A method and system for detecting abnormal targets in campus safety passages

Abstract

The present invention provides a method and system for detecting abnormal targets in campus safety passages, the method comprising: obtaining a monitoring screen of real-time personnel flow; extracting a feature map of the monitoring screen; iterating the feature map using a regression convolution prediction block to generate a new subset; generating abnormal target detection results through a convolutional neural network; wherein the convolutional neural network merges the prediction results of each round of iteration with the prediction results of the previous round to generate a final result. The method for detecting abnormal targets in campus safety passages proposed by the present invention can achieve good results for detecting abnormal targets in campus safety passages. The method can not only identify various objects in crowded and dense environments, but also has the characteristics of accuracy and high speed.

Description

Technical Field

[0001] The present invention relates to the technical field of campus security, and in particular to a method and system for detecting abnormal targets in campus safety passages. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As an important branch of computer vision, object detection has received close attention from many scholars from world-class research institutions such as Stanford University, California Institute of Technology and Tsinghua University in recent years. Object detection can provide valuable information for the semantic understanding of images or videos. It is a basic step towards computer vision applications such as target tracking, pedestrian detection, and cross-border tracking. It is also widely used in intelligent video surveillance, unmanned driving, regional security and other fields.

[0004] Existing survey results show that the number of security guards in most schools is insufficient to cover the entire area of ​​the campus safety passage during the students' commute. When an abnormal target breaks in, it is difficult for school security to respond immediately, and there are great safety hazards in the campus safety passage. Accurate detection of abnormal targets in the campus safety passage is an important method to ensure the traffic safety of students during the commute. Applying the target detection algorithm to the abnormal target detection of the campus safety passage can use the deep convolutional neural network to automatically obtain the characteristics of multiple levels of feature representation from raw pixels to high-level semantic features, automatically detect abnormal targets, and maintain the safety of the campus safety passage, which has important application value in building a safe campus.

[0005] The object detection framework can be mainly divided into two categories: two-stage detectors and one-stage detectors. The two-stage detector first uses a proposal generator to generate a set of sparse proposals, extracts features from each proposal, and then uses a region classifier to predict the category of the proposal region. The one-stage detector directly classifies and predicts the target at each position of the feature map. Compared with the two-stage detector, the one-stage detector has a simple detection process, significantly improved detection speed, and stronger applicability in the detection of real-time objects. Traditional two-stage target detection algorithms such as R-CNN are difficult to achieve multi-scale target detection. When faced with crowded and complex environments such as campus safety passages, problems such as repeated detection, missed detection, and misjudgment are prone to occur, which seriously affect the detection accuracy. Therefore, this method uses a one-stage detector when detecting abnormal targets in campus safety passages.

[0006] The abnormal target detection method for campus safety passages based on iterative extraction of deep expansion features has a low missed detection rate in accurate and high-speed conditions and good performance in highly crowded crowds. It is conducive to the accurate detection of threatening abnormal targets in campus safety passages, effectively solving the problems of repeated detection, missed detection and misjudgment of traditional target detection algorithms, and improving detection efficiency.

[0007] In the process of implementing the present invention, the inventors found that the current technology has the following problems:

[0008] 1. Traditional target detection algorithms are less suitable for detecting abnormal targets in campus channels.

[0009] 2. There are few reports on advanced methods for abnormal target detection in campus channels based on deep learning. Summary of the invention

[0010] In order to solve the above problems, the present invention proposes a method and system for detecting abnormal targets in campus safety passages. The present invention can detect crowded people, small targets and other abnormal targets in campus safety passages.

[0011] According to some embodiments, the present invention adopts the following technical solutions:

[0012] A method for detecting abnormal targets in a campus safety channel, comprising:

[0013] Get real-time monitoring footage of personnel flow;

[0014] Extracting a feature graph of the monitoring screen;

[0015] Iterating the feature map using a regression convolution prediction block to generate a new subset;

[0016] Generate abnormal target detection results through convolutional neural network;

[0017] The convolutional neural network combines the prediction results of each round of iteration with the prediction results of the previous round to generate the final result.

[0018] Furthermore, the extracting of the feature map of the monitoring screen includes inputting the monitoring screen into a sampling network, performing downsampling through several stages, and extracting the feature map.

