Railway perimeter security system and method based on machine vision and sound wave expelling device

Abstract

The invention discloses a railway perimeter security and protection system and method based on machine vision and a sound wave expelling device, and relates to the technical field of railway security and protection, and the system comprises a monitoring module, a target recognition module, a decision module and an expelling module; wherein the monitoring module is used for acquiring a key video frame containing an intrusion target in a to-be-monitored area of a railway system; the target identification module is used for carrying out target identification on the intrusion target in the video frame based on an acoustic dispersion detection model, and determining the type of the intrusion target, the distance of the intrusion target, the moving speed and the existence time; the decision module is used for matching a target sound wave sample from a preset sound source database; and the expelling module is used for emitting corresponding expelling sound waves to the invasion target based on the target sound wave sample. According to the invention, the technical problem that an existing security technology lacks an active intervention means for an emergent event and cannot prevent and intervene the event when the event occurs is solved.

Description

Technical Field

[0001] The present invention relates to the field of railway security technology, and in particular to a railway perimeter security system and method based on machine vision and sound wave repelling devices. Background Art

[0002] Key railway sites, such as stations, railway hubs, and substations, are key components of national infrastructure. The safety of these areas is crucial to ensuring smooth transportation and public safety. However, with the increase in railway lines and the increase in train speeds, safe railway operations have become an important issue that needs to be addressed in railway development. Foreign object intrusion incidents are sudden and random, and the time and place of occurrence are often difficult to predict. Traditional railway perimeter foreign object intrusions are mostly detected by manual inspections, which are often affected by uncertain factors such as time, region, physical condition, and energy. Foreign object intrusion incidents are sudden and unpredictable, and loopholes and mistakes are inevitable during the inspection process. At the same time, the operation of trains will also pose a certain threat to the personal safety of inspectors.

[0003] In order to improve detection efficiency and reduce manual labor, a variety of automated foreign body detection technologies have been applied to the railway industry. At present, the common automated foreign body detection methods are: power grid detection method, fiber grating detection method, video detection method, radar detection method, ultrasonic detection method and infrared barrier detection method. The main technical means of front-end detection equipment in the high-speed railway perimeter intrusion alarm system are light wave cross-beam detection technology, pulse electronic fence detection technology, vibration cable detection technology, vibration fiber optic detection technology, tension fence detection technology, multi-dimensional three-dimensional vibration detection technology and video intelligent analysis detection technology. However, due to the high intensity of manual monitoring, poor equipment reliability, poor equipment generalization performance, and greater installation difficulty, these automated foreign body detection and alarm methods have many limitations.

[0004] Faced with the current situation and difficulties of foreign object detection such as the large variety of foreign objects on railways, the large total railway mileage, the accelerated speed of rail vehicles, and the weak accident prevention capabilities, the perimeter security technical measures for key railway locations in my country mainly use optical technology and Internet of Things technology. At present, these security technologies lack the means to actively intervene in sudden incidents and are unable to prevent and intervene in incidents when they occur. Summary of the invention

[0005] In order to solve the above technical problems existing in the prior art, the embodiment of the present invention provides a railway perimeter security system and method based on machine vision and acoustic wave repelling device. The technical solution is as follows:

[0006] On the one hand, a railway perimeter security system based on machine vision and acoustic wave repelling device is provided, comprising: a monitoring module, a target recognition module, a decision module and a repelling module; wherein the monitoring module is used to obtain a key video frame containing an intrusion target in a monitored area of ​​a railway system; the target recognition module is used to identify the intrusion target in the video frame based on an acoustic dispersion detection model, and determine the intrusion target type, the distance of the intrusion target, the moving speed and the existence time of the intrusion target; the decision module is used to match a target acoustic wave sample from a preset sound source database based on the intrusion target type, the distance of the intrusion target, the moving speed and the existence time of the intrusion target, and the current environmental conditions; the repelling module is used to emit a corresponding repelling acoustic wave to the intrusion target based on the target acoustic wave sample.

[0007] Furthermore, the acoustic dispersion detection model is a target detection model based on the Transformer architecture; wherein the acoustic dispersion detection model includes a dual-branch encoder-decoder structure.

[0008] Furthermore, the encoder of the acoustic dispersion detection model includes an upper branch encoding network and a lower branch encoding network; wherein the upper branch encoding network is used to input the image-rich features extracted by the ResNet backbone network into the Transformer encoder layer to further extract the global features of the image; the lower branch encoding network is used to obtain preliminary global depth features after passing through the ResNet backbone network and then through multiple multi-scale convolutional layers; the decoder of the acoustic dispersion detection model includes an upper branch decoding network and a lower branch decoding network; wherein the upper branch decoding network adopts a Transformer encoder structure; the upper branch decoding network is used to randomly initialize object queries and add position encoding, and extract features through a multi-head self-attention mechanism to obtain output features; the output features include target position features and category probabilities; the lower branch decoding network is used to perform local depth estimation optimization based on the target position features.

