Life detection information presentation method and device based on radar detection and depth analysis and medium

Abstract

The invention discloses a life detection information presentation method and device based on radar detection and depth analysis and a medium, and relates to the technical field of radar detection and information processing. The method comprises the steps of processing and imaging an original echo signal, performing image enhancement on an obtained amplitude image, and performing target detection through a deep learning model; in combination with a target state memory bank and a Kalman filter, performing time sequence optimization and trajectory tracking on a detection result, and outputting target actual position information with dynamic and static attributes; performing clustering and confidence evaluation on static targets, and performing trajectory clustering and second-level behavior recognition on dynamic targets; and finally, synchronously presenting the information on a display control interface in real time in a form of associating the graphical mark with the semantic text description. The problems that a traditional life detection radar is single in information presentation, insufficient in analysis and non-traceable are solved, and the detection accuracy, the information interpretation depth and the system usability are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of radar detection and information processing technology, specifically relating to an intelligent information presentation and analysis method for life detection radar, which is particularly suitable for target detection, behavior analysis and multimodal information presentation in non-contact life detection scenarios such as through-wall radar. Background Technology

[0002] Life detection radar is a technology that utilizes the ability of electromagnetic waves to penetrate non-metallic obstacles (such as walls and ruins) to achieve non-contact detection of concealed human targets. Its basic working principle is as follows: the radar emits electromagnetic waves of a specific frequency band, which penetrate the obstacle and are reflected by the human body. The receiver captures the echo signal, and through signal processing steps such as filtering, imaging, and detection, the target's location information is finally output on the display interface. Due to its strong penetration, real-time performance, and environmental adaptability, this technology has been widely used in civilian and military scenarios such as disaster relief, counter-terrorism, and regional security checks.

[0003] In existing technologies, life detection radar systems typically consist of a radio frequency front-end, a signal processing module, and a display control unit. Regarding information processing and presentation, most systems employ traditional detection algorithms based on threshold judgment (such as constant false alarm rate detection and peak extraction), and present the detection results on the software interface using visual symbols (such as dots) combined with coordinates and color coding. For example, a red dot might represent a moving target, and a blue dot might represent a stationary target; users understand the target distribution and status by observing the position and color changes of the dots.

[0004] However, as application scenarios become increasingly complex and users' demands for in-depth intelligence increase, the aforementioned traditional presentation and analysis methods are gradually showing the following limitations:

[0005] (1) The information presentation dimension is singular and lacks semantic interpretation: the existing system only provides coordinates and color symbols, and fails to provide real-time text description and intelligent analysis of the target's motion behavior, posture changes, trajectory trends, etc. Users need to rely on their own experience to make subjective inferences, which can easily lead to the omission of key information or misjudgment of the situation.

[0006] (2) The detection process cannot be traced back in real time, and the historical status is difficult to trace: The system usually only outputs statistical results such as the number and location of targets after the detection is completed. It cannot support users to view the historical target status and trajectory evolution in real time during the detection process, which is not conducive to continuous tracking and comprehensive analysis.

[0007] (3) Insufficient stability and accuracy of target detection: Traditional detection methods are prone to problems such as missed detection, false detection or tracking interruption in complex scenarios such as low signal-to-noise ratio, multiple target intersection, and coexistence of moving and static targets, which affect the reliability of detection results and system robustness.

[0008] (4) Lack of in-depth behavioral analysis and confidence assessment mechanism: The existing system does not identify and classify the target movement patterns (such as loitering, moving, micro-movement) or provide the confidence probability of the target's existence, which limits the system's support capabilities in advanced applications such as tactical decision-making and situation assessment.

[0009] Therefore, there is an urgent need for a life detection radar information processing method that can integrate intelligent detection and deep analysis capabilities to achieve semantic presentation of the detection process, traceability of historical states, and analysis of target behavior, so as to improve the system's information expression, decision support and overall ease of use. Summary of the Invention

[0010] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a method, device, and medium for presenting life detection information based on radar detection and depth analysis. This addresses the problems of low information interpretation efficiency, weak situational awareness, and poor decision support in traditional life detection radar systems, which suffer from single information presentation dimensions, lack of real-time backtracking of the detection process, insufficient target detection stability, and lack of deep behavioral analysis and confidence assessment mechanisms.

[0011] This invention solves the above-mentioned technical problems through the following technical solution: a method for presenting life detection information based on radar detection and depth analysis, comprising:

[0012] The raw echo signal of the through-wall radar is processed to generate radar imaging results; the amplitude image in the radar imaging results is enhanced, and target detection is performed based on the enhanced image using a deep learning model; the target detection results are optimized and tracked by combining temporal context information, and the actual position information of the target with dynamic and static attributes is output.

[0013] Based on the actual location information and dynamic / static attributes of the target, cluster analysis and confidence assessment are performed on static targets to obtain static target clusters and their confidence probabilities, and trajectory clustering and motion pattern recognition are performed on dynamic targets to obtain their motion patterns.

[0014] The actual location information of the target, the static target cluster and its confidence probability, and the motion pattern of the moving target are presented in real time on the display and control interface in the form of graphical markers and semantic text descriptions; and the semantic text descriptions with timestamps are saved to provide backtracking of historical detection information.

[0015] This invention improves the quality of input data by performing image enhancement processing on amplitude images; it enhances the learning and generalization ability of target features by using a deep learning model instead of a traditional threshold detection algorithm; and in particular, it optimizes the detection results and tracks the trajectory by combining temporal context information, effectively addressing the instantaneous loss of static targets caused by signal-to-noise ratio fluctuations through a prediction compensation mechanism, thus ensuring the continuity of detection. This significantly improves the accuracy and stability of target detection and effectively solves the problems of high false negative and missed detection rates in complex scenarios using traditional methods.

[0016] After acquiring the actual location information of the target, further cluster analysis and confidence assessment are performed on static targets, and trajectory clustering and motion pattern recognition are performed on dynamic targets. This is no longer a simple data statistics process, but a rule-based deep reasoning process that outputs high-order semantic information such as "static target clusters and their confidence probabilities" and "motion patterns." This enables deep semantic interpretation and intelligent analysis of the detection information, greatly enhancing the system's situational awareness and decision support capabilities.

[0017] This invention upgrades the original coordinate points into "graphical markers and semantic text descriptions" and presents them synchronously and in association. This means that while observing the graphic position, users can directly obtain text descriptions about the target's state (dynamic / static), confidence level, and movement pattern (such as moving or loitering), providing an intuitive, rich, and interactive multimodal information presentation method, optimizing human-computer interaction efficiency and reducing user cognitive load.

