Security radar false alarm management method and system based on traction learning

Abstract

The invention relates to the technical field of communication, in particular to a security radar false alarm management method and system based on traction learning, and the method comprises the steps: obtaining the position information and related feature information of a target through a radar, and transmitting the position information to a dome camera; the ball machine performs category identification on the target detection object based on the target position information and extracts an identification result; judging according to system time, and extracting a time judgment result; performing environment information judgment according to the time judgment result, and extracting an environment judgment result; and feeding back an identification result to the radar according to an environment judgment result, training a radar identification model in combination with the feature information, and updating identification model parameters. The objective of the invention is to solve the technical problem of high false alarm rate of a security monitoring system.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a false alarm management method and system for security radar based on traction learning. Background Art

[0002] In modern security monitoring systems, the linkage monitoring of radar and high-speed dome cameras (referred to as dome cameras) is a common technical means. The linkage mechanism follows the following workflow: radar, as a front-end detection device, is responsible for preliminary scanning of the target area and detecting potential targets; once a target is found, the radar transmits the target's location information to the dome camera; then, the dome camera adjusts the viewing angle according to the received location information, and uses its high-resolution camera to conduct secondary confirmation and detailed observation of the target; finally, through comprehensive analysis of the data provided by the radar and the dome camera, the target attributes are determined and reported.

[0003] Although this approach has proven effective in many scenarios, it still has certain limitations. Current methods do not fully utilize the ball camera's capabilities in advanced visual recognition to optimize the performance of the radar system. The recognition models used by radar equipment mainly rely on pre-collected data sets for training, and these data sets often fail to fully cover all possible situations in practical applications. Therefore, when faced with unknown or atypical scenarios, radars may have a high false alarm rate. On the other hand, when the radar frequently sends incorrect target information to the ball camera, even if the latter has a strong classification accuracy, the probability of false alarms will increase due to the increase in the number of false targets and the processing of a large number of invalid signals. Radar models that rely on fixed data sets for training are difficult to adapt to complex and changing actual environments, limiting the flexibility and accuracy of the overall monitoring system. Summary of the invention

[0004] In order to solve the technical problem of high false alarm rate of security monitoring system, the purpose of the present invention is to provide a false alarm management method and system for security radar based on traction learning. The technical scheme adopted is as follows: The technical solution of the first aspect of the present invention provides a security radar false alarm management method based on traction learning, the method comprising: Use radar to obtain the target's location information and related feature information, and transmit the location information to the ball camera; The ball camera identifies the category of the target object based on the target position information and extracts the identification result; Make judgment based on system time and extract time judgment result; Perform environmental information judgment based on the time judgment result and extract the environmental judgment result; The recognition results are fed back to the radar based on the environmental judgment results and the radar recognition model is trained in combination with the feature information to update the recognition model parameters.

[0005] Furthermore, the relevant characteristic information of the target obtained by radar includes: Use radar to obtain the electromagnetic scattering characteristics and target motion characteristics of the target.

[0006] Furthermore, obtaining target-related characteristic information by radar also includes: The radar obtains the target's electromagnetic scattering characteristic information and target motion characteristic information and stores them in the characteristic information structure.

[0007] Furthermore, the radar acquires the target's electromagnetic scattering characteristic information including: Extracting the echo peak energy from the target echo signal; Collect the energy of several noise samples in the background area around the target; The signal-to-noise ratio is calculated as the ratio of the peak energy of the echo to the average energy of several noise samples.

[0008] Furthermore, the target motion characteristic information acquired by the radar includes: Extract speed fluctuation information, distance fluctuation information, and angle fluctuation information based on the target speed, distance, and angle information of the current detection frame and the previous detection frame of the radar; The micro-Doppler frequency shift information is extracted from the radar echo signal using short-time Fourier transform.

[0009] Furthermore, according to the system time, the extraction time determination result includes: Get the current system time, determine whether the current time is day or night, and output the time determination result.

[0010] Further, environmental information is judged according to the time judgment result, and extracting the environmental judgment result includes: If the time determination result is night, the recognition result will not be fed back to the radar and the relevant feature information will be deleted; If the time determination result is daytime, determine whether the external environment is sunny at the current moment and output the environment determination result.

