A radar and visual fusion security protection method suitable for field scenes

Abstract

The application particularly relates to a radar and visual fusion security protection method suitable for a field scene. The method comprises the following steps: acquiring radar target data; selecting a suspicious target from the radar target data and calculating the guiding parameters of the suspicious target, guiding photoelectric acquisition of photoelectric identification data of the suspicious target; taking the radar identification ID of the suspicious target as an index, combining a time stamp to perform space-time alignment on the radar target data and the photoelectric identification data of the suspicious target; constructing a confidence fusion model, acquiring current environmental condition parameters, dynamically adjusting the weight coefficient of the confidence fusion model, and generating a fusion confidence; according to the fusion confidence, combining the consistency judgment rule of the radar target type and the photoelectric target type, determining the final target type of the suspicious target, and generating alarm data according to the final target type. The method overcomes the performance attenuation defects of radar and photoelectric detection equipment caused by complex factors in the field scene, and significantly improves the robustness and accuracy of target identification of the radar and visual fusion security protection system.

Description

Technical Field

[0001] This invention relates to the field of outdoor security technology, and more specifically to a radar-visual fusion security protection method suitable for outdoor scenarios. Background Technology

[0002] Field security systems are a key technology for ensuring the safety of vast areas such as border regions, oil fields, and nature reserves. With the upgrading of sensing requirements, single radar or video surveillance methods are no longer sufficient to cope with the complex and ever-changing field environment. Radar has a long detection range and can operate 24 / 7, but it cannot provide intuitive images; video surveillance offers strong visualization and rich semantic information, but it is significantly affected by lighting and weather conditions. Therefore, current technologies are gradually moving towards the integration of radar and video, aiming to combine the advantages of both to achieve more stable and accurate all-weather security monitoring.

[0003] However, the field environment is highly unstructured: dynamic factors such as sudden changes in lighting, rain and fog interference, and swaying vegetation can significantly alter the perception performance of radar and video. In existing technologies, radar-video fusion systems mostly employ pre-defined fusion rules to combine and judge the identification results. While these methods can operate in controlled environments, they struggle to adapt to the dynamic degradation of sensor performance caused by environmental changes in actual field deployments, leading to a decline in the overall reliability of the system's identification capabilities.

[0004] The limitations of existing technologies prevent the system from establishing a dynamic reliability assessment and decision correction mechanism in the face of the perception uncertainties introduced by field environmental conditions. This results in multi-source information often being isolated or simply superimposed, failing to form complementarity and synergistic enhancement. Consequently, in unstructured field scenarios, the system exposes defects such as insufficient decision robustness and the coexistence of false alarms and missed alarms. Summary of the Invention

[0005] The purpose of this invention is to provide a radar-visual fusion security protection method suitable for outdoor scenarios. This method effectively overcomes the performance degradation defects of radar and photoelectric detection equipment caused by factors such as lighting, weather, and vegetation movement in outdoor scenarios, significantly improves the robustness and accuracy of target recognition in the radar-visual fusion security protection system, and greatly reduces the false alarm and missed alarm rates.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] This invention provides a radar-visual fusion security protection method suitable for outdoor scenarios, comprising:

[0008] S01, acquire radar target data generated by the radar module scanning the environment; the radar target data includes the radar identification ID, radar target type and radar data confidence level of each detected target;

[0009] S02, Select a suspicious target from the radar target data according to the preset tracking strategy and calculate its guidance parameters. Drive the photoelectric module to lock and track the suspicious target based on the guidance parameters and obtain its photoelectric identification data; the photoelectric identification data includes the photoelectric target type and photoelectric data confidence level of the suspicious target;

[0010] S03, using the radar identification ID of the suspicious target as an index, and combining it with the timestamp, the radar target data and photoelectric identification data of the suspicious target are spatiotemporally aligned;

[0011] S04, construct a confidence fusion model and obtain the current environmental condition parameters, dynamically adjust the weight coefficients of the confidence fusion model based on the environmental condition parameters, and use the adjusted confidence fusion model to perform weighted calculation on the radar data confidence and photoelectric data confidence to generate a fused confidence.

