A rescue smart helmet vital sign positioning system

Abstract

The application discloses a rescue intelligent helmet sign positioning system and relates to the technical field of intelligent helmet sign positioning. The system replaces the whole uniform trimming mode with the local re-estimation envelope, jointly incorporates the crack interval and the front and rear credible segments into the local solving boundary, and executes coupling arrangement and linkage convergence on multiple envelopes which are time-continuous and space-adjacent, so that the adjacent cracks are no longer treated separately as abnormal points, and the left and right swing and boundary fragmentation in the path revision process are inhibited. After the local re-estimation, the envelope revision track and the segment-level credibility backfill table are written back to the task microfilm generation link and the mirror state generation link through the back-stable memory film backfill module, the deep feature explanation boundary, the scene restraint trigger boundary, the state candidate boundary, the action template switching boundary and the state chain continuation boundary of the subsequent segment are redefined, and the subsequent interpretation continues to develop along the revised track chain and the state chain.

Description

Technical Field

[0001] This invention relates to the field of intelligent helmet vital sign positioning technology, specifically to a rescue intelligent helmet vital sign positioning system. Background Technology

[0002] Existing rescue helmets or accompanying positioning systems mostly rely on inertial navigation trajectories for position estimation and use heart rate, respiration, voice, or posture data for independent status monitoring. There is a lack of unified segment-level organizational relationships between the position link and the vital signs link, as well as a lack of collaborative interpretation mechanisms around the same task process. When the positioning results drift, existing solutions usually only perform smoothing, filtering, or local correction within the trajectory domain, making it difficult to use head movements, breathing rhythms, heart rate swings, and scene constraint states as evidence to refute the authenticity of the trajectory. At the same time, for complex scenarios such as low clearance, narrow passages, close-to-wall passage, and continuous turning back, existing solutions often use fixed action templates and isolated segment judgment methods, making it difficult to form a complete closed loop covering task micro-screen generation, mirror state determination, crack scanning, local reassessment, and result feedback.

[0003] The aforementioned shortcomings arise from two main reasons. First, the strong asynchronicity of multi-source data in terms of sampling frequency, time boundaries, and action rhythms makes it difficult to establish stable correspondences between changes in physical characteristics, posture, and displacement at the same event inflection point. Second, the subsequent processing relies on single judgment results for a long time, lacking continuous constraints on the continuous relationship between adjacent segments, scene constraint boundaries, coupling relationship between adjacent envelopes, and the feedback relationship of revision results.

[0004] This can easily lead to several abnormal effects in actual rescue operations, including distorted task segmentation boundaries, misjudgment of motion contraction in restricted scenarios as trajectory anomalies, path oscillation caused by the dispersed processing of adjacent cracks in the same area, inability of local revision results to be passed on to subsequent judgment stages, and the continued use of the interpretation boundary already affected by drift in subsequent state judgments, ultimately causing a disconnect between personnel position determination, state recognition, and path revision. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a rescue intelligent helmet vital sign positioning system, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a rescue intelligent helmet vital sign positioning system, including an asynchronous time fold alignment module, a mirror state generation module, a trajectory crack scanning module, a local envelope re-evaluation module, and a stabilization memory film recharge module; The asynchronous time-fold alignment module acquires the vital sign positioning data and the reversal point of the breathing valley of the smart helmet to construct event stakes, and generates a sequence of task micro-screens based on the data interval between adjacent event stakes; The mirror state generation module generates physiological-behavioral mirror state cards for each task micro-screen based on the deep features within the task micro-screen sequence, scene constraint indicator groups, and the continuity relationship between adjacent segments. The trajectory crack scanning module compares the physiological-mirror state card with the inertial navigation trajectory segment by segment, extracts suspicious path segments that cannot be interpreted by the mirror state card, and forms a trajectory crack table. The local envelope re-estimation module uses the suspicious path segments marked in the trajectory crack table as the scope, combines the reliable segments before and after the crack to construct a local re-estimation envelope, and generates an envelope revision trajectory and a segment-level reliability backfill table within the local re-estimation envelope. The stabilization memory membrane recharge module writes the envelope revision trajectory and fragment-level credibility backfill table into the stabilization memory membrane and recharges it into the task micro-screen generation stage and the mirror state generation stage, updating the deep feature interpretation boundary and state determination boundary of subsequent fragments.