[0011] Furthermore, the recognition result is fed back to the radar according to the environmental determination result and the radar recognition model is trained in combination with the feature information. The updating of the recognition model parameters includes: If the environment is judged as not clear, the recognition result will not be fed back to the radar and the relevant feature information will be deleted; If the environment is judged to be clear, the recognition result will be fed back to the radar, and the recognition result and related feature information will be used as training samples to train the radar recognition model. After the training is completed, the related feature information will be deleted.

[0012] Furthermore, using the recognition results and related feature information as training samples to train the radar recognition model includes: Combining relevant feature information into a feature vector, and using the recognition result as a true category label, wherein the relevant feature information at least includes electromagnetic scattering feature information, speed fluctuation information, distance fluctuation information, angle fluctuation information, and micro-Doppler frequency shift information of the target; The radar recognition model is constructed and trained based on the neural network, and the loss function is expressed as:

[0013] In the formula, represents the loss function; Indicates the total number of target categories; Indicates The true label of the class target; express The predicted probability of the class target; represents the first elements; Represents the first feature vector reconstructed by the autoencoder elements; Represents the preset weight matrix; represents the regularization term coefficient; represents the coefficient of the reconstruction error term; Represents the number of elements in the feature vector.

[0014] The technical solution of the second aspect of the present invention provides a security radar false alarm management system based on traction learning, using the security radar false alarm management method based on traction learning provided by the technical solution of the first aspect of the present invention, the system includes: A radar configured to obtain location information and related characteristic information of a target and transmit the location information to a ball camera; The ball camera is configured to identify the category of the target detection object based on the target position information and extract the identification result; A time determination module, configured to make a determination based on the system time and extract a time determination result; An environment determination module is configured to perform environment information determination according to the time determination result and extract the environment determination result; The traction learning module is configured to feed back the recognition results to the radar according to the environmental judgment results and train the radar recognition model in combination with the feature information to update the recognition model parameters.

[0015] The present invention has the following beneficial effects: The security radar false alarm management method and system based on traction learning provided by the present invention utilize the advantages of ball cameras in the field of target recognition and classification. The advantages of ball cameras in the field of image recognition can make up for the problem of insufficient recognition accuracy of radar in some cases, and feedback is given to the results reported by the radar. The radar uses the feedback results to train and learn itself, that is, the radar is pulled by the ball camera to learn, rather than pre-collecting a large amount of data to train the radar. The radar can identify real targets and false targets by continuously receiving feedback and learning, effectively reducing the number of false targets reported by the radar, thereby reducing the number of false targets reported by the overall system and reducing the false alarm rate of the security monitoring system. On the other hand, the radar can perform specific learning for the area that needs to be monitored, which not only reduces the demand for large-scale pre-training data sets, but also enables the radar to learn for the scenes that need to be monitored, thereby improving adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. 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 paying creative work.

[0017] Figure 1 A method flow chart of a security radar false alarm management method based on traction learning provided by one embodiment of the present invention; Figure 2 An algorithm flow chart of a security radar false alarm management method based on traction learning provided by one embodiment of the present invention; Figure 3 A schematic diagram of the structure of a security radar false alarm management system based on traction learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the security radar false alarm management method and system based on traction learning proposed by the present invention, its specific implementation method, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0020] The specific scheme of a security radar false alarm management method and system based on traction learning provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0021] See also Figure 1 and Figure 2 , which shows a method flow chart of a security radar false alarm management method based on traction learning provided by an embodiment of the present invention, the method comprising: Step S100: using the radar to obtain the position information and related feature information of the target, and transmitting the position information to the ball camera; specifically, using the radar to obtain the electromagnetic scattering characteristics and target motion characteristics of the target and store them in a feature information structure; the radar transmits the obtained target position information to the ball camera so that the ball camera can aim at and shoot the target; Step S100 specifically includes: Step S110: using radar to obtain electromagnetic scattering characteristic information of the target, where the electromagnetic scattering characteristic refers to the echo scattering cross-sectional area (RCS) of the target, which is indirectly reflected as the echo signal-to-noise ratio of the target, including: Extracting the echo peak energy from the target echo signal; Collect the energy of several noise samples in the background area around the target; The signal-to-noise ratio is calculated based on the ratio of the echo peak energy to the average energy of several noise samples, which can be expressed as:

[0022] In the formula, represents the signal-to-noise ratio; Indicates the peak energy of the target echo signal; The background area around the target The energy of noise samples; Represents the number of acoustic sample energies, preferably an integer greater than 6.