[0009] Furthermore, the acoustic dispersion detection model also includes a feature pyramid structure, a sub-pixel convolution layer and a fusion layer; wherein the feature pyramid structure is used to connect the output of the upper branch decoding network and the output of the lower branch decoding network, and use the sub-pixel convolution layer to generate local optimized deep features; the fusion layer is used to splice and fuse the output features and the local optimized deep features to obtain target category, target position and target distance information.

[0010] Furthermore, the decision module also includes: a frequency matching unit, an amplitude matching unit, a duration matching unit, a comprehensive matching unit and a decision unit; wherein the frequency matching unit is used to determine the expected sound wave frequency range based on the intrusion target type, and calculate the frequency difference between each sound wave sample in the preset sound source database and the expected sound wave frequency range; the amplitude matching unit is used to determine the expected sound wave amplitude based on the current environmental conditions and the distance of the intrusion target, and calculate the amplitude difference between each sound wave sample in the preset sound source database and the expected sound wave amplitude; the duration matching unit is used to determine the expected sound wave duration based on the moving speed and existence time of the intrusion target, and calculate the duration difference between each sound wave sample in the preset sound source database and the expected sound wave duration; the comprehensive matching unit is used to calculate the comprehensive matching degree of each sound wave sample in the preset sound source database based on the frequency difference, the amplitude difference, the duration difference and the weighted average method; the decision unit is used to match the target sound wave sample from the preset sound source database based on the comprehensive matching degree.

[0011] Furthermore, the expulsion module includes a sonic expulsion device; the sonic expulsion device includes a transmitter, a sonic wave generator, a control system, an energy source and a housing; wherein the sonic wave generator is used to generate expulsion sonic waves; the transmitter is used to propagate the expulsion sonic waves to a target area; the control system is used to control the operation of the transmitter and the sonic wave generator; and the energy source is used to provide electrical energy to the sonic wave expulsion device.

[0012] Furthermore, it also includes a device status monitoring module and a device interface module; wherein the device status monitoring module is used to detect the operating status of each device in the system; and the device interface module is used to provide different device interfaces.

[0013] On the other hand, a railway perimeter security method based on machine vision and an acoustic wave repelling device is provided, comprising: obtaining a key video frame containing an intrusion target in a monitored area of ​​a railway system; identifying the intrusion target in the video frame based on an acoustic dispersion detection model, and determining the intrusion target type, the distance of the intrusion target, the moving speed and the existence time of the intrusion target; matching a target sound wave sample from a preset sound source database based on the intrusion target type, the distance of the intrusion target, the moving speed and the existence time of the intrusion target, and current environmental conditions; and emitting a corresponding repelling sound wave to the intrusion target based on the target sound wave sample.

[0014] Furthermore, based on the intrusion target type, the distance of the intrusion target, the moving speed and existence time of the intrusion target, and the current environmental conditions, the target sound wave samples are matched from the preset sound source database, including: determining the expected sound wave frequency range based on the intrusion target type, and calculating the frequency difference between each sound wave sample in the preset sound source database and the expected sound wave frequency range; determining the expected sound wave amplitude based on the current environmental conditions and the distance of the intrusion target, and calculating the amplitude difference between each sound wave sample in the preset sound source database and the expected sound wave amplitude; determining the expected sound wave duration based on the moving speed and existence time of the intrusion target, and calculating the duration difference between each sound wave sample in the preset sound source database and the expected sound wave duration; calculating the comprehensive matching degree of each sound wave sample in the preset sound source database based on the frequency difference, the amplitude difference, the duration difference and the weighted average method; matching the target sound wave sample from the preset sound source database based on the comprehensive matching degree.

[0015] Furthermore, it also includes: collecting, organizing and storing different types of sound wave sample sets; the sound wave sample set includes a collection of sound wave samples for humans and animals, each sound wave sample includes a frequency parameter, an amplitude parameter, and a duration parameter; based on the sound wave sample set, constructing the preset sound source database.

[0016] The embodiment of the present invention provides a railway perimeter security system and method based on machine vision and acoustic wave repelling devices, which integrates traditional video surveillance technology, artificial intelligence technology, non-lethal acoustic wave repelling technology and sensor technology to achieve active dispersal of personnel and animals, thereby alleviating the technical problem that existing security technologies lack means for active intervention in sudden incidents and are unable to prevent and intervene in incidents when they occur. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 is a schematic diagram of a railway perimeter security system based on machine vision and acoustic wave repelling device provided by an embodiment of the present invention;

[0019] Figure 2 It is a general framework diagram of another railway perimeter security system based on machine vision and acoustic wave repelling device provided by an embodiment of the present invention;

[0020] Figure 3It is an architecture diagram of a railway perimeter security system based on machine vision and acoustic wave repelling device provided by an embodiment of the present invention;

[0021] Figure 4 is a flow chart of a railway perimeter security method based on machine vision and acoustic wave repelling device provided by an embodiment of the present invention;

[0022] Figure 5 It is a flow chart of another railway perimeter security method based on machine vision and acoustic wave repelling device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0024] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0025] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0026] Embodiment 1

[0027] Figure 1 FIG. 1 is a schematic diagram of a railway perimeter security system based on machine vision and acoustic wave repelling device according to an embodiment of the present invention. Figure 1 As shown, the system includes: a monitoring module 10 , a target recognition module 20 , a decision module 30 and a driving away module 40 .