[0019] Furthermore, the sampling network includes a pre-trained ResNet-50 network and a new network layer, and the new network layer includes multiple BottleNeck modules.

[0020] Furthermore, the iterative feature map using the regression convolution prediction block includes connecting the feature map to the new branches for regression and classification respectively.

[0021] Furthermore, the regression convolution prediction block includes a first predictor, a second predictor and a third predictor connected in cascade.

[0022] Further, the feature map is passed through a first predictor to generate splitting and regression information, as well as confidence scores and regression suggestions.

[0023] Furthermore, the second predictor receives the prediction result of the first predictor, and inputs the generated confidence score and regression suggestion into the third predictor to generate the confidence score and regression suggestion.

[0024] A campus safety channel abnormal target detection system, comprising:

[0025] The data acquisition module is configured to acquire real-time monitoring images of personnel flow;

[0026] A feature extraction module is configured to extract a feature map of the monitoring screen;

[0027] A regression analysis module is configured to iterate the feature map using a regression convolution prediction block to generate a new subset;

[0028] The target generation module is configured to generate abnormal target detection results through a convolutional neural network;

[0029] The convolutional neural network combines the prediction results of each round of iteration with the prediction results of the previous round to generate the final result.

[0030] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, a method for detecting abnormal targets in a campus safety channel.

[0031] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor. A method for detecting abnormal targets in a campus safety channel.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention proposes a method for detecting abnormal targets in campus safety passages, which can achieve good results for detecting abnormal targets in campus safety passages. The method can not only identify various objects in crowded and dense environments, but also has the characteristics of accuracy and high speed. The deep expansion feature iterative extraction target detection algorithm uses a series of expansion evolutionary layers to build a new structure, generate and increase the perception field from the deep network, and can extract richer deep features.

[0034] The proposed method of deep dilation feature iterative extraction target detection algorithm includes a new deep feature extraction block, a regression convolution prediction block for regression and classification, and a new loss function to improve network training. The combined effect of these three parts will produce higher accuracy.

[0035] The deep dilation feature iterative extraction target detection algorithm generates a new subset after each iteration. Each iteration considers the target boxes generated by all previous iterations and the previous target boxes are passed to the neural network in the current iteration to ensure that there will be no repeated detection and missed detection in highly dense environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0037] Figure 1 is an architectural diagram of this embodiment;

[0038] Figure 2 This is a diagram of the network layer structure newly added in the 6th stage of this embodiment;

[0039] Figure 3 This is a diagram of the regression convolution prediction block structure of this embodiment. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0041] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0042] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0043] Example 1.

[0044] like Figure 1 As shown, a method for detecting abnormal targets in a campus safety channel includes:

[0045] Get real-time monitoring footage of personnel flow;

[0046] Extracting a feature graph of the monitoring screen;

[0047] Iterating the feature map using a regression convolution prediction block to generate a new subset;

[0048] Generate abnormal target detection results through convolutional neural network;

[0049] The convolutional neural network combines the prediction results of each round of iteration with the prediction results of the previous round to generate the final result.

[0050] The method of extracting the feature map of the monitoring screen includes inputting the monitoring screen into a sampling network, performing downsampling through several stages, and extracting the feature map.

[0051] The sampling network includes a pre-trained ResNet-50 network and a new network layer, and the new network layer includes multiple BottleNeck modules.

[0052] The iterative feature map using the regression convolution prediction block includes connecting the feature map to the new branches for regression and classification respectively.

[0053] The regression convolution prediction block includes a first predictor, a second predictor and a third predictor which are cascaded.

[0054] The feature map is passed through a first predictor to generate splitting and regression information, as well as confidence scores and regression suggestions.

[0055] The second predictor receives the prediction result of the first predictor, and inputs the generated confidence score and regression suggestion into the third predictor to generate the confidence score and regression suggestion.

[0056] Specifically, the experimental platform and environment of the present invention are as follows:

[0057] The experimental platform of the present invention uses the Ubuntu 20.04 operating system, the GPU model of the computer is RTX3090, and the CPU model is i9-10900K. The present invention is carried out under the Pytorch deep learning framework, by calling the existing function packages in Pytorch, reading the pre-trained model of the network, optimizing the algorithm, and building the overall network. The program is programmed in Python language, and by calling function packages such as Pillow, Numpy, Opencv, Matplotlib, etc., code is written for the design of loss functions, calculation of related indicators, etc.

