method thereof | |
| CN107909081B (en) | Method for quickly acquiring and quickly calibrating image data set in deep learning | |
| CN111222478A (en) | Construction site safety protection detection method and system | |
| WO2019114145A1 (en) | Head count detection method and device in surveillance video | |
| CN113553979B (en) | Safety clothing detection method and system based on improved YOLO V5 | |
| CN108647625A (en) | A kind of expression recognition method and device | |
| CN110298297A (en) | Flame identification method and device | |
| CN109360179B (en) | Image fusion method and device and readable storage medium | |
| CN113642474A (en) | Hazardous area personnel monitoring method based on YOLOV5 | |
| CN112926522B (en) | Behavior recognition method based on skeleton gesture and space-time diagram convolution network | |
| CN115620212A (en) | Behavior identification method and system based on monitoring video | |
| CN113177439B (en) | Pedestrian crossing road guardrail detection method | |
| CN111476160A (en) | Loss function optimization method, model training method, target detection method, and medium | |
| CN117197746A (en) | Safety monitoring system and method based on deep learning | |
| CN112597928A (en) | Event detection method and related device | |
| CN116758421A (en) | Remote sensing image directed target detection method based on weak supervised learning | |
| CN114694090A (en) | Campus abnormal behavior detection method based on improved PBAS algorithm and YOLOv5 | |
| CN117475353A (en) | Video-based abnormal smoke identification method and system | |
| CN117132768A (en) | License plate and face detection and desensitization method and device, electronic equipment and storage medium | |
| CN107992799A (en) | Towards the preprocess method of Smoke Detection application |
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 |