[0012] S05. Based on the fusion confidence level of the suspicious target and the consistency judgment rule of its radar target type and photoelectric target type, determine the final target type of the suspicious target, and generate alarm data based on the final target type. The alarm data is used to trigger physical security warnings.

[0013] As a preferred embodiment of the present invention, the radar module consists of two millimeter-wave radars with mutually perpendicular detection axes. The two millimeter-wave radars are arranged horizontally at a 90° central angle interval, covering a 180° horizontal sector.

[0014] As a preferred embodiment of the present invention, the preset tracking strategy used in step S02 includes, but is not limited to, time-priority strategy, distance-priority strategy and speed-priority strategy;

[0015] The time-priority strategy specifically involves selecting the detected target with the earliest timestamp as a suspicious target.

[0016] The distance-priority strategy specifically involves selecting the closest detected target to the radar module as a suspicious target.

[0017] The speed-priority strategy specifically selects the target with the highest movement speed as a suspicious target.

[0018] As a preferred embodiment of the present invention, in step S02, the calculated guidance parameters include the horizontal angle and elevation angle of the suspected target under the detection of the radar module.

[0019] As a preferred embodiment of the present invention, in step S04, the weight coefficients of the confidence fusion model are dynamically adjusted based on the environmental condition parameters, specifically as follows:

[0020] When environmental conditions are deemed favorable, the radar weighting coefficient is set to be less than the photoelectric weighting coefficient; when environmental conditions are deemed unfavorable, the radar weighting coefficient is set to be greater than the photoelectric weighting coefficient.

[0021] As a preferred embodiment of the present invention, the radar target type and the photoelectric target type have the same type label, which includes, but is not limited to, personnel, vehicles, animals and unknown objects.

[0022] As a preferred embodiment of the present invention, in step S04, the consistency determination rule specifically includes:

[0023] If the fusion confidence level is greater than or equal to the first threshold, and the radar target type is consistent with the photoelectric target type, the radar target type is determined as the final target type; if the two types are inconsistent, the target type with the higher confidence level is determined as the final target type.

[0024] If the fusion confidence level is between the second threshold and the first threshold, then the photoelectric target type is determined as the final target type;

[0025] If the fusion confidence level is less than the second threshold, then the unknown is determined as the final target type.

[0026] As a preferred embodiment of the present invention, step S05, which involves generating alarm data based on the final target type and triggering a physical security warning, specifically comprises:

[0027] If the type label corresponding to the final target type is personnel or vehicle, then a level 1 alarm data is generated, and at the same time, an audible and visual physical drive-away is performed, and the level 1 alarm data is uploaded to the portable application terminal.

[0028] If the type label corresponding to the final target type is animal or unknown, then secondary alarm data is generated and uploaded to the portable application terminal.

[0029] In summary, the present invention has the following beneficial effects:

[0030] This invention overcomes the shortcomings of existing technologies that use static, preset fusion rules by introducing environmental condition parameters to dynamically adjust the weight coefficients in the confidence fusion model. In "good" environments with good lighting and high visibility, the system assigns higher weights to photoelectric data, making full use of its advantages of strong visualization and rich semantic information for accurate identification. In "unfavorable" environments such as nighttime, rain, and fog, the system automatically switches to trusted radar data, leveraging its advantages of all-weather operation and strong anti-interference capabilities. This method enables the system to proactively adapt to sensor performance degradation caused by dynamic factors such as sudden changes in lighting and weather interference in the field environment, ensuring the stability and accuracy of the sensing system under different conditions, thereby significantly reducing false alarms or missed detections caused by the temporary failure of a single sensor.