[0028] Specifically, the monitoring module 10 is used to obtain key video frames containing intrusion targets in the area to be monitored in the railway system.

[0029] The target recognition module 20 is used to recognize the intrusion target in the video frame based on the acoustic dispersion detection model, and determine the intrusion target type, the distance of the intrusion target, the moving speed of the intrusion target and the existence time.

[0030] The decision module 30 is used to match the target sound wave sample from the preset sound source database based on the intrusion target type, the moving speed and existence time of the intrusion target, and the current environmental conditions.

[0031] The repelling module 40 is used to transmit corresponding repelling sound waves to the intruding target based on the target sound wave samples.

[0032] Preferably, the monitoring module 10 includes an image monitoring device, such as a video monitoring camera.

[0033] Specifically, the system communication monitoring module 10 provided in the embodiment of the present invention provides a real-time monitoring function, and the operator can view the video images of important areas around the railway in real time at the control center. Optionally, the monitoring module 10 has the characteristics of multi-screen segmentation and high-definition image transmission, ensuring that the operator can fully and clearly grasp the dynamic changes around the railway.

[0034] Optionally, the monitoring module 10 also has a camera function, which can take screenshots of specific areas as needed. This function helps to quickly capture abnormal situations and provide specific image support for subsequent evidence collection and analysis.

[0035] Optionally, the monitoring module 10 also has a recording and playback function. Specifically, the monitoring module 10 supports a recording function and can record monitoring images in real time. At the same time, operators can replay according to needs, quickly locate and find historical events, and improve the efficiency of post-analysis.

[0036] Preferably, the system provided by the embodiment of the present invention further includes an environmental sensor for obtaining current environmental conditions. Specifically, the environmental sensor includes a light sensor, a temperature sensor, a humidity sensor, etc., which are respectively used to obtain information such as light intensity, temperature, humidity, etc. of the current environment.

[0037] Specifically, the target recognition module 20 is further used for:

[0038] Perform image preprocessing on the video frames, including denoising, edge detection and other preprocessing operations on the images captured by the camera to improve the accuracy and stability of subsequent target recognition;

[0039] Then, the AI ​​recognition model Acoustic Dispersion Detection Model (ADDM) is used to classify the preprocessed images and determine whether the intruder target is a human or an animal.

[0040] Specifically, the acoustic dispersion detection model is a target detection model based on the Transformer architecture; wherein the acoustic dispersion detection model includes a dual-branch encoder-decoder structure.

[0041] Specifically, the encoder of the acoustic dispersion detection model includes an upper-branch encoding network and a lower-branch encoding network; the upper-branch encoding network is used to input the rich features of the image extracted by the ResNet backbone network into the Transformer encoder layer to further extract the global features of the image; the lower-branch encoding network is used to obtain preliminary global depth features after passing through the ResNet backbone network and then through multiple multi-scale convolutional layers.

[0042] Specifically, the decoder of the acoustic dispersion detection model includes an upper-branch decoding network and a lower-branch decoding network; the upper-branch decoding network adopts a Transformer encoder structure; the upper-branch decoding network is used to randomly initialize object queries and add position encoding, and extracts features through a multi-head self-attention mechanism to obtain output features; the output features include target position features and category probabilities; the lower-branch decoding network is used to perform local depth estimation optimization based on target position features.

[0043] Specifically, the acoustic dispersion detection model also includes a feature pyramid structure, a sub-pixel convolution layer and a fusion layer; wherein the feature pyramid structure is used to connect the output of the upper branch decoding network with the output of the lower branch decoding network, and use the sub-pixel convolution layer to generate local optimized deep features; the fusion layer is used to splice and fuse the output features and the local optimized deep features to obtain the target category, target position and target distance information.

[0044] In an embodiment of the present invention, the railway perimeter is monitored in real time by a camera. The monitored video images are transmitted to the AI ​​recognition model. The AI ​​recognition model uses the ADDM (Acoustic Dispersion Detection Model) algorithm, which is responsible for analyzing and identifying the monitored images. It can accurately detect foreign objects in the railway perimeter and identify the type and existence time of the target, including people or small animals. When the algorithm is deployed on the edge box, key information such as the target category, distance, and existence time can be displayed.