[0018] This invention transforms the dynamic detection process into a structured timeline record by saving semantic text descriptions with timestamps. Users can review any snapshot of the detection process and textual conclusions at any time for comparative analysis, achieving full-cycle traceability of the detection process. It supports historical state review and result verification, improving the rigor and reliability of detection operations and completely solving the limitations of traditional technologies where the detection process cannot be reviewed in real time and historical states are difficult to trace.

[0019] Furthermore, the amplitude image in the radar imaging result is subjected to image enhancement processing, specifically: the amplitude image is processed using an image enhancement algorithm based on gamma transform to generate an intermediate target image.

[0020] This invention processes amplitude images using a gamma transform-based image enhancement algorithm, effectively improving the visibility of weak targets in the original radar image while suppressing background noise, generating a higher-quality "intermediate target image." This improves the input quality of subsequent deep learning models from the data source, laying a crucial foundation for achieving higher accuracy and more stable target detection.

[0021] Furthermore, the target detection results are optimized and trajectory tracked by incorporating temporal context information, including:

[0022] Based on the historical trajectory information recorded in the target state memory, the Kalman filter is used to predict the target's state in the current frame, thus obtaining the predicted position of the target.

[0023] When a target is not detected in the current frame, a compensation operation is performed, that is, the compensated detection box generated based on the predicted position is used as the final detection box of the target;

[0024] When a target is detected in the current frame, calculate the deviation between the current detection box position and the predicted position;

[0025] If the deviation is less than or equal to the first preset threshold, a selection operation is performed, that is, the current detection box is used as the final detection box of the target.

[0026] If the deviation is greater than the first preset threshold, a fusion operation is performed, that is, the current detection box is fused with the compensated detection box generated based on the predicted position, and the fusion result is used as the final detection box of the target.

[0027] The target state memory is updated based on the final detection frame.

[0028] This invention effectively solves the problems of missed static target detection and interrupted dynamic target tracking caused by the fluctuation of signal-to-noise ratio of through-wall radar in traditional single-frame detection by introducing a timing prediction and correction mechanism based on target state memory and Kalman filter. It can predict and compensate for targets lost instantaneously and intelligently correct abnormal detection results, thereby significantly improving the continuity of target detection, trajectory smoothness and overall system robustness.

[0029] This invention achieves dynamic and adaptive correction of target detection results through an intelligent decision-making mechanism based on deviation thresholds. It can accurately distinguish between normal target movement and abnormal fluctuations. While ensuring high confidence in the adoption of reliable detection results, it effectively suppresses instantaneous false detections and compensates for target loss caused by occlusion or low signal-to-noise ratio. Thus, it achieves the optimal balance between detection stability and accuracy in complex detection environments.

[0030] Furthermore, cluster analysis and confidence assessment are performed on the static targets, including:

[0031] Calculate the first distance between the newly detected static target and the center position of the existing static target cluster. If the first distance is not greater than the second preset threshold, the newly detected static target is assigned to the corresponding static target cluster. Otherwise, a new static target cluster is created with the newly detected static target as the center.

[0032] After the detection period ends, the confidence probability of the real static target corresponding to the static target cluster is calculated based on the number of targets and the dispersion of their spatial distribution within the static target cluster. The confidence probability is positively correlated with the number of targets within the cluster in a segmented manner. Within a first interval where the number of targets within the cluster does not exceed a preset threshold, the confidence probability increases with a first growth rate as the number of targets within the cluster increases. Within a second interval where the number of targets within the cluster exceeds the preset threshold, the confidence probability increases with a second growth rate as the number of targets within the cluster increases, and the second growth rate is lower than the first growth rate.

[0033] This invention utilizes a distance-threshold-based clustering algorithm to intelligently group discrete static target detection points into spatially consistent target clusters, effectively distinguishing static targets at different locations. Furthermore, by introducing a confidence assessment model that is piecewise correlated with the number of targets, this model can quickly establish initial confidence when the number of target points is small, and then increase confidence at a more robust rate as the number of target points accumulates to a certain scale. This avoids over-reliance on occasional noise points while ensuring high reliability of conclusions when sufficient evidence is obtained, thus achieving an intelligent and quantitative assessment of the existence of static targets.

[0034] Furthermore, trajectory clustering and motion pattern recognition are performed on dynamic targets, including trajectory clustering steps and motion pattern recognition steps; wherein, the trajectory clustering step is used to associate observation points belonging to the same moving target in consecutive frames as the same trajectory; the motion pattern recognition step is used to identify typical motion behaviors of dynamic targets based on the positional changes of the trajectory within a second-level time window.

[0035] This invention decomposes the moving target processing flow into two steps: "trajectory clustering" and "motion pattern recognition," achieving an intelligent transformation from raw location points to high-level behavioral semantics. Trajectory clustering ensures the continuity of the same target's movement history, providing a reliable data foundation for behavioral analysis; while motion pattern recognition, based on second-level time windows, can accurately capture and classify the target's instantaneous behavioral intentions (such as movement, loitering, etc.), thereby upgrading the detection results from simple "location reports" to tactically valuable "behavioral situational awareness," greatly enhancing the depth of information interpretation and decision support capabilities.

[0036] Furthermore, the trajectory clustering step includes:

[0037] For a newly detected dynamic target, calculate its second distance to all dynamic targets in the previous time step;

[0038] If the second distance is not greater than the third preset threshold, it is determined to be a continuation of the trajectory of the same moving target, and the position information of the newly detected dynamic target is added to the trajectory sequence of the corresponding dynamic target at the previous moment;

[0039] If the second distance is greater than the third preset threshold, then calculate the third distance between the newly detected dynamic target and the center position of each existing trajectory sequence;

[0040] If the third distance is not greater than the fourth preset threshold, the location information of the newly detected dynamic target is added to the corresponding existing trajectory sequence;

[0041] If the third distance is greater than the fourth preset threshold, then a new trajectory sequence is initialized starting from the newly detected dynamic target.

[0042] This invention achieves efficient and robust trajectory clustering through a two-level distance matching mechanism: first, it attempts to precisely match the target from the previous time step to maintain trajectory continuity; if this fails, it performs spatial similarity matching with the center of an existing trajectory sequence to handle situations where the target is briefly lost or reappears. This strategy can effectively handle complex scenarios such as intermittent target appearance and cross-movement while ensuring stable tracking of continuously moving targets, thereby constructing a complete and accurate target motion trajectory and providing high-quality input for subsequent behavior analysis.

[0043] Furthermore, the motion pattern recognition step includes:

[0044] Based on the trajectory sequence of a dynamic target, the position of the dynamic target at the start and end times is obtained within a second-level time window;

[0045] Based on the positional changes between the start and end times, at least one of the following motion patterns is identified:

[0046] If the corresponding trajectory sequence contains only one observation point, it is determined to be a newly emerging dynamic target;

[0047] If the displacement of the position change is less than the fifth preset threshold, it is determined to be local lingering or micro-movement;

[0048] If the displacement of the position change is greater than or equal to the fifth preset threshold, it is determined to be a directional movement from the starting position to the ending position.