[0031] This invention achieves synergy between radar and visual data sources based on a comprehensive judgment rule that integrates confidence level and type consistency. The system not only performs confidence-weighted fusion but also handles consistency or conflict between radar and photoelectric target types in a hierarchical manner according to the fusion result. At high confidence levels, type consistency is prioritized or the higher-confidence target is selected; at medium confidence levels, the visualization results of the image are relied upon; and at low confidence levels, the target is cautiously classified as an "unknown object." This mechanism provides a quantifiable framework for comprehensive judgment when there is ambiguity in information from different sensors. Its deep coupling of multi-source information significantly improves the reliability of the overall system in identifying intrusive targets in complex, unstructured scenarios and effectively distinguishes between real threats and interference. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of the method of the present invention;

[0034] Figure 2 This is a schematic diagram of a radar module consisting of two millimeter-wave radars in one embodiment. Detailed Implementation

[0035] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described in some examples may be combined in other examples.

[0036] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0037] Combination Figure 1 As shown, this embodiment provides a specific implementation of a radar-visual fusion security protection method suitable for outdoor scenarios. This method is based on a radar-visual fusion rapid deployment device, which can communicate with a portable application terminal via a wireless communication module to achieve system-level security protection and alarm notifications. The method of this invention relies on the following radar-visual integrated rapid deployment system, which consists of a radar-visual fusion rapid deployment device (radar module, photoelectric module, photosensitive module, main control module, servo module, warning module), a wireless communication module, and a portable application terminal. Among them, the radar module and the optoelectronic module are the two core information acquisition modules. The former is responsible for continuously scanning the environment and acquiring radar data such as target distance, azimuth, radial velocity, RCS characteristics, latitude and longitude; the latter is responsible for image-level precise tracking and analysis based on the data provided by the radar module. The photosensitive module is used to sense the ambient light intensity; the servo module is used to drive the optoelectronic module to detect suspicious targets; the warning module is responsible for executing physical security warning actions. Each module is electrically connected to the main control module for data reporting and command reception. The main control module is implemented using a high-performance SoC and is the computing core, responsible for visual recognition, image processing, target tracking strategy driving, multi-source data fusion analysis, energy-saving management, alarm linkage control, etc. It also has data storage functions.

[0038] First, the equipment is deployed. In this embodiment, the radar module consists of two 24GHz millimeter-wave radars with mutually perpendicular detection axes. The two millimeter-wave radars are arranged horizontally at a 90° central angle, covering a 180° horizontal sector. The optoelectronic module is integrated with a visible light lens and an uncooled infrared thermal imaging lens. The operator sets up the equipment at a high point, using a tripod for stable installation, and connects it to a power source. After the equipment is powered on, the radar module automatically starts and begins environmental scanning. Simultaneously, it establishes a communication connection with the wireless self-organizing network module inserted in the portable application terminal through its built-in wireless communication module, forming a wireless communication link between the equipment and the portable application terminal. Subsequently, the operator completes the configuration of basic equipment parameters (including equipment name, equipment ID, product number, etc.), time synchronization, user management, network, system upgrade, video encoding, OSD information, display mode, and PTZ control function in the client software interface of the portable application terminal. The above configuration information is sent to the radar-visual converged rapid deployment equipment through the established wireless communication link, completing the system initialization settings.

[0039] Execute S01 to acquire radar target data generated by the radar module's environmental scan; this step specifically involves:

[0040] The two millimeter-wave radars in the radar module periodically transmit linear frequency modulated continuous waves. When the signal encounters the target being detected, it generates an echo that returns to the corresponding millimeter-wave radar, thereby acquiring information such as the target's distance, speed, and azimuth relative to the radar module. The radar module then sends all of the above information to the main control module, which performs analysis and calculation (in other possible implementations, this analysis and calculation can also be performed by the high-performance radar module itself, which carries the computing core), to obtain radar target data including the radar identification ID, radar target type, and radar data confidence level of each target being detected.