[0007] Preferably, the asynchronous time-fold alignment module includes a stake point extraction and merging unit and an interval slice arrangement unit; The pile extraction and merging unit acquires the inertial navigation displacement increment sequence, attitude sequence, heartbeat interval sequence, respiratory rhythm sequence and voice activity sequence output by the smart helmet, and maps each sequence to the same time reference for alignment processing. Gait discontinuity points are identified based on inertial navigation displacement increments and attitude changes; Identify the head turning inflection point based on changes in head yaw and pitch; Identify the start and end boundaries of speech based on the rise and fall boundaries of speech activity energy; Identify the reversal point of the respiratory trough based on the change in the direction of the respiratory rhythm trough; The identified candidate points are sorted according to time sequence. Candidate points in the same short time neighborhood are merged into the same cluster. Isolated candidate points that only appear in a single sequence and do not meet the neighborhood continuity condition are removed. Candidate points that fall into the same action transition interval across sequences are combined into the same event stump. Time location identifier, source identifier, and adjacency relationship identifier are written for each event stump to obtain an ordered set of event stumps.

[0008] Preferably, the interval slicing unit reads the ordered set of event stakes and uses the time interval between adjacent event stakes as the original slice boundary to extract the corresponding data segments of the inertial navigation displacement increment sequence, attitude sequence, heartbeat interval sequence, respiratory rhythm sequence and speech activity sequence within the time interval to form candidate task micro-screens. Perform interval legality checks on each candidate task micro-screen, including interval span checks, sequence coverage checks, and boundary coherence checks; When the span of the candidate task micro-screen is less than the preset lower limit and the adjacent intervals have continuous action directions, the current interval is spliced ​​with the previous or next interval. When the span of the candidate task micro-screen is greater than the preset upper limit, a secondary segmentation point is inserted within the current interval based on the displacement turning point, posture change, or local reversal of breathing rhythm, and then the current interval is re-segmented. When a sequence has a missing segment in the current interval, write a missing segment identifier for the candidate task micro-screen and retain the remaining sequence data; After all candidate task micro-screens have been spliced, re-segmented, and segmented, the task micro-screen sequence is output in chronological order.

[0009] Preferably, the mirror state generation module includes a candidate state construction and switching unit and a continuous verification card generation unit; The candidate state construction and switching unit extracts deep features and scene constraint indication groups for each task micro-screen sequence; Deep features include one or more of the following: head stationary wake width, respiratory trough distance viscous plaque, heartbeat fall-off tail, displacement-respiration decoupling residual pattern, and head position-displacement deviation characteristics. The scenario restraint instruction group includes one or more of the following: low headroom collision symptoms, lateral head-turning and width-reduction symptoms, continuous crouching passage symptoms, and forward wall-hugging passage symptoms; Based on the combination form, occurrence order and co-occurrence relationship of each deep feature in the current task micro-screen, construct the candidate state set corresponding to the task micro-screen; The candidate state set includes at least one of the following: horizontal movement, short-term observation, continuous scanning and turning, partial turning back, vertical ascent, obstructed adjustment, and low-posture traversal; and it checks whether the current task micro-screen meets the scene-constrained action template switching conditions; When at least two signs in the scene constraint indication group appear consecutively in the current task micro-screen and the adjacent task micro-screen, the current task micro-screen is switched from the regular action template branch to the scene constraint action template branch, and the head scan arc interpretation boundary, the stationary wake interpretation boundary, and the head position-displacement deviation interpretation boundary are rewritten simultaneously to generate the state candidate record corresponding to the current task micro-screen.

[0010] Preferably, the continuous verification card unit is based on the state candidate record, and combines the continuity relationship between the current task micro-screen and the adjacent task micro-screen to perform continuous verification, compatibility screening and target state freezing processing on each candidate state in sequence; Continuity relationships include the direction of action continuation, the continuation of bodily signs, the continuation of head behavior, and the continuation of scene constraints; First, determine whether the candidate states of the current task's micro-screen meet the continuous transition conditions with the frozen state of the previous task's micro-screen. Then, determine whether the state trend of the next task's micro-screen remains consistent. When a candidate state simultaneously satisfies the requirement of continuous transition between the preceding and following segments, the candidate state is retained as the priority state. If a candidate state of the current task micro-screen conflicts with the state chain between adjacent task micro-screens, or does not satisfy at least one of the continuous relationships, then the candidate state is downgraded or eliminated. After the screening is completed, the state that is most coherent with the preceding and following segments is determined from the remaining candidate states and used as the target state for the current task's micro-screen. After determining the target state, a physiological-behavioral mirror state card corresponding to the current task micro-screen is generated.