[0049] If the starting and ending times are close in position and their trajectories form a closed profile, it is determined that there may be circular motion behavior.

[0050] This invention achieves rapid and automated identification and classification of dynamic target behavior patterns by analyzing target position changes within a second-level time window. It accurately distinguishes typical target states such as "appearance," "micro-movement," "directional movement," and "circling," transforming raw positional data into high-level information with clear tactical or behavioral semantics. This allows the system to go beyond simple target tracking, directly providing the operator with a preliminary assessment of the target's intentions, significantly enhancing the depth of understanding and response speed regarding personnel activities in concealed areas.

[0051] Furthermore, the related information is presented in real-time on the display interface in the form of graphical markers and semantic text descriptions, including:

[0052] For each detected target, a corresponding graphical marker unit and a semantic text description unit are generated and maintained on the display and control interface;

[0053] The graphical marking unit updates its display position in real time based on the actual position information of the target, and the semantic text description unit updates its description content in real time based on the attributes of the target; wherein, for static targets, the description content includes at least its position and confidence probability information; for dynamic targets, the description content includes at least its position and current motion mode information.

[0054] Furthermore, the graphical marker unit and the semantic text description unit share the same target identifier; in response to the user's selection operation on any of the units, the system locates and highlights the corresponding other unit on the display interface based on the target identifier;

[0055] The storage of the semantic text description with timestamps to provide backtracking of historical exploration information includes:

[0056] Semantic text descriptions generated at each moment are stored as historical records in chronological order of detection time, associated with the corresponding timestamps.

[0057] The display and control interface provides a history viewing panel that presents the historical records in chronological order. The history viewing panel supports user interaction. When a user selects a historical record, the display and control interface is triggered to synchronously display snapshot information of the historical detection scene corresponding to the timestamp of the record. The snapshot information includes at least the distribution of the target graphical markers at the corresponding time.

[0058] This invention achieves deep integration and intuitive expression of detection information by establishing a one-to-one, data-linked graphical and textual presentation unit for each target. The graphical marker unit provides an intuitive spatial situational awareness, while the semantic textual description unit provides precise semantic interpretation; their real-time synchronization ensures information consistency. In particular, the interactive linkage function achieved through shared target identifiers allows users to quickly perform graphical and textual cross-referencing and precise positioning in complex multi-target scenarios, greatly reducing the cognitive load and risk of misjudgment for operators, and significantly improving human-computer interaction efficiency and the accuracy of situational awareness.

[0059] This invention achieves full-cycle, verifiable backtracking of the detection process by using structured storage of timestamps and semantic descriptions, coupled with an interactive historical viewing interface. Users can not only review historical conclusions chronologically, but also intuitively verify the complete detection situation at any given moment through interactive operations such as "clicking on text to recreate the scene." This completely changes the limitation of traditional radar systems where "conclusions are fixed upon detection completion," providing strong technical support for post-event analysis, process verification, and collaborative decision-making, and greatly enhancing the auditability and reliability of detection operations.

[0060] Based on the same concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the life detection information presentation method based on radar detection and depth analysis as described above.

[0061] Based on the same concept, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the life detection information presentation method based on radar detection and depth analysis as described above.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] This invention fundamentally improves the detection accuracy and continuous tracking capability of through-wall radar for life targets in complex scenarios (such as low signal-to-noise ratio, multiple targets, and mixed dynamic and static targets) through a technology chain of "amplitude image enhancement → deep learning detection → time-series optimized tracking", effectively solving the problems of missed detection and high false detection rate of traditional methods.

[0064] This invention performs "static target clustering and confidence assessment" and "dynamic target trajectory clustering and motion pattern recognition" on the detection results. This upgrades the system output from the original coordinate points to in-depth analysis results with semantic information (such as "target confidence probability 85%" and "target is moving to the right"), which greatly enhances the system's situational understanding and decision support capabilities.

[0065] This invention achieves intuitive and interactive information presentation and full-cycle traceability of the detection process by "presenting graphical markers and semantic text descriptions in conjunction" and "saving text records to provide historical backtracking." Users can not only obtain rich information with complementary graphics and text in real time, but also review the detection status at any point in history at any time, significantly improving the ease of use, reliability, and auditability of the operation. Attached Figure Description

[0066] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a flowchart of the life detection information presentation method based on radar detection and depth analysis in an embodiment of the present invention;

[0068] Figure 2 This is a dot pattern map based on the traditional CFAR algorithm under the "one moving and five static" test scenario in this embodiment of the invention;

[0069] Figure 3 This is a dot pattern map based on a deep learning model under the "one moving and five static" test scenario in this embodiment of the invention;

[0070] Figure 4 This is a dot pattern map based on the traditional CFAR algorithm in a complex scenario of "nine people stationary and the target moving slightly" in an embodiment of the present invention.

[0071] Figure 5 This is a dot pattern map based on a deep learning model in the complex scenario of "nine people stationary and the target moving slightly" in this embodiment of the invention.

[0072] Figure 6 This is a flowchart of static target depth analysis in an embodiment of the present invention;

[0073] Figure 7 This is a schematic diagram of the display and control interface of a static target during the detection process in an embodiment of the present invention;

[0074] Figure 8 This is a schematic diagram of the display and control interface of a static target after the detection is completed in an embodiment of the present invention;

[0075] Figure 9 This is a flowchart of dynamic target depth analysis in an embodiment of the present invention;

[0076] Figure 10 This is a schematic diagram of the dynamic target display and control interface during the detection process in an embodiment of the present invention;

[0077] Figure 11 This is a schematic diagram of the dynamic target display and control interface during the directional movement in an embodiment of the present invention;

[0078] Figure 12 This is a schematic diagram of the dynamic target display and control interface after the detection is completed in an embodiment of the present invention. Detailed Implementation

[0079] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0080] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0081] Example 1

[0082] This invention provides a method for presenting life detection information based on radar detection and depth analysis. Its core lies in transforming the original echo signal of through-wall radar into intuitive, reliable and traceable intelligent situational information through deep learning, deep analysis, multimodal presentation and historical backtracking.

[0083] Figure 1 A flowchart illustrating an overall method provided by an embodiment of the present invention is shown. Figure 1 As shown, the method for presenting life detection information includes the following steps:

[0084] Step S1: Process the raw echo signal from the through-wall radar to generate radar imaging results.