[0041] In this invention, it should be noted that, as Figure 2As shown, the arrangement of dual millimeter-wave radars creates a small overlap area (e.g., [-5°, 5°]). Therefore, a target may be identified by any single millimeter-wave radar (within its own area), or it may be identified by both radars simultaneously because it is within the overlap area. In practice, the two millimeter-wave radars independently acquire and analyze signals, obtaining multiple raw point data points arranged based on timestamps. The main control module transforms the azimuth angles of these raw point data points into a unified angle system of [-90°, 90°], filters out raw data points with azimuth angles belonging to the [-5°, 5°] interval, and clusters them according to time periods (multiple timestamps). Data points within the same time period are grouped into a data sequence, resulting in multiple data sequences. Data from different radars within the sequences are paired (note: the two data points cannot come from the same radar, as this is to remove duplicate targets seen by both radars), and their distances are transformed into two dimensions in a unified coordinate system. The coordinate values ​​are used to calculate the distance. If the calculated distance is less than or equal to a preset threshold (e.g., 1 meter), the data is determined to be the same target, and the signal-to-noise ratio (SNR) deduplication method is activated to determine which radar data to discard. Specifically, the SNR values ​​of the two original point data are compared. The millimeter-wave radar identification data with the larger SNR value is always retained, and the other data is deleted. If the SNR values ​​are the same, either one is retained or the one with the smallest radar identification ID is retained. This saves computing and storage resources and assigns a unique radar identification ID to it. If the distance is greater than the threshold, the data is determined to be two different targets. Subsequently, the two original point data are synchronously retained and radar identification IDs are assigned to them respectively.

[0042] Next, micro-Doppler feature extraction is performed to calculate the beat signal acquired by the radar module for the suspicious target. The time spectrum STFTt,f is obtained through short-time Fourier transform, and feature vectors are extracted from STFTt,f. :

[0043]

[0044] in, For the centroid of the spectrum, For spectrum bandwidth, This is the spectral entropy.

[0045] Next, RCS feature extraction is performed to calculate the RCS features of the suspicious target under radar module detection. The specific calculation is as follows:

[0046]

[0047] For transmission power, Let R be the antenna gain, where R is the distance between the target being detected and the radar module. For millimeter-wave radar operating wavelength, This represents the echo power received by the millimeter-wave radar.

[0048] Extracting RCS statistical features: RCS mean μ σ RCS variance Fluctuation factor

[0049] Based on micro-Doppler and RCS features, construct the feature vector x:

[0050]

[0051] The feature vector x is then standardized.

[0052]

[0053] A one-to-one multi-classification strategy is adopted to establish 6 SVM classifiers for the categories that need to be classified (in this embodiment, the radar target type and the photoelectric target type have the same type label, and the type label includes, but is not limited to, personnel, vehicles, animals and unknown objects):

[0054] The classification categories are defined as follows:

[0055] C = {c1 = pedestrian, c2 = vehicle, c3 = animal, c4 = unknown object}

[0056] Each Classifier uses categories and Use the samples for training:

[0057]

[0058] The RBF kernel function and decision function used by each SVM classifier are as follows:

[0059] The RBF kernel function is:

[0060]

[0061] Here, γ is the kernel function parameter, used to control the influence range of a single sample, and a grid search method is used.

[0062] The Euclidean distance between two samples;

[0063] The decision function is :

[0064]

[0065] , It is a symbolic function; The sample to be classified; The Lagrange multipliers for support vectors; The label for the support vector (+1 or -1); For bias terms; The number of support vectors; For RBF kernel functions;

[0066] For each new sample x, six SVM classifiers are used for voting:

[0067]

[0068] in, For category Total number of votes received; This is an indicator function; its value is 1 when the condition is true, and 0 otherwise. Represents classifier Classify the samples as ;

[0069] After voting, obtain the voting result vector:

[0070]

[0071] Based on the voting results, calculate the confidence level of each category for sample x:

[0072]

[0073] For category The final confidence level; the sum of all confidence levels is 1: ;

[0074] Finally, based on the principle of maximizing confidence, the final radar target type y and the corresponding radar data confidence level are determined:

[0075]

[0076]

[0077] S02, select a suspicious target from the radar target data according to the preset tracking strategy and calculate its guidance parameters, drive the photoelectric module to lock the suspicious target based on the guidance parameters, and obtain the photoelectric identification data of the suspicious target; the photoelectric identification data includes the photoelectric target type and photoelectric data confidence level of the suspicious target.

[0078] In step S02, the preset tracking strategy is implemented based on a program pre-stored in the main control module's register area. The preset strategy includes, but is not limited to, time-priority strategy, distance-priority strategy, and speed-priority strategy.