[0011] Physiological-behavioral mirror state cards should include at least the state to which the segment belongs, the interpretable range of head movements, the interpretable range of respiratory rhythms, the interpretable range of heartbeat swings, the coherence condition of adjacent segments, and the template source marker. Specifically, when the current task micro-screen uses a scene-constrained action template branch, the template source marker is used to indicate that the subsequent trajectory crack scanning process calls the scene-constrained crack judgment boundary for the task micro-screen instead of calling the regular head scanning judgment boundary. Preferably, the trajectory crack scanning module includes a fragment comparison and discrimination unit and a crack aggregation and tabulation unit; The segment comparison and discrimination unit extracts displacement change segments, turning change segments, and vertical change segments that are consistent with the time interval of the task micro-screen film from the inertial navigation trajectory. Then, the displacement change segments, turning change segments, and vertical change segments are matched and reviewed with the segment status, head movement interpretation range, respiratory rhythm interpretation range, heartbeat swing interpretation range, adjacent segment coherence conditions, and template source markers in the corresponding physiological-behavioral mirror status card. After completing the matching review, select the corresponding difference boundary according to the template source mark; When the template source marker indicates that the current task micro-screen uses a conventional action template, conventional head scanning behavior, vital sign swinging behavior, and displacement continuation behavior are used as the primary comparison criteria. When the template source marker indicates that the current task micro-screen film uses a scene-constrained action template, the primary comparison criteria are limited observation instead of symptoms, scene constraint continuity, and displacement continuity, to avoid directly judging head movement contraction under low headroom, narrow passage, or wall-hugging conditions as trajectory abnormality. After the comparison is completed, the trajectory changes that cannot be interpreted by the corresponding mirror state card are marked as difference segments; they are marked as velocity-stationary misalignment candidate segments, steering noise candidate segments, or vertical detuning candidate segments, and (the segment number, candidate type, difference source, template source interval, and initial connection relationship with the preceding and following segments are recorded) to form a candidate crack record set.

[0012] Among them, when the inertial navigation trajectory shows a large translational change, and the corresponding mirror status card indicates that the current mission micro-screen is in a short-term observation, static transition, or restricted observation state, and the head behavior and breathing behavior do not give a support relationship that matches the translational change, the segment is marked as a velocity-stationary misalignment candidate segment. When the inertial navigation trajectory shows continuous azimuth changes, and the corresponding mirror state card does not provide a head steering support relationship that matches the azimuth change, the segment is marked as a candidate segment for steering noise. When the inertial navigation trajectory exhibits a vertical displacement change, and the vertical motion interpretation boundary, attitude continuity relationship, and symmetric swing relationship in the corresponding mirror state card cannot jointly support the vertical displacement change, the segment is marked as a candidate segment for vertical detuning. Preferably, the crack aggregation and table compilation unit rearranges the candidate crack records in chronological order according to the fragment number, and determines whether adjacent candidate crack records meet the aggregation conditions of temporal continuity, near continuity, correlation of trajectory change direction, and consistency of local intervals. When two or more candidate crack records meet the conditions, they are merged into the same crack segment. When adjacent candidate crack records, although of different types, point to the same chain of conflicting interpretations, they are associated and grouped. After merging, spatial adjacency checks are performed on each crack segment (the checks include spatial proximity between crack segments, sharing of reliable segments before and after, and common boundary intervals); and adjacent crack adjacency relationship identifiers are written for crack segments that meet the adjacency conditions. Based on the number of continuous segments, candidate type composition, template source interval, connection tension, and adjacent crack adjacency relationship of crack segments, each crack segment is sorted by level, and the crack start and end segment number, crack type, crack spanning segment duration, template source interval, connection tension identifier, spatial adjacency identifier, and adjacent crack adjacency relationship identifier are written, and the trajectory crack table is output.