[0023] Step S120: The radar acquires target motion characteristic information of the target, including: According to the target speed, distance, and angle information of the current radar detection frame and the previous detection frame, the speed fluctuation information, distance fluctuation information, and angle fluctuation information are extracted, where: The speed fluctuation information is calculated as follows:

[0024] The distance fluctuation information is calculated as follows:

[0025] The angle fluctuation information is calculated as follows:

[0026] In the formula, Indicates the current detection frame. Indicates the last detection frame; Indicates speed fluctuation information; Indicates distance fluctuation information; Indicates angle fluctuation information; The micro-Doppler frequency shift information is extracted from the radar echo information signal using short-time Fourier transform. Short-time Fourier transform can convert the time domain signal into the time-frequency domain, thereby extracting the information of frequency variation over time, which is helpful to identify the subtle motion characteristics of the target. The feature information obtained above is combined into a feature vector and stored in the feature information structure. The feature vector can be expressed as: , represents micro-Doppler frequency shift information; in this embodiment, by combining electromagnetic scattering feature information and target motion feature information, the radar can more accurately describe the characteristics and behavior patterns of the target, thereby improving the accuracy of target recognition; wherein, the calculation of the signal-to-noise ratio (SNR) helps to distinguish between real targets and background noise, and reduce false alarms caused by noise; the speed fluctuation information, the distance fluctuation information, and the angle fluctuation information provide the dynamic characteristics of the target, which helps to filter out stationary or slowly moving false targets; the micro-Doppler frequency shift information reveals the motion characteristics of the internal structure or part of the target, which helps to identify targets with specific internal motion patterns, and further improves the robustness of the system.

[0027] Step S200: The ball camera identifies the category of the target object based on the target position information and extracts the identification result. Specifically, the ball camera (usually a pan / tilt camera with a high-resolution camera) performs image recognition on the target. The ball camera indexes the target according to the target position provided by the radar and obtains the image information of the target. It uses the image processing algorithm to identify and extract the features of the target and inputs them into a pre-trained classification model to classify the target, and outputs the category result of the target, such as people, cars, animals, etc.

[0028] Step S300: make a judgment according to the system time and extract the time judgment result; specifically, the current time information can be obtained from the system clock or the network time protocol (NTP) server, the time information can be parsed, and the current date and specific time (hour, minute, second) can be determined; according to the preset time period (for example, 6 pm to 6 am is night, and other times are day), it is judged whether the current time is night or day; if the time judgment result is night, the recognition result is not fed back to the radar, and the relevant feature information is deleted from the feature information structure; if the time judgment result is day, it is judged whether the external environment is clear at the current moment, and the environment judgment result is output; due to the poor lighting conditions during the night period, the accuracy of image recognition will be greatly reduced, therefore, not feeding back the recognition result at night can avoid introducing a large amount of low-quality data and reduce the false alarm rate; by deleting the feature information during the night period, the storage space can be effectively saved and the resource utilization of the system can be improved; this embodiment only feeds back the recognition result when the daytime and the environment is clear, so as to ensure that the data used for radar recognition model training has high quality and reliability, thereby improving the recognition accuracy of the model.

[0029] Step S400: Perform environmental information judgment based on the time judgment result and extract the environmental judgment result; specifically, environmental sensors (such as light sensors, weather sensors, etc.) can be used to collect current environmental data, including light intensity and weather conditions (sunny, cloudy, rainy, etc.); based on the data of the light sensor, judge whether the current light intensity reaches a preset threshold (for example, greater than a certain brightness value indicates sunny); combined with the data of the weather sensor, judge whether there is currently adverse weather conditions such as rainfall; comprehensively consider the light intensity and weather conditions, and output the environmental judgment result, for example, if the light intensity is high and there is no adverse weather condition, it is judged to be sunny; otherwise, it is judged to be If the environment is determined to be not clear, the recognition result will not be fed back to the radar, and the relevant feature information will be deleted from the feature information structure. If the environment is determined to be clear, the recognition result will be fed back to the radar, and the radar recognition model will be trained with the recognition result and the relevant feature information as training samples, and the relevant feature information will be deleted after the training is completed. In this embodiment, the recognition result is only fed back in a clear environment to ensure the quality of the data used for radar recognition model training, and further improve the training effect and recognition accuracy of the model. Under non-clear conditions, the radar system will not use low-quality data for training, thereby reducing the possibility of radar false alarms.