[0079] The time-priority strategy specifically involves selecting the detected target with the earliest timestamp as a suspicious target.

[0080] The distance-priority strategy specifically involves selecting the closest detected target to the radar module as a suspicious target.

[0081] The speed priority strategy specifically selects the target with the highest moving speed (i.e., the speed that moves closer to the device is positive, and the speed that moves further away from the device is negative) as a suspicious target.

[0082] Different strategies are selected to identify suspicious targets in different application scenarios. Suspicious targets are selected from the set of multiple targets detected by the radar modules. They are identified as high-risk targets through the screening of preset tracking strategies under specific security scenarios. Therefore, the main control module controls the servo module to drive the photoelectric module to lock onto the suspicious target. Before this process, the main control module needs to calculate the guidance parameters. This step is as follows:

[0083] After selecting a suspected target, the main control module needs to calculate the horizontal angle that the photoelectric module should point to. and pitch angle (i.e., guidance parameters). The horizontal angle is directly taken from the azimuth angle of the suspected target provided by the radar module. Regarding pitch angle Because the radar module does not provide elevation angle It needs to be estimated using the target distance R and the target height preset by the system.

[0084] Assuming the suspicious target is located at an altitude of (Assigning values ​​based on the target type initially identified by the radar), the radar installation height is... Based on geometric relationships, the formula for calculating the pitch angle is:

[0085]

[0086] in, The altitude at which the suspicious target is located; R represents the installation height of the radar module; R is the distance between the suspected target and the radar module.

[0087] The calculated horizontal angle and pitch angle The parameters are sent from the main control module to the servo module, enabling the photoelectric module to quickly and smoothly turn towards the target direction.

[0088] Next, to ensure the suspicious target is clear and of appropriate size in the photoelectric image, automatic focusing and zooming are required. The main control module adjusts the focus and zoom according to the selected target distance. The image distance (focusing value) is calculated using the Gaussian formula and sent to the photoelectric module.

[0089]

[0090] in, Indicates image distance; Indicates object distance (corresponding to target distance) ); Indicates focal length;

[0091] The next step is to control the pixel size of the target in the image by zooming in, in order to achieve the focal length required to reach the desired pixel size. The calculation method is as follows:

[0092]

[0093] The pixel size that the suspected target is expected to occupy in the visible light image; The physical size of a single pixel in a visible light image sensor; Typical physical dimensions of a suspicious target.

[0094] The required focal length value is further sent by the main control module to the visible light camera, which then drives its internal motor to the specified focal length. For the infrared lens, since its focal length is fixed, the expected pixel size of the target in the infrared image can be calculated.

[0095]

[0096] During tracking, the photoelectric module continuously acquires dual-spectral video streams at a rate of 25fps and sends them to the main control module, thus successfully acquiring image recognition data of the suspicious target. Then, based on the deep learning algorithm model in the main control module, the suspicious target is identified and classified, and its photoelectric target type L is output. optic Confidence level C of photoelectric data optic Captured images, tracking videos, and radar identification IDs of suspicious targets.

[0097] S03, using the radar identification ID of the suspicious target as an index, and combining it with the timestamp, the radar target data and image recognition data of the suspicious target are spatiotemporally aligned;

[0098] Specifically, in this step, the radar identification ID of the suspected target is used as an index (that is, it also corresponds to the radar identification ID of the detected target in the radar target data obtained in S01, because the preceding identity definition of each suspected target is a detected target), based on the timestamp T. n Aligning radar target data and photoelectric identification data of suspicious targets to form fused data information:

[0099]

[0100] It should be noted that in this step, at any given moment, the target locked by the center of the field of view of the optoelectronic module is also the target that the radar module is steadily tracking, and corresponds to a unique radar identification ID for the suspicious target.

[0101] S04, construct a confidence fusion model and obtain the current environmental condition parameters, dynamically adjust the weight coefficients of the confidence fusion model based on the environmental condition parameters, and use the adjusted confidence fusion model to perform weighted calculation on the confidence of the radar data and the confidence of the photoelectric data to generate a fused confidence score.