[0085] In this embodiment, a three-dimensional through-wall radar is used as the detection device. The radar emits electromagnetic waves and receives echo signals from the area behind the wall, obtaining the raw echo signal. The signal processing module preprocesses the raw echo signal, including filtering and pulse compression, and then performs two-dimensional imaging processing. The imaging processing generates two images with complementary characteristics: an amplitude image and a decibel image (DB image). The amplitude image has lower background noise and can better distinguish between stationary and moving targets, but weak targets are easily lost; the DB image can better preserve weak targets, but the background noise is more obvious.

[0086] Step S2: Perform image enhancement processing on the amplitude image in the radar imaging results, and perform target detection based on the enhanced image using a deep learning model.

[0087] Step S2 aims to achieve stable and accurate detection of life targets from radar imaging through innovative image preprocessing and advanced deep learning models.

[0088] S2.1: Image enhancement processing based on gamma transform.

[0089] In this embodiment, the amplitude image generated in step S1 is selected for image enhancement processing. Amplitude images have the advantage of low background noise, but their visibility of weak targets (especially stationary or slightly moving targets) is poor. To address the technical problem of balancing target visibility and background noise in through-wall radar imaging, instead of directly using the original amplitude image or decibel map, a gamma transform-based image enhancement algorithm is creatively employed to process the amplitude image.

[0090] The formula for the gamma transform is defined as follows:

[0091] (1)

[0092] in, The normalized grayscale value of the input pixel, with a value range of [0,1]; The output grayscale value after gamma transformation; This is the grayscale scaling factor, which is a constant value of 1 in this embodiment; The gamma factor is a key parameter that controls the scaling of the entire transformation.

[0093] Through experimental optimization, a suitable one was selected. Values ​​(e.g., based on experimental data) The gamma transform of formula (1) can be applied to each pixel of the entire amplitude image (which can be optimized between 0.5 and 2.0). After this transformation, the regions with insufficient contrast (such as weak target regions) in the original amplitude image are enhanced, while the changes in background noise regions are relatively gradual. This process generates a new, higher-quality intermediate target image. The intermediate target image combines the advantages of a clean background in the amplitude image and the sensitivity of the decibel map to weak targets, providing input data with more distinct features and better noise suppression for subsequent deep learning models. This improves the performance ceiling of the detection task from the data source and effectively balances the precision and recall of detection.

[0094] S2.2: Object detection based on deep learning models.

[0095] The intermediate target image generated in step S2.1 is input into a pre-trained deep learning model for inference. In this embodiment, the YOLOv5 network is selected as the deep learning model, which adopts a single-stage detection framework and simultaneously completes feature extraction, target localization, and classification through an end-to-end convolutional neural network.

[0096] A typical YOLOv5 network consists of an input layer, a backbone, a feature fusion layer, and a head. The backbone is responsible for extracting multi-scale features from the image, effectively capturing the spatial information of the target using multiple convolutional layers and residual connections. The feature fusion layer integrates feature maps from different scales, providing rich contextual information for the head. The head further utilizes a pre-defined anchor box mechanism to generate candidate boxes in different regions of the feature maps, and simultaneously optimizes target classification, bounding box regression, and confidence scoring through a multi-task loss function.

[0097] Before deploying deep learning models, the YOLOv5 network needs to be trained using a large-scale, well-annotated through-wall radar imaging dataset (the expected output of each sample in the dataset, obtained by setting UWB labels on the target to obtain its true location). Through iterative optimization, the network can learn the unique features of living targets in 3D through-wall radar images (such as shape, texture, contrast relationship with the background, etc.), thus possessing strong generalization ability and robustness.

[0098] After training, the deep learning model is deployed on the display and control system. For each intermediate target image frame input in real time, the model performs forward inference and outputs the initial detection box in the image pixel coordinate system and its corresponding class confidence score. Each detection box is marked as a rectangle to indicate the suspected target region, and the confidence score reflects the model's confidence in the presence of a target in that region.

[0099] S2.3: Coordinate transformation and output of detection results.

[0100] The coordinates of the initial detection box center point in the image coordinate system output by the model are transformed into the real world coordinate system (e.g., a Cartesian coordinate system with the radar as the origin) through a pre-calibrated coordinate mapping relationship (usually a perspective transformation matrix or a polynomial fitting model) to obtain the actual position information of the target (e.g., X, Y coordinates in meters).

[0101] At this point, step S2 completes the intelligent detection process from raw radar imaging to the target's actual location. Subsequent steps will build upon this foundation, incorporating time-series information for further optimization and in-depth analysis.

[0102] Experimental Verification: In comparative tests, the method of this invention demonstrates significant advantages over traditional algorithms such as CFAR. For example, in a comparative experiment of a "one moving, five stationary" test scenario, both the method of this invention and traditional CFAR identified all targets, such as... Figure 2 and Figure 3 As shown. By comprehensively analyzing the features of the dot pattern map and real-time monitoring data from the software interface, the method of this invention has a more stable detection process, faster target detection response, and better trajectory continuity and consistency (e.g. Figure 3 As shown, blue dots represent static targets, and red dots represent dynamic targets. In a complex scenario where nine people are stationary while the target moves slightly, comprehensive analysis of the dot pattern features and real-time monitoring data reveals that the method of this invention can detect all targets faster and more stably, and the accuracy and reliability of the judgment results are superior to traditional methods (e.g., ...). Figure 4 and Figure 5 (As shown).

[0103] Step S3: Combine temporal context information to optimize the target detection results and track the trajectory, and output the actual position information of the target with dynamic and static attributes.

[0104] S3.1: Establish and maintain the target state memory.

[0105] The system initializes a target state memory for continuous tracking and management of all detected targets. The target state memory maintains a state record for each confirmed target; this record is a data structure containing, but is not limited to, the following core fields:

[0106] Target unique identifier (ID): used to distinguish different targets;

[0107] Historical trajectory sequence: Stores the world coordinate system position (e.g., center point coordinates (x, y)) of the target in the past several frames (e.g., the first N=30 frames) in chronological order;

[0108] Target motion state vector: including position, velocity, etc., used as input for Kalman filtering.

[0109] Target static / dynamic attributes: The current state label of the target, derived from historical trajectory analysis, such as "stationary" or "moving".

[0110] Associated bounding box information, such as size and confidence history.

[0111] The target state memory is updated after each frame is processed, removing target records that have not been associated or confirmed for a long time (such as M=5 consecutive frames) and adding newly initialized target records.

[0112] S3.2: State prediction based on Kalman filter.

[0113] For each currently active target in the target state memory, the system uses a Kalman filter to predict its possible position in the current frame before starting to process a new frame of data.

[0114] For static targets, a constant position model is used for state prediction; for dynamic targets, a uniformly accelerated motion model is used. The choice of model can be adaptively determined based on the smoothness of the target's historical trajectory.

[0115] The optimal state (posterior estimate) of the target after updating in the previous frame is input into the prediction equation of the Kalman filter to calculate its predicted position in the current frame and the prediction uncertainty (covariance matrix). This predicted position represents the system's best estimate of its current position based on the target's historical behavior.