[0030] Step S500: Feedback the recognition result to the radar according to the environmental determination result and train the radar recognition model in combination with the feature information to update the recognition model parameters; the radar inputs the recognition result of the target and the corresponding relevant feature information into the radar recognition model for training and learning, and deletes the relevant feature information from the feature information structure after the training is completed; Specifically, the relevant feature information is combined into a feature vector, and the relevant feature information at least includes the electromagnetic scattering feature information, speed fluctuation information, distance fluctuation information, angle fluctuation information and micro-Doppler frequency shift information of the target, namely: , , the recognition result of the ball camera is used as the true category label; The neural network results include: input layer: the dimension of the feature vector; hidden layer: multiple fully connected layers, each layer contains several neurons, and uses the ReLU activation function; output layer: fully connected layer, the number of output nodes is equal to the number of categories, and the softmax activation function is used; the feature vector is forward propagated through the neural network to obtain the predicted probability distribution and reconstructed features; then the loss function is used to calculate the total loss, the gradient is calculated through the back propagation algorithm, and the model parameters are updated to minimize the loss function. The recognition model is iteratively trained until the preset maximum number of iterations is met. After the training is completed, the data used for training is deleted from the feature information structure; the expression of the loss function is:

[0031] In the formula, represents the loss function; Indicates the total number of target categories; Indicates The true label of the class target; express The predicted probability of the class target; represents the first elements; Represents the first feature vector reconstructed by the autoencoder elements; Represents the preset weight matrix; represents the regularization term coefficient; represents the coefficient of the reconstruction error term; Represents the number of elements in the feature vector; Step S500 effectively improves the recognition accuracy and robustness of the radar recognition model through the filtered high-quality training data, reduces the false alarms caused by low-quality data, and through continuously receiving the recognition results of the ball camera for training, the radar can dynamically adjust its recognition model and gradually improve the recognition capability.

[0032] In summary, the security radar false alarm management method and system based on traction learning provided by the present application utilize the advantages of the ball camera in the field of target identification and classification. The advantages of the ball camera in the field of image recognition can make up for the problem of insufficient recognition accuracy of the radar in some cases, and provide feedback on the results reported by the radar. The radar uses the feedback results to train and learn itself, that is, the radar is pulled by the ball camera to learn, rather than pre-collecting a large amount of data to train the radar. The radar can identify real targets and false targets by continuously receiving feedback and learning, effectively reducing the number of times the radar reports false targets, thereby reducing the number of times the overall system reports false targets and reducing the false alarm rate of the security monitoring system; on the other hand, the radar can perform specific learning for the area that needs to be monitored, which not only reduces the demand for large-scale pre-training data sets, but also enables the radar to learn for the scenes that need to be monitored, thereby improving adaptability.

[0033] See also Figure 3 , which shows a schematic diagram of the structure of a security radar false alarm management system based on traction learning provided by an embodiment of the present invention. The technical solution of the second aspect of the present invention provides a security radar false alarm management system based on traction learning, which adopts the security radar false alarm management method based on traction learning provided by the technical solution of the first aspect of the present invention, and the system includes: A radar configured to obtain location information and related characteristic information of a target and transmit the location information to a ball camera; The ball camera is configured to identify the category of the target detection object based on the target position information and extract the identification result; A time determination module, configured to make a determination based on the system time and extract a time determination result; An environment determination module is configured to perform environment information determination according to the time determination result and extract the environment determination result; The traction learning module is configured to feed back the recognition results to the radar according to the environmental judgment results and train the radar recognition model in combination with the feature information to update the recognition model parameters.