[0102] The purpose of this step is to obtain the fusion confidence score, and its specific calculation operation is as follows:

[0103] Confidence fusion of target types is performed using a weighted average method:

[0104]

[0105] in: To integrate confidence levels; Confidence level of radar data; The confidence level of the aligned photoelectric data; These are the weighting coefficients, and .

[0106] Next, adaptive weight allocation is performed, dynamically adjusting the weights based on environmental conditions, specifically as follows:

[0107] When environmental conditions are deemed favorable, the radar weighting coefficient is set to be less than the photoelectric weighting coefficient; when environmental conditions are deemed unfavorable, the radar weighting coefficient is set to be greater than the photoelectric weighting coefficient.

[0108] This is because the photoelectric module provides a stronger final interpretation of the target, but its reliability is lower when environmental conditions are poor. Therefore, when environmental conditions are good, increasing the photoelectric weight can achieve better interpretation with higher reliability. However, when environmental conditions are poor, increasing the radar weight to be greater than the photoelectric weight can ensure the reliability of the target detection.

[0109] Specifically, in this embodiment, the experimental data or statistical model assumes good daytime weather and sets a radar weighting coefficient. This is solely to enhance the impact of high-quality photoelectric images; photoelectric performance degrades at night / inclement weather (rain, fog, snow), so radar weighting coefficients are set accordingly. .

[0110] S05. Based on the fusion confidence level of the suspicious target and the consistency judgment rule of its radar target type and photoelectric target type, determine the final target type of the suspicious target, and generate alarm data based on the final target type. The alarm data is used to trigger physical security warnings.

[0111] As shown in S04 above, the fusion confidence level of each suspicious target has been obtained. Then, based on the fusion confidence of the suspicious targets By combining the consistency judgment rules of radar target type and photoelectric target type, the final target type of the suspicious target is determined.

[0112] The consistency determination rule is explained in detail below:

[0113] It is known that the first threshold and the second threshold are preset fusion confidence judgment thresholds, and the first threshold is greater than the second threshold. In this embodiment, the first threshold is 0.8 and the second threshold is 0.2 as an example for explanation.

[0114] For each suspected target, there are known radar target type and electro-optical target type; the two type labels may be the same or different. The fusion confidence level of a suspected target... A value greater than 0.8 corresponds to a high confidence level determination. When the radar target type and the electro-optical target type are consistent, the radar target type is determined as the final target type; when the two types are inconsistent, the target type with the higher confidence level is determined as the final target type. This can achieve optimal target type determination based on confidence level.

[0115] If the fusion confidence of a suspicious target If the confidence level is less than 0.8 but greater than 0.2, it corresponds to a medium confidence level determination. The photoelectric target type is determined as the final target type and marked as "to be confirmed". At this time, confidence level is still the core consideration, but the objective contribution of the photoelectric module to the target type is enhanced to obtain relatively high reliability.

[0116] If the fusion confidence of a suspicious target If the confidence level is less than 0.2, it corresponds to a low confidence level judgment, and the unknown object is directly identified as the final target type.

[0117] Based on the above steps, the final target type label for each suspicious target can be obtained.

[0118] Finally, step S05 also includes a step of generating alarm data based on the final target type, wherein the alarm data is used to trigger a physical security warning.

[0119] Specifically, if the type label corresponding to the final target type is personnel or vehicle, then a level 1 alarm data is generated, and at the same time, an audio-visual physical drive-away is performed, and the level 1 alarm data is uploaded to the portable application terminal.

[0120] If the type label corresponding to the final target type is animal or unknown, then secondary alarm data is generated and uploaded to the portable application terminal.

[0121] The portable application terminal will display pop-up prompts for first-level or second-level alarm data and generate relevant alarm record data.

[0122] In other possible implementations, operators can view alarm notifications via portable application terminals, and further examine data such as target type, tracking video, radar trajectory, and time within the alarm data. They can also access an online preview interface to observe and analyze the real-time video and motion trajectory of the target of interest. For alarm data with the target type unknown and the tag "to be confirmed," operators will manually process and verify the data to confirm the actual target type, which will be used for radar and image recognition algorithm optimization.