[0116] S3.3: Correlation and comparison between detection results and predicted locations.

[0117] The target detection result of the current frame output in step S2 (i.e., the actual position information after the initial detection box is converted) is correlated with the predicted positions of all targets in the target state memory, and their consistency is compared.

[0118] The nearest neighbor association algorithm is used to find the best-matching predicted location (i.e., the closest location within a preset association threshold) for each detection box. Successfully matched detection box-target pairs are marked as "associated".

[0119] Unmatched targets in the target state memory: Targets that exist in the target state memory but for which no matching detection box was found in the current frame. These targets are classified as "potentially lost targets," usually due to low instantaneous signal-to-noise ratio, occlusion, or missed detection of weak targets.

[0120] Unmatched detection boxes in the current frame: Detection boxes that are detected in the current frame but cannot match the predicted location of any target in the target state memory. These detection boxes may represent newly emerging targets or transient false detections.

[0121] S3.4: Intelligent corrective decision-making and execution based on comparison results.

[0122] Based on the results of the correlation comparison, the system executes a set of intelligent correction decision logic to generate a stable final detection box for each target.

[0123] For "associated" targets (i.e., targets detected in the current frame): calculate the Euclidean distance deviation between the current detection box position and the predicted position; if the deviation is less than or equal to a first preset threshold (e.g., 0.5 meters), the current target detection result is considered reliable and consistent with the prediction. Perform a selection operation, directly adopting the current detection box as the final detection box for the target.

[0124] If the deviation exceeds a first preset threshold, it is considered that the current detection may have a large error or the target may have undergone a sudden change. The system performs a fusion operation, which involves weightedly fusing the current detection box position with a compensated detection box generated based on the predicted position (usually centered at the predicted position, with the size taken as the historical average size of the target), and using the fusion result as the final detection box for the target. This operation can effectively smooth the trajectory and suppress instantaneous jumps.

[0125] For "potentially lost" targets (i.e., targets not detected in the current frame): the system determines that the target is temporarily lost due to signal-to-noise ratio fluctuations or other reasons. In this case, the system performs a compensation operation, directly using the compensated detection box generated in step S3.2 based on the target's historical trajectory prediction as its final detection box. This ensures that static or weak targets can still be "virtually" maintained during their brief disappearance, preventing tracking interruption.

[0126] For "newly appearing detection boxes": After filtering (such as checking whether their confidence is higher than the new target threshold), new target records are initialized in the memory for these detection boxes. Their initial position is the detection box position. The dynamic and static attributes are temporarily set as "newly appearing". Their final detection box is the initial detection box.

[0127] S3.5: Update the target state memory and determine its dynamic / static attributes.

[0128] Based on the final detection frame determined in step S3.4, update the state of the corresponding target in the target state memory.

[0129] For targets whose final detection boxes are obtained through selection, fusion, or compensation operations, the position information of the final detection boxes is used to correct their state vectors and uncertainties through the update equation of the Kalman filter.

[0130] Based on the updated trajectory sequence, the average displacement or velocity of the target within the most recent historical period (e.g., the last second) is calculated. If this value is below a preset dynamic / static threshold, the target is marked as static; otherwise, it is marked as dynamic. This attribute will be used as the basis for depth analysis.

[0131] S3.6: Output the optimized target information.

[0132] The world coordinates (i.e., the actual position information of the target) and its dynamic and static attributes corresponding to the final detection box of each target after time-series optimization are used as the output of step S3 for subsequent depth analysis processing.

[0133] By implementing step S3, the system effectively solves the technical problem of "unstable static target detection and easy tracking interruption." Experiments show that in low signal-to-noise ratio environments or multi-target intersection scenarios, the method of this invention can significantly reduce the target loss rate and improve the continuity and smoothness of the trajectory, such as... Figures 2 to 5 As shown (compared to the discontinuous points of traditional methods, the trajectory of the method of this invention is more coherent), thus providing a stable and reliable data foundation for subsequent in-depth analysis.

[0134] Step S4: Based on the actual location information of the target and its dynamic and static attributes, perform cluster analysis and confidence assessment on the static targets to obtain the static target clusters and their confidence probabilities.

[0135] Step S4 aims to intelligently summarize and quantify targets identified as static, addressing the reliability issue in determining the existence of static life targets. Cluster analysis integrates discrete detection points into spatially consistent target clusters, and an innovative confidence calculation model outputs a quantified reliability probability for each static target cluster. The specific implementation is as follows:

[0136] S4.1: Initialization and Data Filtering.

[0137] The system receives the actual target location information and its dynamic / static attributes from the output of step S3. First, based on the dynamic / static attributes, targets marked as "static" are selected from all targets. The coordinate sequences of these static targets will serve as the input data stream for step S4. The system initializes a static target cluster set, which is initially empty, to store and manage all formed static target clusters.

[0138] S4.2: Real-time clustering analysis based on distance threshold.

[0139] like Figure 6 As shown, for each newly detected static target, iterate through each existing static target cluster (Cluster_i) in the current static target cluster set, and calculate the first distance (Euclidean distance) between the position coordinates of the newly detected static target and the center position coordinates of the existing static target cluster (Cluster_i). The center position coordinates of the existing static target cluster (Cluster_i) are defined as the arithmetic mean of the coordinates of all existing target points within the existing static target cluster (Cluster_i).

[0140] If there exists a static target cluster Cluster_k such that the first distance is less than or equal to a second preset threshold (e.g., 0.5 meters), then the newly detected static target is considered to belong to the same entity as the static target cluster Cluster_k in space. The newly detected static target is added to the target point list of the static target cluster Cluster_k. Based on the newly added target point, the center position coordinates of the static target cluster Cluster_k are recalculated in real time, and the number of target points in the static target cluster Cluster_k is incremented by 1.

[0141] If the first distance between the location coordinates of a newly detected static target and the center location coordinates of all existing static target clusters is greater than a second preset threshold, then the newly detected static target is considered to represent a new potential static entity. A new static target cluster, Cluster_new, is created centered on the location coordinates of the newly detected static target. The target point list of Cluster_new is initialized, and its center location coordinates are set to the location coordinates of the newly detected static target. The number of target points is 1. The static target cluster Cluster_new is then added to the static target cluster set.

[0142] After completing the above clustering operations (whether adding or creating new clusters), the system immediately updates the display of static targets on the control interface using an icon + coordinates + color. Simultaneously, in the text information area of ​​the interface, a summary of the depth analysis of all current static target clusters is output in text form every second (or every fixed cycle), such as... Figure 7 As shown.

[0143] S4.3: Confidence assessment and output after the detection cycle ends.