[0013] Preferably, the partial envelope re-estimation module includes an envelope extraction and coupling unit and a branch re-estimation and backfilling unit; The envelope extraction coupling unit uses the start and end segment range corresponding to each crack segment as the local re-estimation center interval, and then extracts the continuous trajectory segments that are directly connected and not marked as crack segments from the front and back sides of the center interval as the reliable segments before and after the crack; this is used to limit the front and back boundaries of the local re-estimation envelope. The central interval, the front trusted segment, and the rear trusted segment are combined to form a local reassessment envelope; and each local reassessment envelope is written with an envelope number, an envelope boundary segment number, a central crack type, a corresponding template source interval, and a trusted anchor segment location identifier. Based on the adjacency relationship of adjacent cracks in the trajectory crack table, multiple local revaluation envelopes are coupled and arranged. For locally re-estimated envelopes that are temporally continuous, spatially adjacent, and share the same preceding and following reliable segments or common boundary intervals, they are merged into the same adjacent envelope coupling subdomain; and envelope boundary inheritance relationship, shared anchor segment relationship, and swing direction suppression relationship are written into the adjacent envelope coupling subdomain; the envelope boundary inheritance relationship is used to ensure that the solution starting boundary of the next envelope inherits the converged boundary of the previous envelope, the shared anchor segment relationship is used to ensure that the reliable segments shared by multiple envelopes maintain a unified reference, and the swing direction suppression relationship is used to restrict adjacent envelopes from alternating between left and right candidate paths; Perform envelope validity processing on each local revaluation envelope. The processing includes the integrity of the central crack interval, the continuity of the preceding and following reliable segments, the consistency of the template source boundary, and the attribution relationship of the coupled subdomain. After processing, output the set of local revaluation envelopes and their corresponding envelope boundary parameters, reliable anchor segment parameters, and coupled subdomain parameters.

[0014] Preferably, the branch re-estimation backfill unit constructs multiple candidate path branches on the original inertial navigation trajectory based on the central crack type, template source interval, credible anchor segment location, and coupling subdomain parameters corresponding to the local re-estimation envelope, and matches the candidate path branches with the physiological-behavioral mirror state cards of the corresponding segments respectively; During the matching process, static stability repulsion constraints are applied to short-term observation or statically stable transition segments, steering lock constraints are applied to continuous scanning and turning segments, stabilization closure constraints are applied to vertical displacement segments, and scene constraint boundary constraints are applied to segments using scene-constrained action templates. By applying constraints, each candidate path branch is screened and reduced segment by segment, and candidate path branches that do not meet the interpretation boundary of the mirror state card and the connection conditions of the trusted anchor segment are removed from the re-estimation sequence. After the candidate path branches are reduced, the remaining candidate path branches are subjected to envelope-based orbit determination. For locally re-evaluated envelopes located within the coupling subdomains of adjacent envelopes, the candidate path branches in multiple adjacent envelopes are converged in a coordinated manner by combining the envelope boundary inheritance relationship, the shared anchor segment relationship, and the sway suppression relationship, so that the revision trajectory between adjacent envelopes extends continuously along the same reference direction and maintains the consistency of the common boundary interval. From the retained candidate path branches, select the path branch that has the most stable connection with the preceding and following credible segments, the fewest conflicts with the physiological-behavioral mirror state card, and the smallest deviation from the orientation of adjacent envelopes within the coupled subdomain as the envelope revision trajectory; after determining the envelope revision trajectory, generate a segment-level credibility backfill table according to the task micro-screen number.