[0123] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A radar-visual fusion security protection method suitable for outdoor scenarios, characterized in that, include: S01, acquire radar target data generated by the radar module through environmental scanning; The radar target data includes the radar identification ID, radar target type, and radar data confidence level for each detected target; S02, Select a suspicious target from the radar target data according to the preset tracking strategy and calculate its guidance parameters. Drive the photoelectric module to lock and track the suspicious target based on the guidance parameters and obtain its photoelectric identification data; the photoelectric identification data includes the photoelectric target type and photoelectric data confidence level of the suspicious target; S03, using the radar identification ID of the suspicious target as an index, and combining the timestamp, the radar target data and photoelectric identification data of the suspicious target are spatiotemporally aligned; S04, Construct a confidence fusion model and obtain the current environmental condition parameters, and dynamically adjust the weight coefficients of the confidence fusion model based on the environmental condition parameters; The confidence scores of the radar data and the photoelectric data are weighted and calculated using the adjusted confidence fusion model to generate a fused confidence score. S05. Based on the fusion confidence level of the suspicious target and the consistency judgment rule of its radar target type and photoelectric target type, determine the final target type of the suspicious target, and generate alarm data based on the final target type. The alarm data is used to trigger physical security warnings.

2. The radar-visual fusion security protection method applicable to outdoor scenarios according to claim 1, characterized in that, The radar module consists of two millimeter-wave radars with their detection axes perpendicular to each other. The two millimeter-wave radars are arranged horizontally with a 90° central angle between them, covering a 180° horizontal sector.

3. The radar-visual fusion security protection method applicable to outdoor scenarios according to claim 1, characterized in that, In step S02, the preset tracking strategy used includes, but is not limited to, time-priority strategy, distance-priority strategy, and speed-priority strategy; The time-priority strategy specifically involves selecting the detected target with the earliest timestamp as a suspicious target. The distance-priority strategy specifically involves selecting the closest detected target to the radar module as a suspicious target. The speed-priority strategy specifically selects the target with the highest movement speed as a suspicious target.

4. A radar-visual fusion security protection method suitable for outdoor scenarios according to claim 2, characterized in that, In step S02, the calculated guidance parameters include the horizontal and vertical angles of the suspected target as detected by the radar module.

5. A radar-visual fusion security protection method suitable for outdoor scenarios according to claim 1, characterized in that, In step S04, the weight coefficients of the confidence fusion model are dynamically adjusted based on the environmental condition parameters, specifically as follows: When environmental conditions are deemed favorable, the radar weighting coefficient is set to be less than the photoelectric weighting coefficient; when environmental conditions are deemed unfavorable, the radar weighting coefficient is set to be greater than the photoelectric weighting coefficient.

6. The radar-visual fusion security protection method applicable to outdoor scenarios according to claim 1, characterized in that, The radar target type and the optoelectronic target type have the same type label, which includes, but is not limited to, personnel, vehicles, animals, and unknown objects.

7. A radar-visual fusion security protection method suitable for outdoor scenarios according to claim 1, characterized in that, In step S04, the consistency determination rule specifically includes: If the fusion confidence level is greater than or equal to the first threshold, and the radar target type is consistent with the photoelectric target type, the radar target type is determined as the final target type; if the two types are inconsistent, the target type with the higher confidence level is determined as the final target type. If the fusion confidence level is between the second threshold and the first threshold, then the photoelectric target type is determined as the final target type; If the fusion confidence level is less than the second threshold, then the unknown is determined as the final target type.

8. A radar-visual fusion security protection method suitable for outdoor scenarios according to claim 6, characterized in that, In step S05, the specific steps for generating alarm data based on the final target type and triggering a physical security warning are as follows: If the type label corresponding to the final target type is personnel or vehicle, then a level 1 alarm data is generated, and at the same time, an audible and visual physical drive-away is performed, and the level 1 alarm data is uploaded to the portable application terminal. If the type label corresponding to the final target type is animal or unknown, then secondary alarm data is generated and uploaded to the portable application terminal.

Images