[0144] When a complete detection cycle ends (e.g., when the user stops scanning or the preset detection duration is reached), the system performs a final statistical analysis on each cluster (Cluster_i) in the static target cluster set and calculates its confidence probability. The calculation of the confidence probability is mainly based on two factors:

[0145] Number of target points within a cluster: The more targets there are, the more times the location has been detected, and the greater the likelihood that it is a real target.

[0146] Spatial dispersion: For example, calculate the standard deviation of the distances from all target points within a cluster to the center coordinates. The lower the dispersion (the more concentrated the points), the better the detection consistency and the higher the confidence probability.

[0147] This embodiment uses a piecewise linear growth model to calculate the confidence probability. This model quantifies the nonlinear relationship between the number of target points within a cluster and the confidence probability.

[0148] If the number of target points within the static target cluster Cluster_i does not exceed a preset threshold (e.g., 10), i.e., the first interval, the confidence probability is equal to the product of the number of target points within the static target cluster Cluster_i and the first growth rate. In this embodiment, the first growth rate is set to 5%. Within the first interval, the confidence probability increases linearly at a rate of 5% for each additional detection, aiming to quickly establish initial confidence in potential targets.

[0149] If the number of target points within the static target cluster Cluster_i exceeds a preset threshold (e.g., 10), i.e., the second interval, the confidence probability = 50% + (number of target points within the static target cluster Cluster_i - preset threshold) × second growth rate. In this embodiment, the second growth rate is set to 2%. Within the second interval, based on the existing 50% base confidence probability, the confidence probability increases by only 2% for each additional probe. This reflects the principle of "diminishing returns," meaning that when the number of probes is sufficiently high, the marginal contribution of newly added points to "confirming" the target decreases, and the model tends to become robust.

[0150] For each static target cluster (Cluster_i), the system generates a final report, including the target identifier, center coordinates, and confidence probability. Simultaneously, on the graphical interface, the corresponding static target icon will display its confidence probability in the form of "icon + coordinates + color," or its visual style will change according to the confidence probability (e.g., color depth). Figure 8 As shown.

[0151] By implementing step S4, the system not only intelligently aggregates messy static points into meaningful "target areas," but more importantly, through a scientific segmented confidence model, it provides operators with a quantitative and reliable basis for judging the key question of "whether there is a stationary living organism in this location," greatly reducing false alarms from occasional noise and enhancing confidence in confirming the real target.

[0152] Step S5: Based on the actual location information of the target and its dynamic and static attributes, perform trajectory clustering and motion pattern recognition on the dynamic target to obtain its motion pattern.

[0153] Step S5 aims to intelligently track and understand the behavior of targets identified as dynamic, addressing the issues of maintaining the continuity of moving targets' trajectories and interpreting their behavioral intentions. A continuous and complete motion trajectory is constructed using a two-level trajectory clustering algorithm, and the trajectory morphology is analyzed based on second-level time windows, transforming the original coordinate points into high-level behavioral semantics. The specific implementation is as follows:

[0154] S5.1: Initialization and Data Filtering.

[0155] The system receives the target's actual location information and its dynamic / static attributes from the output of step S3. First, based on the dynamic / static attributes, targets marked as "dynamic" are filtered out from all targets. The coordinate sequences of these dynamic targets will serve as the input data stream for step S5. The system initializes a dynamic target trajectory set, which is used to store and manage all tracked dynamic target trajectory sequences. Each trajectory sequence is a list of location points ordered by timestamp and associated with a unique trajectory ID.

[0156] S5.2: Real-time trajectory clustering based on two-level distance matching.

[0157] like Figure 9 As shown, for each newly detected dynamic target, the last known position points of all dynamic targets at the previous time step (i.e., the tail points of each trajectory sequence) are traversed, and the second distance between the newly detected dynamic target and these tail points is calculated.

[0158] If there exists a tail point of a trajectory Traj_k such that the second distance is less than or equal to the third preset threshold (e.g., 0.8 meters), then the newly detected dynamic target is determined to be a continuation of the trajectory Traj_k and belongs to the same moving target. The newly detected dynamic target is added to the end of the trajectory Traj_k, and the tail point of the trajectory Traj_k is updated to the newly detected dynamic target.

[0159] If the second distance between the newly detected dynamic target and the tail point of each trajectory sequence at the previous time is greater than the third preset threshold, it indicates that the newly detected dynamic target cannot directly continue with any existing trajectory sequence. The system calculates the third distance between the newly detected dynamic target and the center position of all existing trajectory sequences in the dynamic target trajectory set (which can be approximated by the geometric center of the trajectory point set or the average of several nearest points).

[0160] If there exists a center position of a trajectory Traj_m such that the third distance is less than or equal to the fourth preset threshold (usually slightly greater than the third preset threshold, for example, 1.2 meters), then it is considered that the newly detected dynamic target, although discontinuous in time, belongs to the same activity area in space (it may be a target that has been briefly lost and then reappeared), and the newly detected dynamic target is added to the trajectory Traj_m.

[0161] If the third distance between the newly detected dynamic target and the center position of all existing trajectories is greater than the fourth preset threshold, then the newly detected dynamic target is considered to represent a brand new dynamic target or a brand new starting point of activity. Starting from the newly detected dynamic target, a new trajectory sequence is initialized and added to the dynamic target trajectory set.

[0162] After completing the clustering operation, the system immediately updates the display of dynamic targets on the control interface in the form of "icon + coordinates + color". Simultaneously, in the "Target Information" or "Depth Analysis in Progress..." area of ​​the control interface, the depth analysis information of the dynamic targets is output in scrolling text every second (or every fixed cycle), for example: the dynamic target is moving near coordinates (1.5, 5.7); the dynamic target moves from coordinates (2.1, 6.4) to coordinates (2.8, 6.4), etc. Figure 10 and Figure 11 As shown.

[0163] S5.3: Motion pattern recognition based on second-level time windows.

[0164] The system periodically (e.g., every second) performs behavior analysis on the trajectory of each active dynamic target. For trajectory Traj_i, all trajectory points within the most recent second-level time window are extracted. Let the position at the start of the window be P_start, and the position at the end of the window (i.e., the latest position) be P_end; calculate the displacement between the start position P_start and the end position P_end.

[0165] If the trajectory Traj_i contains only one observation point within the current time window (i.e., it has just been initialized or detected for the first time), then its motion mode is determined to be "newly emerging dynamic target".

[0166] If the displacement is less than the fifth preset threshold (e.g., 0.5 meters), the dynamic target is determined to be in a "local hovering or slight movement" state within that time window. This usually corresponds to the target adjusting its attitude within a small range or making slight movements in place.

[0167] If the displacement is greater than or equal to the fifth preset threshold and the movement directions are relatively consistent, then it is determined that the target has undergone "directional movement from the starting position to the ending position" within that time window. Figure 12 As shown, the output text description is "The dynamic target moves from coordinates (4.3, 6.4) to coordinates (4.2, 6.4)".