[0015] The fragment-level credibility backfill table should include at least the fragment number, the envelope number, the crack type before revision, the path branch number after revision, the credibility level after revision, the template source range, the credibility anchor inheritance identifier, the envelope coupling identifier, and the swing suppression status. Preferably, the stabilization memory film recharge module includes an in-membrane writing cataloging unit and a boundary recharge update unit; The cataloging unit written into the membrane performs fragmented expansion processing on the cover revision trajectory, maps the revised trajectory results to the corresponding task micro-screen segment interval, and then matches the information segments in the segment-level credibility backfill table with the fragmented revision trajectory one by one to form a revision record for single segments. The information fragment includes the fragment number, the envelope number, the crack type before revision, the path branch number after revision, the credibility level after revision, the template source range, the credibility anchor segment inheritance identifier, the envelope coupling identifier, and the swing suppression status. Based on the sequence of the mission micro-episodes, the position of the envelope boundary, and the affiliation of the coupled subdomains, each revision record is cataloged and organized to ensure that the revision records within the same envelope are arranged continuously, that the revision records within the coupled subdomains of adjacent envelopes are kept to the same lineage, and that they are written into the stabilization memory film. During the writing process of the stabilizing memory membrane, an intra-membrane index relationship is generated, which includes a fragment index, an envelope index, a template source index, and a coupling lineage index. Among them, the fragment index is used to directly associate each revision record with the corresponding task micro-screen, the envelope index is used to identify the local re-evaluation envelope to which each revision record belongs, the template source index is used to identify the action template boundary interval of the corresponding fragment before and after revision, and the coupling genealogy index is used to identify the continuous envelope relationship after the coupling and swaying processing of adjacent envelopes. Through write processing, a structured in-membrane revision record set organized by fragment, envelope, template boundary, and envelope lineage is formed within the stabilizing memory membrane; Boundary smoothing is performed on the structured intra-membrane revision record set. Boundary smoothing includes smoothing the corresponding boundaries between the revision trajectory and the original trajectory, smoothing the continuous boundaries of the segment-level confidence level, smoothing the connection between the template source intervals, and smoothing the continuation of the envelope lineage boundaries. After smoothing, an intra-membrane re-injection dataset with segment positioning relationships, template boundary relationships, and envelope lineage relationships is output. The boundary re-injection update unit redefines the boundaries of the displacement-respiration decoupling residual pattern interpretation interval, head stationary wake interpretation interval, respiratory valley distance viscous speckle interpretation interval, and heartbeat fallback trail interpretation interval corresponding to the subsequent task micro-screen images based on the revision trajectory, fragment index, and confidence level in the intramembrane re-injection dataset. It also synchronously corrects the trigger intervals of low clearance collision sign, lateral head swing amplitude reduction sign, continuous bending passage sign, and forward wall-hugging passage sign by combining the template source index and envelope genealogy relationship. This allows the subsequent task micro-screen images to directly inherit the revised fragment boundaries and feature interpretation boundaries during segmentation and feature annotation. After the correction is completed, based on the data in the intramembrane reinfusion dataset, the state candidate boundary, action template switching boundary and state chain continuation boundary of the subsequent task micro-screen are reset, so that the mirror state generation link will preferentially inherit the revised trusted behavior chain and template source boundary when generating the subsequent physiological-behavioral mirror state card. The data in the intramembrane reinjection dataset includes template source intervals, revised path branch numbers, trusted anchor segment inheritance identifiers, envelope coupling identifiers, and swing suppression status. For segments within scene-constrained action template branches, maintain the extension relationship of their template continuity intervals; For segments within the influence range of adjacent envelope coupling subdomains, the envelope coupling identifier and swing suppression state are written into the state chain continuation boundary; this ensures that subsequent state determination results remain continuous with the previous envelope revision trajectory. After processing, a boundary update result set is generated, including the boundary update results of deep feature interpretation, scene constraint triggered boundary update, state candidate boundary update, action template switching boundary update, and state chain continuation boundary update.

[0016] This invention provides a rescue smart helmet vital sign positioning system, which has the following beneficial effects: (1) By sorting, merging clustered points, removing isolated candidate points, and merging across sequences, event stumps are no longer determined by a single data sequence, but are jointly limited by multiple actions and changes in vital signs. As a result, the event stumps obtained are closer to the state transition positions in the actual action process of rescuers, which helps to reduce the segmentation deviation caused by the instantaneous fluctuations of a single sequence. Using the time interval between adjacent event stumps as the original slice boundary, candidate task micro-screens are formed, so that the generation basis of task segments changes from a fixed time window to event-driven interval segmentation. This enables each task micro-screen to correspond to a relatively clear action transition or state transition, thereby providing a more realistic segment basis for subsequent mirror state generation and trajectory crack scanning.

[0017] (2) Through the mirror state generation module, personnel actions are no longer judged solely based on inertial navigation trajectories. Instead, physiological-behavioral mirror state cards are generated by combining the deep features within the task micro-screen, scene constraint indicator groups, and the continuity relationship between adjacent segments. In this way, the position link and the vital sign link are no longer separated from each other, but rather a unified interpretation basis is established around the same segment. This is particularly suitable for state interpretation in scenarios such as low headroom, narrow passages, and close-to-wall passages, and can address the shortcomings of "fixed action templates being difficult to adapt to complex scenarios" mentioned in the background section.