[0058] The specific embodiments of the present invention are as follows:

[0059] 1. Use the existing monitoring equipment in the school to collect real-time monitoring images of the safe passage during the students' commute. After the image is input into the network, the feature map is extracted. Downsampling is performed in five stages, and the downsampling coefficients in each stage are 2, 4, 8, 16, and 32 respectively. Figure 1 As shown in , the network backbone is ResNet-50 pre-trained on ImageNet, and then a new network layer is added at the end. Figure 2 As shown in Figure 1, in the new network layer, three BottleNeck modules are used, one of which has a 1×1 convolution branch and the other two have no convolution branches. And the same number of channels as in the fifth stage is used to reduce the computational effort of the dilated convolution.

[0060] 2. The network is truncated in stages 3, 4, and 5, and the feature maps are output. A deeper feature extraction is performed in stage 6. The feature maps extracted in stages 3, 4, 5, and 6 are connected to the new branches for regression and classification respectively.

[0061] 3. The part used for regression and classification on the new branch is called the regression convolution prediction block, such as Figure 3 As shown in the figure, it is composed of three convolutional prediction blocks in cascade. First, the input passes through a 3×3 convolutional layer with 256 channels to generate classification and regression information, and then generates confidence scores and regression suggestions. Then, the output of the first predictor is input to the second identical predictor to perform classification and regression again, generating confidence scores and regression suggestions. Then, the output of the second predictor is input to the third identical predictor to perform classification and regression, generating confidence scores and regression suggestions, and the final result is the result of merging the results generated each time.

[0062] 4. The loss function consists of two parts: classification loss and regression loss, which alleviates the imbalance between the number of positive samples and the number of negative samples, as shown below:

[0063] L=L classfication +L location

[0064] Where L represents the total loss, L classfication and L location represent classification loss and regression loss respectively.

[0065] The classification loss function is defined as:

[0066]

[0067] The regression loss function is defined as:

[0068]

[0069]

[0070] Where P a represents the probability of a positive sample, β represents the weighting parameter, γ represents the focusing parameter, and N loc represents the number of anchor box locations, P*a represents the predictability of the anchor box, and L reg represents the smooth L1 loss function, o a represents the bounding box coordinates generated during anchor box prediction, and o*a represents the ground truth coordinates of the anchor box.

[0071] 5. The new subsets generated in each iteration are merged with the previous prediction results through the convolutional neural network. When the iteration process no longer generates new subsets, the iteration ends. At this time, the generated detection result is the final abnormal target detection result.

[0072] 6. In order to verify the detection advantage of the present invention in a crowded and dense environment, the present invention has conducted a large number of experiments under the WiderPerson data set. The experimental results show that the abnormal target detection algorithm for campus safety passages based on deep expansion feature iterative extraction proposed by the present invention can achieve good results. This method can not only identify various objects in a crowded and dense environment, but also has the characteristics of accuracy and high speed. This shows that the target detection algorithm proposed by the present invention is effective, and provides a better method for obtaining high-accuracy and high-quality detection results, and has certain practical value.

[0073] 7. The background sets motor vehicles, non-motor vehicles, etc. around the campus safety passage as abnormal targets. When students go to and from school, when abnormal targets approach the campus safety passage, the system will automatically detect the abnormal targets and record them in the background.

[0074] Example 2.

[0075] A campus safety channel abnormal target detection system, comprising:

[0076] The data acquisition module is configured to acquire real-time monitoring images of personnel flow;

[0077] A feature extraction module is configured to extract a feature map of the monitoring screen;

[0078] A regression analysis module is configured to iterate the feature map using a regression convolution prediction block to generate a new subset;

[0079] The target generation module is configured to generate abnormal target detection results through a convolutional neural network;

[0080] The convolutional neural network combines the prediction results of each round of iteration with the prediction results of the previous round to generate the final result.

[0081] Example 3.

[0082] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device. A method for detecting abnormal targets in a campus safety channel provided in this embodiment.

[0083] Example 4.

[0084] A terminal device includes a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor. A campus safety channel abnormal target detection method provided in this embodiment.