[0168] If the starting position P_start and the ending position P_end are very close (the distance is less than the fifth preset threshold), but analysis of the complete trajectory point sequence within this time window reveals that it forms an obvious non-linear, closed, or nearly closed contour, then it is suggested that the target "may exhibit circular motion behavior".

[0169] S5.4: Final motion mode output and display.

[0170] At the end of the detection cycle (e.g., when manually stopped by the user), the system provides a comprehensive summary of the behavior of all dynamic targets throughout the detection process. For example... Figure 12 As shown, in the "Deep Analysis Completed" area, the system outputs summary information, such as "Detection Results, Analysis Results: Dynamic Target Exists," and may list a summary of the main movement paths. Simultaneously, on the graphical interface, the trajectory lines of the dynamic target can be drawn with different colors or line types to visually distinguish different movement patterns (e.g., a solid red line indicates directional movement, and a dashed yellow line indicates loitering).

[0171] By implementing step S5, the system can not only robustly track moving targets and overcome the impact of temporary loss of tracking, but also transform the target's movement into a behavioral description with clear semantics. For example... Figures 10 to 12 As shown, users no longer just see moving points, but can read intuitive descriptions such as "moving near..." or "walking from..." in real time, which greatly enhances the depth of understanding and speed of judgment of the activity situation of people behind the wall, and provides direct and efficient information support for command and decision-making.

[0172] Step S6: The actual location information of the target, the static target cluster and its confidence probability, and the motion pattern of the moving target are presented in real time on the display and control interface in the form of graphical markers and semantic text descriptions; and the semantic text descriptions with timestamps are saved to provide backtracking of historical detection information.

[0173] Step S6 aims to present all the aforementioned processing results (actual target location, static confidence probability, and dynamic behavior pattern) in an intelligent and integrated manner, and provide complete traceability capability for the detection process, thereby solving the problems of fragmented and untraceable information in traditional radar display and control interfaces.

[0174] S6.1: Dynamic Creation and Association of Multimodal Information Units

[0175] Whenever step S4 or step S5 confirms and outputs a new static target cluster or dynamic target trajectory (or the target state changes significantly), the system synchronously performs the following operations on the display and control interface:

[0176] Graphical Marker Unit: On the 2D / 3D situation view (main view), a graphical marker is instantiated based on the target's actual position coordinates. Static targets are typically represented by specific icons (such as red dots or red human figures), while dynamic targets are represented by another type of icon (such as green dots). This marker is assigned a unique target identifier.

[0177] Semantic Text Description Unit: A new text entry is added to the "Target Information" list area in the sidebar or bottom of the interface. This entry is also bound to the aforementioned target identifier. Its content is dynamically generated based on the target attributes.

[0178] For static targets: Generate a description such as "static target, coordinates, confidence probability", e.g. Figure 8 As shown.

[0179] For dynamic targets: generate descriptions such as "dynamic target, motion pattern", etc. Figure 12 As shown.

[0180] Graphical marker units: As the actual location of the target is updated in real time (especially for dynamic targets), the position of its corresponding graphical marker on the situation view is updated smoothly and in real time. The content of the semantic text description units is also refreshed in real time. For dynamic targets, their coordinates and motion pattern descriptions are updated every second; for static targets, their confidence probability may be updated periodically as new points are added.

[0181] S6.2: Interactive linkage implementation.

[0182] To achieve deep association between graphics and text, the system establishes an interactive linkage mechanism based on target identifiers:

[0183] Graphical markers and text entries are tightly bound together in the background data model through a shared target identifier. When a user clicks on a graphic marker in the situation view, the system captures the target identifier of the event, then scrolls through the text list area to locate the text entry corresponding to the target identifier and highlights that line (e.g., changing the background color and bolding the font).

[0184] Conversely, when a user clicks on a target description in the text list, the system also obtains its target identifier, and then pans or zooms the view on the situation view to place the corresponding graphic mark in the center of the view or in a prominent position, and highlights the mark (such as by zooming in, flashing, or changing the border color).

[0185] This linkage function enables users to easily "find text from images" or "locate text from images" in complex multi-objective scenarios, greatly improving the efficiency of information retrieval and verification.

[0186] S6.3: Structured storage of historical exploration information.

[0187] During operation, the system continuously records historical data, laying the foundation for backtracking functionality. At fixed time intervals (e.g., per second) or whenever a new semantic text description is generated, the system packages the semantic text descriptions of all active targets at the current moment (i.e., the set of text entries generated in step S6.1) and associates them with a precise timestamp to form a historical record. This historical record is stored in a specially designed time-series database or a file with a time index. During storage, in addition to the text content and timestamp, the system can also associate and store the target identifiers and their simplified states for all targets at that moment, enabling rapid recovery.

[0188] S6.4: Implementation of history viewing and scene replay functions.

[0189] The display interface provides a separate "History Viewing Panel" or "Playback Control Area". In this panel, all stored historical records are presented in a list format in reverse or ascending chronological order, with each record displaying its timestamp and key summary.

[0190] When a user selects a historical record in the history view panel, the system retrieves the corresponding complete semantic description set from storage based on the record's timestamp. The system replaces the text description of that historical moment in the "Target Information" list area, clears the current real-time display in the situation view, and redraws (restores) the graphical marker distribution of all targets at that moment based on the retrieved historical data. This is equivalent to generating a snapshot of the scene at that moment, which includes not only the target locations (graphical markers) but also detailed text descriptions of each target at that time, accessible through linked functions. Users can sequentially select records from different time points in the history panel to "replay" the detection process frame by frame or jump to key nodes, thereby reviewing, analyzing, and verifying the entire detection process.

[0191] Through the implementation of step S6, the system constructs an integrated intelligent display and control environment that combines real-time perception and historical insight. Users can not only grasp the current situation behind the wall in real time, but also "go back in time" to review the detection details of any key moment. For example... Figures 10 to 12 As shown in the interface evolution from "Deep analysis in progress..." to "Deep analysis complete", the final result is a complete detection report that is rich in information, well-organized, interactive, and auditable. This completely changes the limitations of traditional radar software, which presents information in a simple way and has no traceability of the process. It significantly improves the practical value and credibility of the system in key tasks such as emergency rescue and security monitoring.

[0192] Example 2

[0193] This invention also provides an electronic device, which includes a memory, a processor, and a computer program or instructions stored in the memory. The processor executes the computer program or instructions to implement the life detection information presentation method based on radar detection and depth analysis in this invention.

[0194] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may contain multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for device operation are also stored in RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0195] The processor and memory described above are used together to execute programs / instructions stored in the memory. When the program / instructions are executed by the computer, they can implement the methods, steps, or functions described in the above embodiments.