[0034] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0035] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A security radar false alarm management method based on traction learning, characterized in that: The method comprises: Use radar to obtain the target's location information and related feature information, and transmit the location information to the ball camera; The ball camera identifies the category of the target object based on the target position information and extracts the identification result; Make judgment based on system time and extract time judgment result; Perform environmental information judgment based on the time judgment result and extract the environmental judgment result; The recognition results are fed back to the radar based on the environmental judgment results and the radar recognition model is trained in combination with the feature information to update the recognition model parameters.

2. The security radar false alarm management method based on traction learning as claimed in claim 1, characterized in that: The relevant characteristic information of the target obtained by radar includes: Use radar to obtain the electromagnetic scattering characteristics and target motion characteristics of the target.

3. The security radar false alarm management method based on traction learning as claimed in claim 2, characterized in that: The use of radar to obtain target-related feature information also includes: The radar obtains the target's electromagnetic scattering characteristic information and target motion characteristic information and stores them in the characteristic information structure.

4. The security radar false alarm management method based on traction learning as claimed in claim 3 is characterized in that: The electromagnetic scattering characteristic information of the target obtained by the radar includes: Extracting the echo peak energy from the target echo signal; Collect the energy of several noise samples in the background area around the target; The signal-to-noise ratio is calculated as the ratio of the peak energy of the echo to the average energy of several noise samples.

5. The security radar false alarm management method based on traction learning as claimed in claim 4, characterized in that: The target motion characteristic information acquired by the radar includes: Extract speed fluctuation information, distance fluctuation information, and angle fluctuation information based on the target speed, distance, and angle information of the current detection frame and the previous detection frame of the radar; The micro-Doppler frequency shift information is extracted from the radar echo signal using short-time Fourier transform.

6. The security radar false alarm management method based on traction learning according to any one of claims 1 to 5, characterized in that: According to the system time, the extraction time judgment results include: Get the current system time, determine whether the current time is day or night, and output the time determination result.

7. The security radar false alarm management method based on traction learning as claimed in claim 6, characterized in that: The environmental information is judged based on the time judgment result, and the environmental judgment result is extracted including: If the time determination result is night, the recognition result will not be fed back to the radar and the relevant feature information will be deleted; If the time determination result is daytime, determine whether the external environment is sunny at the current moment and output the environment determination result.

8. The security radar false alarm management method based on traction learning as claimed in claim 6, characterized in that: According to the environmental judgment result, the recognition result is fed back to the radar and the radar recognition model is trained in combination with the feature information. The recognition model parameters are updated including: If the environment is judged as not clear, the recognition result will not be fed back to the radar and the relevant feature information will be deleted; If the environment is judged to be clear, the recognition result will be fed back to the radar, and the recognition result and related feature information will be used as training samples to train the radar recognition model. After the training is completed, the related feature information will be deleted.

9. The security radar false alarm management method based on traction learning as claimed in claim 8, characterized in that: Using the recognition results and related feature information as training samples to train the radar recognition model includes: Combining relevant feature information into a feature vector, and using the recognition result as a true category label, wherein the relevant feature information at least includes electromagnetic scattering feature information, speed fluctuation information, distance fluctuation information, angle fluctuation information, and micro-Doppler frequency shift information of the target; The radar recognition model is constructed and trained based on the neural network, and the loss function is expressed as: In the formula, represents the loss function; Indicates the total number of target categories; Indicates The true label of the class target; express The predicted probability of the class target; represents the first elements; Represents the first feature vector reconstructed by the autoencoder elements; Represents the preset weight matrix; represents the regularization term coefficient; represents the coefficient of the reconstruction error term; Represents the number of elements in the feature vector.

10. A security radar false alarm management system based on traction learning, characterized in that: The security radar false alarm management method based on traction learning according to any one of claims 1 to 9 is adopted, and the system comprises: A radar configured to obtain location information and related characteristic information of a target and transmit the location information to a ball camera; The ball camera is configured to identify the category of the target detection object based on the target position information and extract the identification result; A time determination module is configured to make a determination based on the system time and extract a time determination result; An environment determination module is configured to perform environment information determination according to the time determination result and extract the environment determination result; The traction learning module is configured to feed back the recognition results to the radar according to the environmental judgment results and train the radar recognition model in combination with the feature information to update the recognition model parameters.

Images