[0018] (3) The local re-evaluation envelope replaces the whole segment unified adjustment method. The crack interval and the preceding and following reliable segments are included in the local solution boundary. Coupled arrangement and linkage convergence are performed on multiple envelopes that are continuous in time and adjacent in space, so that adjacent cracks are no longer treated as separate anomalies. The left and right swing and boundary fragmentation phenomena in the path revision process are suppressed. After local re-evaluation, the envelope revision trajectory and segment-level reliability backfill table are written back to the task micro-screen generation stage and the mirror state generation stage through the stabilization memory film backfilling module. The deep feature interpretation boundary, scene constraint trigger boundary, state candidate boundary, action template switching boundary and state chain continuation boundary of the subsequent segments are redefined, so that the subsequent interpretation continues to unfold along the revised trajectory chain and state chain.

[0019] (4) By selecting the path branch from the retained candidate path branches that is more stable in connection with the preceding and following reliable segments, has fewer conflicts with the mirror state card, and has smaller deviations in orientation from adjacent envelopes as the envelope revision trajectory, the system output is not just "a path has been changed", but a local revision result that converges around the relationship between the preceding and following segments, the relationship between the state interpretation and the relationship between adjacent envelopes. This result organization method is more suitable as the basis for subsequent refeeding.

[0020] After determining the cover revision trajectory, a segment-level credibility backfill table is generated according to the task micro-screen number, so that each segment not only has the revised trajectory result, but also the corresponding credibility level, the cover to which it belongs, and the revision source. Attached Figure Description

[0021] Figure 1 This is a schematic flowchart of a rescue smart helmet vital sign positioning system according to the present invention; Figure 2 This is a schematic diagram of the same-cluster merging and isolated elimination judgment process of the present invention; Figure 3 This is a schematic diagram of the splicing and re-segmentation judgment process of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] Example 1 This invention provides a rescue intelligent helmet vital sign positioning system. Please refer to [link / reference]. Figure 1 It includes an asynchronous time-fold alignment module, a mirror state generation module, a trajectory crack scanning module, a local envelope re-estimation module, and a stabilizing memory film recharge module; The asynchronous time-fold alignment module acquires the vital sign positioning data and the reversal point of the breathing valley of the smart helmet to construct event stakes, and generates a sequence of task micro-screens based on the data interval between adjacent event stakes; The mirror state generation module generates physiological-behavioral mirror state cards for each task micro-screen based on the deep features within the task micro-screen sequence, scene constraint indicator groups, and the continuity relationship between adjacent segments. The trajectory crack scanning module compares the physiological-mirror state card with the inertial navigation trajectory segment by segment, extracts suspicious path segments that cannot be interpreted by the mirror state card, and forms a trajectory crack table. The local envelope re-estimation module uses the suspicious path segments marked in the trajectory crack table as the scope, combines the reliable segments before and after the crack to construct a local re-estimation envelope, and generates an envelope revision trajectory and a segment-level reliability backfill table within the local re-estimation envelope. The stabilization memory membrane recharge module writes the envelope revision trajectory and fragment-level credibility backfill table into the stabilization memory membrane and recharges it into the task micro-screen generation stage and the mirror state generation stage, updating the deep feature interpretation boundary and state determination boundary of subsequent fragments.

[0024] In this embodiment, the mirror state generation module no longer relies solely on inertial navigation trajectories to determine personnel actions. Instead, it combines deep features within the task micro-screen, scene constraint indicator groups, and the continuity relationship of adjacent segments to generate a physiological-behavioral mirror state card. This eliminates the separation between the positional and vital sign links, establishing a unified interpretation basis around the same segment. This approach is particularly suitable for state interpretation in scenarios with low headroom, narrow passages, and close-to-wall movement, addressing the deficiency mentioned in the background section regarding the difficulty of adapting fixed action templates to complex scenarios.

[0025] This embodiment uses a trajectory crack scanning module to compare the inertial navigation trajectory with the mirror state card segment by segment. It no longer only checks the smoothness of the path within the trajectory domain, but also uses head movements, respiratory rhythm, heart rate fluctuations, and scene constraint states as the basis for trajectory interpretation. In this way, trajectory drift no longer relies solely on the inertial navigation system's internal self-calibration, but introduces a physiological-behavioral counter-evidence link to examine whether it can be explained by the current state. This is suitable for addressing the problem of "lack of linkage between position determination and vital sign monitoring" mentioned in the background section.