[0045] Specifically, ADDM is an object detection model based on the Transformer architecture. Its design idea is to transform the object detection task into an end-to-end conversion problem. The following is a detailed description of the ADDM architecture:

[0046] Encoder-Decoder structure:

[0047] ADDM adopts a dual-branch encoder-decoder structure. In the encoder stage, it can be divided into two branches, the upper branch is responsible for extracting the edge features of the input image, and the lower branch is responsible for extracting the image depth features. First, the ResNet backbone network is used to extract the rich features of the input image. Secondly, the upper branch network passes through ResNet and then enters the Transformer encoder layer to further extract the global features of the image. The Transformer encoder is not used at the beginning because the Transformer has a huge amount of computation. After feature extraction, the image size can be reduced, the Transformer's computation can be reduced, and the inference speed can be improved. The Transformer encoder uses positional encoding technology to divide the image features into multiple patches, and then uses the multi-head self-attention mechanism to extract the global features of the patches. For the lower branch of the encoder, the lower branch network passes through ResNet and then passes through multiple multi-scale convolutional layers to extract preliminary global depth features.

[0048] Since the encoding upper branch adopts the Transformer encoder structure, the corresponding decoder upper branch also uses the Transformer structure. First, the object query is randomly initialized (position encoding is added), and then the multi-head self-attention mechanism is used, which is understood as first dividing different areas and finding the relationship between different areas. Then the self-attention results are queries, and the encoder output results are keys and values ​​for feature extraction. The output of the last layer of the decoder is divided into two parts: one part is used to predict the category probability of the target, and the other part is used to predict the position of the target (usually represented by the coordinates of the bounding box). The generated target position and category prediction will go through a series of post-processing steps (such as non-maximum suppression) to obtain the final detection result. Since the encoding lower branch does not adopt the Transformer structure, its corresponding decoder lower branch is a local depth optimization function, which uses the target position features extracted by the decoding upper branch to perform local depth estimation optimization. Therefore, when designing the algorithm, the features of the upper decoding branch are output to the lower branch so that the lower encoding branch can further focus on the local depth features to be optimized based on the preliminary global depth features. ADDM uses the feature pyramid (FPN) structure to connect the output of the upper decoding branch and the output of the lower encoding branch, and uses the Pixel-Shuffle Layer and convolution operations to generate local optimized depth features. The Pixel-Shuffle Layer uses the learned feature map to generate an upsampled feature map by shuffling pixels from the channel dimension to the spatial dimension. Using the Pixel-Shuffle Layer instead of the transposed convolution can avoid expensive computational costs. Generate a local fine depth estimate. Finally, the position and category prediction features of the target output by the upper decoding branch and the depth features output by the lower decoding branch are spliced ​​and fused through a fusion layer to obtain the category, position, and distance information of the target. In addition, the algorithm obtains the time when the target appears by detecting the duration of the target appearing in the window. Therefore, the ADDM algorithm can output the category, position, distance, and existence time information of the target.

[0049] The design of the ADDM architecture enables end-to-end object detection without relying on manually designed features or region proposals, thereby simplifying the object detection process and achieving satisfactory performance on some standard datasets.

[0050] Once the AI ​​recognition model confirms the target type, location distance and other information in the monitoring area, the system will immediately select the most suitable sound wave from the pre-built sound source database according to the target type and location distance for expulsion. The sound source database carefully organizes the frequency, amplitude and expulsion effect of various sound waves, and associates them with the target type (such as people or small animals) and distance information.

[0051] For example, when the recognition model determines that the target is a human, the system will automatically select the sound wave that is most suitable for affecting humans for repelling; conversely, if the target is a small animal, it will select the sound wave that has the best effect on it. At the same time, according to the distance, the closer the distance, according to the properties of the power function, the greater the amplitude of the sound wave, and vice versa. This intelligent sound wave selection process aims to maximize the repelling effect and reduce the waste of system resources.

[0052] Specifically, the decision-making module 30 further includes: a frequency matching unit 31, an amplitude matching unit 32, a duration matching unit 33, a comprehensive matching unit 34, and a decision-making unit 35.

[0053] Specifically, the frequency matching unit 31 is used to determine the expected sound wave frequency range based on the type of intrusion target, and calculate the frequency difference between each sound wave sample in the preset sound source database and the expected sound wave frequency range.

[0054] Specifically, determining the expected sound wave frequency range based on the type of intrusion target. For example, for humans, the expected sound wave frequency range is determined to be [X1, X2] Hz; for a certain animal, the expected sound wave frequency range is determined to be [Y1, Y2] Hz. Taking the above-mentioned expected sound wave frequency range [X1, X2] for humans as an example, the method for calculating the frequency difference is illustrated as follows:

[0055] For the frequency f of a sound wave sample in the preset sound source database sample , if f sample < X1, then the frequency difference is △F = X1 - f sample ; if f sample > X2, then the frequency difference is △F = f sample - X2; if X1 ≤ f sample ≤ X2, then the frequency difference is △F = 0.

[0056] The amplitude matching unit 32 is used to determine the expected sound wave amplitude based on the current environmental conditions and the distance of the intrusion target, and calculate the amplitude difference between each sound wave sample in the preset sound source database and the expected sound wave amplitude.

[0057] Specifically, the amplitude difference is:

[0058]

[0059] where A target is the expected sound wave amplitude, and A sample is the amplitude of a sound wave sample in the preset sound source database.