[0196] Although not shown, embodiments of the present invention also provide a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the method for presenting life detection information based on radar detection and depth analysis in embodiments of the present invention.

[0197] Readable storage media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0198] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for presenting life detection information based on radar detection and depth analysis, characterized in that, The method includes: The raw echo signal of the through-wall radar is processed to generate radar imaging results; the amplitude image in the radar imaging results is enhanced, and target detection is performed based on the enhanced image using a deep learning model; the target detection results are optimized and tracked by combining temporal context information, and the actual position information of the target with dynamic and static attributes is output. Based on the actual location information and dynamic / static attributes of the target, cluster analysis and confidence assessment are performed on static targets to obtain static target clusters and their confidence probabilities, and trajectory clustering and motion pattern recognition are performed on dynamic targets to obtain their motion patterns. The actual location information of the target, the static target cluster and its confidence probability, and the motion pattern of the moving target are presented in real time on the display and control interface in the form of graphical markers and semantic text descriptions; and the semantic text descriptions with timestamps are saved to provide backtracking of historical detection information.

2. The method for presenting life detection information based on radar detection and depth analysis according to claim 1, characterized in that, The amplitude image in the radar imaging result is subjected to image enhancement processing, specifically: the amplitude image is processed using an image enhancement algorithm based on gamma transform to generate an intermediate target image.

3. The method for presenting life detection information based on radar detection and depth analysis according to claim 1, characterized in that, Optimize target detection results and track trajectories by incorporating temporal context information, including: Based on the historical trajectory information recorded in the target state memory, the Kalman filter is used to predict the target's state in the current frame, thus obtaining the predicted position of the target. When a target is not detected in the current frame, a compensation operation is performed, that is, the compensated detection box generated based on the predicted position is used as the final detection box of the target; When a target is detected in the current frame, calculate the deviation between the current detection box position and the predicted position; If the deviation is less than or equal to the first preset threshold, a selection operation is performed, that is, the current detection box is used as the final detection box of the target. If the deviation is greater than the first preset threshold, a fusion operation is performed, that is, the current detection box is fused with the compensated detection box generated based on the predicted position, and the fusion result is used as the final detection box of the target. The target state memory is updated based on the final detection frame.

4. The method for presenting life detection information based on radar detection and depth analysis according to claim 1, characterized in that, Cluster analysis and confidence assessment of static targets include: Calculate the first distance between the newly detected static target and the center position of the existing static target cluster. If the first distance is not greater than the second preset threshold, the newly detected static target is assigned to the corresponding static target cluster. Otherwise, a new static target cluster is created with the newly detected static target as the center. After the detection period ends, the confidence probability of the real static target corresponding to the static target cluster is calculated based on the number of targets and the dispersion of their spatial distribution within the static target cluster. The confidence probability is positively correlated with the number of targets within the cluster in a segmented manner. Within a first interval where the number of targets within the cluster does not exceed a preset threshold, the confidence probability increases with a first growth rate as the number of targets within the cluster increases. Within a second interval where the number of targets within the cluster exceeds the preset threshold, the confidence probability increases with a second growth rate as the number of targets within the cluster increases, and the second growth rate is lower than the first growth rate.

5. The method for presenting life detection information based on radar detection and depth analysis according to any one of claims 1 to 4, characterized in that, Trajectory clustering and motion pattern recognition for dynamic targets include trajectory clustering steps and motion pattern recognition steps; wherein, the trajectory clustering step is used to associate observation points belonging to the same moving target in consecutive frames as the same trajectory; the motion pattern recognition step is used to identify typical motion behaviors of dynamic targets based on the positional changes of the trajectory within a second-level time window.

6. The method for presenting life detection information based on radar detection and depth analysis according to claim 5, characterized in that, The trajectory clustering steps include: For a newly detected dynamic target, calculate its second distance to all dynamic targets in the previous time step; If the second distance is not greater than the third preset threshold, it is determined to be a continuation of the trajectory of the same moving target, and the position information of the newly detected dynamic target is added to the trajectory sequence of the corresponding dynamic target at the previous moment; If the second distance is greater than the third preset threshold, then calculate the third distance between the newly detected dynamic target and the center position of each existing trajectory sequence; If the third distance is not greater than the fourth preset threshold, the location information of the newly detected dynamic target is added to the corresponding existing trajectory sequence; If the third distance is greater than the fourth preset threshold, then a new trajectory sequence is initialized starting from the newly detected dynamic target.

7. The method for presenting life detection information based on radar detection and depth analysis according to claim 5, characterized in that, The motion pattern recognition steps include: Based on the trajectory sequence of a dynamic target, the position of the dynamic target at the start and end times is obtained within a second-level time window; Based on the positional changes between the start and end times, at least one of the following motion patterns is identified: If the corresponding trajectory sequence contains only one observation point, it is determined to be a newly emerging dynamic target; If the displacement of the position change is less than the fifth preset threshold, it is determined to be local lingering or micro-movement; If the displacement of the position change is greater than or equal to the fifth preset threshold, it is determined to be a directional movement from the starting position to the ending position. If the starting and ending times are close in position and their trajectories form a closed profile, it is determined that there may be circular motion behavior.

8. The method for presenting life detection information based on radar detection and depth analysis according to claim 1, characterized in that, The information is presented in real-time on the display interface in a related manner, using both graphical markers and semantic text descriptions, including: For each detected target, a corresponding graphical marker unit and a semantic text description unit are generated and maintained on the display and control interface; The graphical marking unit updates its display position in real time based on the actual position information of the target, and the semantic text description unit updates its description content in real time based on the attributes of the target; wherein, for static targets, the description content includes at least its position and confidence probability information; for dynamic targets, the description content includes at least its position and current motion mode information. Furthermore, the graphical marker unit and the semantic text description unit share the same target identifier; in response to the user's selection operation on any of the units, the system locates and highlights the corresponding other unit on the display interface based on the target identifier; The storage of the semantic text description with timestamps to provide backtracking of historical exploration information includes: Semantic text descriptions generated at each moment are stored as historical records in chronological order of detection time, associated with the corresponding timestamps. The display and control interface provides a history viewing panel that presents the historical records in chronological order. The history viewing panel supports user interaction. When a user selects a historical record, the display and control interface is triggered to synchronously display snapshot information of the historical detection scene corresponding to the timestamp of the record. The snapshot information includes at least the distribution of the target graphical markers at the corresponding time.

9. An electronic device comprising a memory, a processor, and a computer program or instructions stored in the memory, characterized in that, The processor executes the computer program or instructions to implement the life detection information presentation method based on radar detection and depth analysis as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they implement the life detection information presentation method based on radar detection and depth analysis as described in any one of claims 1 to 8.

Images