[0026] This embodiment utilizes a local envelope re-estimation module to re-estimate only the suspicious path segments marked on the trajectory crack table, and constructs a local re-estimation envelope by combining reliable segments before and after the crack. This processing method does not modify the entire trajectory at once, but introduces before-and-after anchoring relationships within a local area, making the re-estimation boundary clearer. At the same time, the envelopes corresponding to adjacent cracks are linked and arranged, which can resolve the path back-and-forth swinging phenomenon that occurs when there are continuous back-and-forths and adjacent cracks clustered together. The problem of "path oscillation caused by the dispersed processing of adjacent cracks in the same area" mentioned in the background section is addressed here.

[0027] This embodiment, after partial envelope reassembly, not only outputs the envelope revision trajectory but also simultaneously generates a fragment-level credibility backfill table. In other words, the system records not only "what the trajectory changed to," but also "which fragments were revised, what type of crack they belonged to before the revision, what credibility level the revised fragments are at, and which envelope and template boundary each fragment is associated with." This result organization method facilitates direct retrieval in subsequent backfeeding stages, and the problem mentioned in the background section—that "partial revision results cannot be passed to subsequent judgment stages"—is structurally addressed here.

[0028] This embodiment uses a stabilization memory film backfeed module to write the envelope revision trajectory and fragment-level credibility backfill table into the stabilization memory film and backfeed it to the task micro-screen generation stage and the mirror state generation stage. In this way, when subsequent fragments are re-segmented, re-annotated with deep features, switched action templates, and re-determined, they no longer use the previously affected interpretation boundaries, but instead continue to unfold along the revised fragment boundaries. Compared to a one-time trajectory revision with no further feedback, this approach creates a continuous sequential relationship within the system, addressing the issue described in the background where "subsequent state determinations still use the old boundaries."

[0029] Example 2 Please refer to Figure 2 and Figure 3 Specifically: the asynchronous time-fold alignment module includes a stub extraction and merging unit and an interval slice arrangement unit; The pile extraction and merging unit acquires the inertial navigation displacement increment sequence, attitude sequence, heartbeat interval sequence, respiratory rhythm sequence and voice activity sequence output by the smart helmet, and maps each sequence to the same time reference for alignment processing. Gait discontinuity points are identified based on inertial navigation displacement increments and attitude changes; Identify the head turning inflection point based on changes in head yaw and pitch; Identify the start and end boundaries of speech based on the rise and fall boundaries of speech activity energy; Identify the reversal point of the respiratory trough based on the change in the direction of the respiratory rhythm trough; The identified candidate points are sorted according to time sequence. Candidate points in the same short time neighborhood are merged into the same cluster. Isolated candidate points that only appear in a single sequence and do not meet the neighborhood continuity condition are removed. Candidate points that fall into the same action transition interval across sequences are combined into the same event stump. Time location identifier, source identifier, and adjacency relationship identifier are written for each event stump to obtain an ordered set of event stumps.

[0030] The interval slicing unit reads the ordered set of event stakes and uses the time interval between adjacent event stakes as the original slice boundary to extract the corresponding data segments of the inertial navigation displacement increment sequence, attitude sequence, heartbeat interval sequence, respiratory rhythm sequence and speech activity sequence within the time interval to form candidate task micro-screens. Perform interval legality checks on each candidate task micro-screen, including interval span checks, sequence coverage checks, and boundary coherence checks; When the span of the candidate task micro-screen is less than the preset lower limit and the adjacent intervals have continuous action directions, the current interval is spliced ​​with the previous or next interval. When the span of the candidate task micro-screen is greater than the preset upper limit, a secondary segmentation point is inserted within the current interval based on the displacement turning point, posture change, or local reversal of breathing rhythm, and then the current interval is re-segmented. When a sequence has a missing segment in the current interval, write a missing segment identifier for the candidate task micro-screen and retain the remaining sequence data; After all candidate task micro-screens have been spliced, re-segmented, and segmented, the task micro-screen sequence is output in chronological order.

[0031] In this embodiment, by sorting, merging clustered points, removing isolated candidate points, and merging across sequences, event studs are no longer determined independently by a single data sequence, but are jointly defined by multiple actions and changes in vital signs. Therefore, the resulting event studs more closely resemble the state transition positions during the actual actions of rescue personnel, which helps reduce segmentation deviations caused by instantaneous fluctuations in a single sequence. Using the time interval between adjacent event studs as the original slice boundary, candidate task micro-scenes are formed. This transforms the generation of task segments from a fixed time window to event-driven interval segmentation, enabling each task micro-scene to correspond to a relatively clear action or state transition. This provides a more realistic segment basis for subsequent mirror state generation and trajectory crack scanning.