[0060] The duration matching unit 33 is used to determine the expected sound wave duration based on the moving speed and existence time of the intrusion target, and calculate the duration difference between each sound wave sample in the preset sound source database and the expected sound wave duration.

[0061] Specifically, the duration difference is:

[0062]

[0063] Among them, T target is the expected duration of the sound wave, T sample The duration of a sound wave sample in the preset sound source database.

[0064] The comprehensive matching unit 34 is used to calculate the comprehensive matching degree of each sound wave sample in the preset sound source database based on the frequency difference, amplitude difference, duration difference and weighted average method.

[0065] Specifically, the comprehensive matching degree is:

[0066]

[0067] Where M represents the comprehensive matching degree, w1, w2, and w3 are the frequency weight coefficient, amplitude weight coefficient, and duration weight coefficient, respectively. To ensure that the comprehensive matching degree has no unit, the unit of w1 is Hz. -1 , the unit of w2 is m -1 , the unit of w3 is s -1 .

[0068] The decision unit 35 is used to match the target sound wave sample from the preset sound source database based on the comprehensive matching degree.

[0069] Preferably, the sound wave sample with the smallest comprehensive matching degree M is selected as the target sound wave sample.

[0070] Specifically, the driving module 40 includes a sonic driving device.

[0071] The principle of sonic repelling technology is to use sound waves of a specific frequency to match or approach the natural frequency of certain organs or tissues of humans or animals, thereby inducing a resonance effect. This resonance can cause displacement, deformation, or even rupture of organs, making people or animals feel uncomfortable, and then forcing them to leave the area to achieve the purpose of protecting the target area. Using the principle of sound wave resonance, infrasound devices use infrasound waves with a frequency close to the natural frequency of human organs to resonate with human organs. When the frequency of infrasound waves is in the range of 4 to 18 Hz, it will resonate with the human nervous system or certain organs. Mild resonance can cause discomfort, decreased concentration, and emotional distress. In severe cases, it may cause mental confusion, breathing difficulties, and even life-threatening.

[0072] The acoustic wave drive device, the whole shell is made of high-strength aluminum alloy material, with comprehensive functions and high stability. Specifically, the acoustic wave drive device includes a transmitter, an acoustic wave generator, a control system, an energy source and a shell;

[0073] The sound wave generator is used to generate the repelling sound waves. The sound wave generator is a device used to generate sound waves, and usually uses components such as piezoelectric ceramics or sound wave oscillators. The sound wave generator generates sound waves by exciting the piezoelectric ceramics or oscillators to generate high-frequency mechanical vibrations. The frequency and amplitude of the sound waves can be adjusted as needed to adapt to the repelling effect in different scenarios.

[0074] The transmitter is used to transmit the repelling sound waves to the target area. The transmitter is the core component of the acoustic repelling device and is responsible for transmitting high-intensity sound waves. It usually uses a speaker or other sound emitting device to effectively transmit the sound waves to the target area. The transmitter has a certain power and frequency range, which can be adjusted and controlled according to actual needs.

[0075] The control system is used to control the operation of the transmitter and the sound wave generator. The control system is the intelligent core of the sound wave repellent device and is responsible for controlling the operation of the transmitter and the sound wave generator. The control system usually uses a microprocessor or an embedded system and has the functions of automatic control and intelligent adjustment. It can automatically adjust the frequency, amplitude and emission direction of the sound wave according to real-time monitoring data and preset parameters to achieve the best repellent effect.

[0076] The energy source is used to provide electrical energy for the acoustic wave dispersing device. The acoustic wave dispersing device requires a stable energy supply to ensure its normal operation. The energy source usually uses a battery, an AC power supply or a solar cell to provide the required electrical energy. For an acoustic wave dispersing device that needs to work continuously for a long time, a stable and reliable energy source is essential.

[0077] Protective housing: Sonic repellent devices are usually installed in outdoor environments, so it is very important to have good waterproof, dustproof and corrosion resistance. The protective housing can protect the electronic components inside the device from the harsh environment and ensure the long-term stable operation of the device.

[0078] Through the synergistic effect of the above components, the acoustic repellent device can effectively emit high-frequency sound waves and produce harsh sounds, thereby achieving the purpose of repelling intruders around the railway. Its intelligent control system and adjustable parameter design enable the device to be flexibly adjusted and optimized according to actual conditions, thereby improving the repellent effect and applicability.

[0079] Figure 2 is a general framework diagram of another railway perimeter security system based on machine vision and acoustic wave repelling device provided according to an embodiment of the present invention, Figure 3 1 is an architecture diagram of a railway perimeter security system based on machine vision and acoustic wave repelling devices provided according to an embodiment of the present invention.