[0085] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0086] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0090] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal targets in campus safety passages, characterized in that: include: Get real-time monitoring footage of personnel flow; Set the motor vehicles and non-motor vehicles around the campus safety passage in the background as abnormal targets. When there are abnormal targets approaching the campus safety passage during the students' commuting time, the system will automatically detect the abnormal targets and record them in the background; Extracting a feature graph of the monitoring screen; Iterating the feature map using a regression convolution prediction block to generate a new subset; Generate abnormal target detection results through convolutional neural network; The convolutional neural network combines the prediction results of each round of iteration with the prediction results of the previous round to generate the final result. Specifically, the new subset generated by each round of iteration is combined with the previous prediction result through the convolutional neural network. When the iterative process no longer generates new subsets, the iteration ends. At this time, the generated detection result is the final abnormal target detection result. The extracting feature graph of the monitoring picture includes inputting the monitoring picture into a sampling network, performing downsampling through several stages, and extracting the feature graph; The regression convolution prediction block includes a first predictor, a second predictor and a third predictor cascaded; The loss function used in the network training consists of two parts: classification loss and regression loss: ;in L represents the total loss, L classfication and L location Represent classification loss and regression loss respectively; the classification loss function is defined as: ; The regression loss function is defined as: ; ; in P a represents the probability of a positive sample, β represents the weighting parameter, γ represents the focusing parameter, N loc represents the number of anchor box locations, P*a represents the predictability of the anchor box, L reg represents the smooth L1 loss function, o a represents the bounding box coordinates generated during anchor box prediction, o*a Represents the ground truth coordinates of the anchor box.

2. A campus safety channel abnormal target detection method as claimed in claim 1, characterized in that: The sampling network includes a pre-trained ResNet-50 network and a new network layer, and the new network layer includes multiple BottleNeck modules.

3. A campus safety channel abnormal target detection method as claimed in claim 1, characterized in that: The iterative feature map using the regression convolution prediction block includes connecting the feature map to the new branches for regression and classification respectively.

4. A campus safety channel abnormal target detection method as claimed in claim 1, characterized in that: The feature map is passed through a first predictor to generate classification and regression information, as well as confidence scores and regression suggestions.

5. A campus safety channel abnormal target detection method as claimed in claim 4, characterized in that: The second predictor receives the prediction result of the first predictor, and inputs the generated confidence score and regression suggestion into the third predictor to generate the confidence score and regression suggestion.

6. A campus safety channel abnormal target detection system, characterized in that: include: The data acquisition module is configured to acquire real-time monitoring images of personnel flow; Set the motor vehicles and non-motor vehicles around the campus safety passage in the background as abnormal targets. When there are abnormal targets approaching the campus safety passage during the students' commuting time, the system will automatically detect the abnormal targets and record them in the background; A feature extraction module is configured to extract a feature map of the monitoring screen; A regression analysis module is configured to iterate the feature map using a regression convolution prediction block to generate a new subset; The target generation module is configured to generate abnormal target detection results through a convolutional neural network; The convolutional neural network combines the prediction results of each round of iteration with the prediction results of the previous round to generate the final result. Specifically, the new subset generated by each round of iteration is combined with the previous prediction result through the convolutional neural network. When the iterative process no longer generates new subsets, the iteration ends. At this time, the generated detection result is the final abnormal target detection result. The extracting feature graph of the monitoring picture includes inputting the monitoring picture into a sampling network, performing downsampling through several stages, and extracting the feature graph; The regression convolution prediction block includes a first predictor, a second predictor and a third predictor cascaded; The loss function used in the network training consists of two parts: classification loss and regression loss: ;in L represents the total loss, L classfication and L location Represent classification loss and regression loss respectively; the classification loss function is defined as: ; The regression loss function is defined as: ; ; in P a represents the probability of a positive sample, β represents the weighting parameter, γ represents the focusing parameter, N loc represents the number of anchor box locations, P*a represents the predictability of the anchor box, L reg represents the smooth L1 loss function, o a represents the bounding box coordinates generated during anchor box prediction, o*a Represents the ground truth coordinates of the anchor box.

7. A computer-readable storage medium, characterized in that: A plurality of instructions are stored therein, and the instructions are suitable for being loaded and executed by a processor of a terminal device, according to any one of claims 1-5, in a method for detecting abnormal targets in a campus safety passage.

8. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executed by a campus safety channel abnormal target detection method described in any one of claims 1-5.

Images