[0080] like Figure 2 and Figure 3 As shown, the system provided by the embodiment of the present invention also includes a sound wave expulsion strategy library. Specifically, the key component of the system provided by the embodiment of the present invention is the sound wave expulsion strategy library, which is responsible for selecting appropriate sound waves for expulsion according to the identified intrusion target type and environmental conditions. The design of the sound wave expulsion strategy library includes two main aspects, namely, an AI recognition model and a sound source database, and a decision layer.

[0081] The AI ​​recognition model includes an acoustic dispersion detection model; the decision layer is used to select appropriate sound waves from the sound source database for expulsion based on the target type and scenario requirements. Specifically, it includes:

[0082] Sound source database construction: collect, organize and store different types of sound wave samples, including sounds for humans and animals. Each sound wave sample should contain relevant parameters such as frequency, amplitude, duration, etc.

[0083] Sound wave matching: Based on the target type recognition results and current environmental conditions, the best matching sound wave sample is selected from the sound source database. The matching needs to consider the sound wave's frequency, amplitude, duration and other parameters.

[0084] Comprehensive decision-making: The results of the target type recognition algorithm and the sound source database matching algorithm are combined to formulate the final acoustic repellent strategy. The strategy should take into account factors such as the type, distance, direction of the intruder target, as well as current environmental conditions such as noise level and wind direction.

[0085] In the sound wave repelling strategy library, a complete sound source database has been established, which contains data on the impact of various sound wave frequencies and amplitudes on humans and animals. Based on the results of the AI ​​recognition model, the system dynamically selects the appropriate sound source for repelling. For example, when confirming that there is a human intrusion in the monitoring screen, the system will use one sound wave, and when an animal intrusion is detected, another sound wave will be used. This process can switch the sound source in real time to ensure that the sound wave repelling effect is more accurate and efficient. In order to maintain the timeliness of the system, the sound source database will be updated regularly, integrating new research results and data to ensure that the sound source information in the database is always kept up to date.

[0086] Specifically, Figure 3 As shown, the system provided by the embodiment of the present invention also includes a device status monitoring module and a device interface module; wherein,

[0087] Equipment status monitoring module, used to detect the operating status of each device in the system;

[0088] The device interface module is used to provide different device interfaces.

[0089] Specifically, the equipment status monitoring module provides status monitoring of the acoustic wave repellent device and other key equipment. Operators can view the equipment operating status in real time, including power status, communication status, etc., to ensure the normal operation of the system.

[0090] Optionally, the equipment status monitoring module also supports equipment policy control. Operators can flexibly adjust the working mode, sound wave frequency, etc. of the sonic repellent device according to actual conditions to meet the security needs of different scenarios.

[0091] Optionally, the equipment status monitoring module also integrates the alarm management function, which can monitor equipment abnormalities in real time and issue alarms. Operators in the control center can quickly respond to alarm information and take corresponding measures to ensure the stability and reliability of the system.

[0092] Specifically, the device interface module provides a surveillance camera interface and supports integration with various surveillance cameras. This enables the electronic sentry system to be compatible with multiple camera models and achieve a wider range of surveillance coverage.

[0093] Specifically, the device interface module also provides an interface for the acoustic wave drive device. In order to achieve flexible control of the acoustic wave drive device, the system has designed a special interface. Through this interface, the operator can adjust the frequency, amplitude and other parameters of the acoustic wave drive to adapt to different environments and scenarios.

[0094] Specifically, the device interface module also supports an infrared detection interface, which can be connected to infrared sensors and other devices. Such an interface design enables the system to effectively monitor intrusion behavior in low-light or no-light environments, improving the all-weather performance of the system.

[0095] From the above description, it can be seen that the embodiment of the present invention provides a railway perimeter security system based on machine vision and acoustic wave repelling device, which uses sound waves of specific frequencies to cause temporary psychological or physiological discomfort to people, thereby achieving the effect of repelling. The application of acoustic wave repelling technology in key places of railways can enhance the defense capability of existing security systems and realize dynamic protection of key areas, which has the following advantages:

[0096] 1) Reduce illegal intrusion: By creating an invisible protective barrier through sound waves, intruders can be quickly driven away without being noticed by the intruders, thus reducing the incidence of security accidents.

[0097] 2) Enhanced monitoring efficiency: Combined with traditional monitoring systems, acoustic wave technology can automatically start when abnormal behavior is detected, improving the response speed and efficiency of the entire security system.

[0098] 3) Reduce maintenance costs and safety risks: Acoustic wave equipment generally requires lower maintenance costs, and due to its non-contact nature, it reduces the possibility of human confrontation and improves the work safety of security personnel.

[0099] By studying and implementing an electronic sentry system based on acoustic repelling technology, the present invention can significantly improve the safety management level of key railway sites and modernize the protection system, which is in line with the current trend of high-tech security protection. In addition, this research may also provide feasible technical solutions and strategies for other similar public safety fields, with broad application prospects and far-reaching social impact.

[0100] Embodiment 2

[0101] Figure 4 1 is a flow chart of a railway perimeter security method based on machine vision and acoustic wave repelling device provided according to an embodiment of the present invention. Figure 4 As shown, the method specifically comprises the following steps:

[0102] Step S402, obtaining a key video frame containing an intrusion target in the area to be monitored of the railway system;

[0103] Step S404, identifying the intrusion target in the video frame based on the acoustic dispersion detection model, and determining the intrusion target type, the distance of the intrusion target, the moving speed of the intrusion target, and the existence time;

[0104] Step S406, matching the target sound wave sample from a preset sound source database based on the intrusion target type, the distance of the intrusion target, the moving speed and existence time of the intrusion target, and the current environmental conditions;

[0105] Step S408: transmitting corresponding repelling sound waves to the intruding target based on the target sound wave samples.

[0106] Specifically, step S406 further includes the following steps:

[0107] Step S4061, determining an expected sound wave frequency range based on the intrusion target type, and calculating a frequency difference between each sound wave sample in a preset sound source database and the expected sound wave frequency range.

[0108] Specifically, the expected sound wave frequency range is determined based on the intrusion target type. For example, for humans, the expected sound wave frequency range is determined to be [X1, X2] Hz; for certain animals, the expected sound wave frequency range is determined to be [Y1, Y2] Hz. Taking the above expected sound wave frequency range [X1, X2] for humans as an example, the frequency difference calculation method is illustrated as follows:

[0109] For a frequency f of a sound wave sample in the preset sound source database sample , if fsample <If f < X1, the frequency difference is ΔF = X1 - f sample ; if f sample > X2, the frequency difference is ΔF = f sample - X2; if X1 ≤ f sample ≤ X2, the frequency difference is ΔF = 0.

[0110] Step S4062: Determine the expected sound wave amplitude based on the current environmental conditions and the distance to the intrusion target, and calculate the amplitude difference between each sound wave sample in the preset sound source database and the expected sound wave amplitude.

[0111] Specifically, the amplitude difference is:

[0112]

[0113] where A target is the expected sound wave amplitude, and A sample is the amplitude of a sound wave sample in the preset sound source database.

[0114] Step S4063: Determine the expected sound wave duration based on the moving speed and presence time of the intrusion target, and calculate the duration difference between each sound wave sample in the preset sound source database and the expected sound wave duration.

[0115] Specifically, the duration difference is:

[0116]

[0117] where T target is the expected sound wave duration, and T sample is the duration of a sound wave sample in the preset sound source database.

[0118] Step S4064: Calculate the comprehensive matching degree of each sound wave sample in the preset sound source database based on the frequency difference, amplitude difference, duration difference, and weighted average method.

[0119] Specifically, the comprehensive matching degree is:

[0120]

[0121] where M represents the comprehensive matching degree, and w1, w2, and w3 are the frequency weight coefficient, amplitude weight coefficient, and duration weight coefficient, respectively. Among them, to ensure that the comprehensive matching degree has no unit, the unit of w1 is Hz -1 , the unit of w2 is m -1 , and the unit of w3 is s -1 .

[0122] Step S4065, based on the comprehensive matching degree, matching the target sound wave sample from the preset sound source database.

[0123] Preferably, the sound wave sample with the smallest comprehensive matching degree M is selected as the target sound wave sample.

[0124] Specifically, the method provided by an embodiment of the present invention also includes: collecting, organizing and storing different types of sound wave sample sets; the sound wave sample set includes a collection of sound wave samples for humans and animals, each sound wave sample includes a frequency parameter, an amplitude parameter, and a duration parameter; based on the sound wave sample set, a preset sound source database is constructed.

[0125] Figure 5 FIG. 1 is a flow chart of another railway perimeter security method based on machine vision and acoustic wave repelling device provided according to an embodiment of the present invention. Figure 5 As shown, after the sound wave expulsion strategy library selects the corresponding sound source from the sound source database according to the target type, it first selects a more stimulating sound source at a medium distance and with increasing decibels to issue a warning command and judge the target behavior. If the target behavior is under control, the time is over; if the target behavior is not under control, the most effective stimulating sound source is selected at a close distance and with the highest decibel value to provide auditory stimulation and interference to the target and expel it to the maximum extent.

[0126] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0127] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0128] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0130] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0131] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0132] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0133] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0134] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A railway perimeter security system based on machine vision and acoustic wave repelling device, characterized in that: include: Monitoring module, target recognition module, decision module and expulsion module; among them, The monitoring module is used to obtain key video frames containing intrusion targets in the monitored area of ​​the railway system; The target recognition module is used to identify the intrusion target in the video frame based on the acoustic dispersion detection model, and determine the intrusion target type, the distance of the intrusion target, the moving speed and existence time of the intrusion target; The decision module is used to match the target sound wave sample from the preset sound source database based on the type of the intrusion target, the distance of the intrusion target, the moving speed and existence time of the intrusion target, and the current environmental conditions; The repelling module is used to transmit corresponding repelling sound waves to the intruding target based on the target sound wave sample.

2. The system according to claim 1, characterized in that The acoustic dispersion detection model is a target detection model based on the Transformer architecture; wherein the acoustic dispersion detection model includes a dual-branch encoder-decoder structure.

3. The system according to claim 2, characterized in that The encoder of the acoustic dispersion detection model includes an upper branch encoding network and a lower branch encoding network; wherein, The upper branch encoding network is used to input the rich image features extracted by the ResNet backbone network into the Transformer encoder layer to further extract the global features of the image; The lower branch encoding network is used to obtain preliminary global depth features after passing through the ResNet backbone network and then through multiple multi-scale convolutional layers; The decoder of the acoustic dispersion detection model includes an upper branch decoding network and a lower branch decoding network; wherein the upper branch decoding network adopts a Transformer encoder structure; The upper branch decoding network is used to randomly initialize object queries and add position encoding, and extract features through a multi-head self-attention mechanism to obtain output features; the output features include target position features and category probabilities; The lower branch decoding network is used to perform local depth estimation optimization based on the target position features.

4. The system according to claim 3, characterized in that The acoustic dispersion detection model also includes a feature pyramid structure, a sub-pixel convolution layer and a fusion layer; wherein, The feature pyramid structure is used to connect the output of the upper branch decoding network and the output of the lower branch decoding network, and generate local optimized depth features using the sub-pixel convolution layer; The fusion layer is used to splice and fuse the output features and the local optimized depth features to obtain target category, target position and target distance information.

5. The system according to claim 1, characterized in that The decision module further includes: a frequency matching unit, an amplitude matching unit, a duration matching unit, a comprehensive matching unit and a decision unit; wherein, The frequency matching unit is used to determine the expected sound wave frequency range based on the intrusion target type, and calculate the frequency difference between each sound wave sample in the preset sound source database and the expected sound wave frequency range; The amplitude matching unit is used to determine the expected sound wave amplitude based on the current environmental conditions and the distance of the intrusion target, and calculate the amplitude difference between each sound wave sample in the preset sound source database and the expected sound wave amplitude; The duration matching unit is used to determine the expected sound wave duration based on the moving speed and existence time of the intrusion target, and calculate the duration difference between each sound wave sample in the preset sound source database and the expected sound wave duration; The comprehensive matching unit is used to calculate the comprehensive matching degree of each sound wave sample in the preset sound source database based on the frequency difference, the amplitude difference, the duration difference and the weighted average method; The decision unit is used to match the target sound wave sample from the preset sound source database based on the comprehensive matching degree.

6. The system according to claim 1, characterized in that The expulsion module includes a sonic expulsion device; the sonic expulsion device includes a transmitter, a sonic generator, a control system, an energy source and a housing; wherein, The sound wave generator is used to generate a driving sound wave; The transmitter is used to transmit the repelling sound waves to the target area; The control system is used to control the operation of the transmitter and the sound wave generator; The energy source is used to provide electrical energy to the sound wave driving device.

7. The system according to claim 1, characterized in that It also includes a device status monitoring module and a device interface module; wherein, The device status monitoring module is used to detect the operating status of each device in the system; The device interface module is used to provide different device interfaces.

8. A railway perimeter security method based on machine vision and acoustic wave repelling device, characterized in that: include: Acquire key video frames containing intrusion targets in the area to be monitored of the railway system; Based on the acoustic dispersion detection model, target recognition is performed on the intrusion target in the video frame to determine the type of the intrusion target, the distance of the intrusion target, the moving speed of the intrusion target and the existence time; Matching target sound wave samples from a preset sound source database based on the type of the intrusion target, the distance of the intrusion target, the moving speed and existence time of the intrusion target, and the current environmental conditions; Based on the target sound wave sample, a corresponding repelling sound wave is emitted toward the intruder target.

9. The method according to claim 8, characterized in that Based on the type of the intrusion target, the distance of the intrusion target, the moving speed and existence time of the intrusion target, and the current environmental conditions, the target sound wave sample is matched from a preset sound source database, including: Determine an expected sound wave frequency range based on the intrusion target type, and calculate the frequency difference between each sound wave sample in the preset sound source database and the expected sound wave frequency range; Determine the expected sound wave amplitude based on the current environmental conditions and the distance of the intrusion target, and calculate the amplitude difference between each sound wave sample in the preset sound source database and the expected sound wave amplitude; Determine the expected sound wave duration based on the moving speed and existence time of the intrusion target, and calculate the duration difference between each sound wave sample in the preset sound source database and the expected sound wave duration; Calculate the comprehensive matching degree of each sound wave sample in the preset sound source database based on the frequency difference, the amplitude difference, the duration difference and the weighted average method; Based on the comprehensive matching degree, the target sound wave sample is matched from the preset sound source database.

10. The method according to claim 8, characterized in that Also includes: Collect, organize and store different types of sound wave sample sets; the sound wave sample set includes a collection of sound wave samples for humans and animals, and each sound wave sample includes a frequency parameter, an amplitude parameter, and a duration parameter; Based on the sound wave sample set, the preset sound source database is constructed.

Images