[0032] By checking the interval span, sequence coverage, and boundary coherence, the candidate task micro-screens are screened for legality. This avoids fragment distortion caused by intervals that are too short, too long, or incomplete coverage of multi-source data, making the output task micro-screens more suitable for subsequent status determination processes in terms of time scale and data integrity.

[0033] When the candidate task micro-screen interval is too short, forward or backward splicing is performed based on the direction of continuous movement; when the interval is too long, secondary segmentation points are inserted and re-segmented based on displacement transitions, sudden changes in posture, or local reversals of breathing rhythm. This processing method allows the segment length to converge around the actual movement rhythm, avoiding fragmented segments that lead to fragmented state judgments, and also avoiding fragmented segments that cause multiple action processes to be mixed in the same segment. For sequences with missing segments, a missing segment marker is written and the remaining sequence data is retained, so that the system can continue to form task micro-screens even when local acquisition is incomplete, rather than discarding the data for the corresponding time period as a whole. This helps maintain continuous processing capabilities during the rescue process and reduces the breakage of the segment chain caused by short-term loss of a single channel.

[0034] Example 3 Please refer to Figure 1 Specifically: the mirror state generation module includes a candidate state construction and switching unit and a continuous verification card generation unit; The candidate state construction and switching unit extracts deep features and scene constraint indication groups for each task micro-screen sequence; Based on the combination form, occurrence order and co-occurrence relationship of each deep feature in the current task micro-screen, construct the candidate state set corresponding to the task micro-screen; and detect whether the current task micro-screen satisfies the scene-constrained action template switching condition. When at least two signs in the scene constraint indication group appear consecutively in the current task micro-screen and the adjacent task micro-screen, the current task micro-screen is switched from the regular action template branch to the scene constraint action template branch, and the head scan arc interpretation boundary, the stationary wake interpretation boundary, and the head position-displacement deviation interpretation boundary are rewritten simultaneously to generate the state candidate record corresponding to the current task micro-screen.

[0035] The continuous verification card unit is based on the status candidate record and combines the continuity relationship between the current task micro-screen and the adjacent task micro-screen to perform continuous verification, compatibility screening and target status freezing processing on each candidate status in sequence. Continuity relationships include the direction of action continuation, the continuation of bodily signs, the continuation of head behavior, and the continuation of scene constraints; First, determine whether the candidate states of the current task's micro-screen meet the continuous transition conditions with the frozen state of the previous task's micro-screen. Then, determine whether the state trend of the next task's micro-screen remains consistent. When a candidate state simultaneously satisfies the requirement of continuous transition between the preceding and following segments, the candidate state is retained as the priority state. If a candidate state of the current task micro-screen conflicts with the state chain between adjacent task micro-screens, or does not satisfy at least one of the continuous relationships, then the candidate state is downgraded or eliminated. After the screening is completed, the state that is most coherent with the preceding and following segments is determined from the remaining candidate states and used as the target state for the current task's micro-screen. After determining the target state, a physiological-behavioral mirror state card corresponding to the current task micro-screen is generated.

[0036] The trajectory crack scanning module includes a fragment comparison and discrimination unit and a crack aggregation and tabulation unit; The segment comparison and discrimination unit extracts displacement change segments, turning change segments, and vertical change segments that are consistent with the time interval of the task micro-screen film from the inertial navigation trajectory. Then, the displacement change segments, turning change segments, and vertical change segments are matched and reviewed with the segment status, head movement interpretation range, respiratory rhythm interpretation range, heartbeat swing interpretation range, adjacent segment coherence conditions, and template source markers in the corresponding physiological-behavioral mirror status card. After completing the matching review, select the corresponding difference boundary according to the template source mark; When the template source marker indicates that the current task micro-screen uses a conventional action template, conventional head scanning behavior, vital sign swinging behavior, and displacement continuation behavior are used as the primary comparison criteria. When the template source marker indicates that the current task micro-screen film uses a scene-constrained action template, the primary comparison criteria are limited observation instead of symptoms, scene constraint continuity, and displacement continuity, to avoid directly judging head movement contraction under low headroom, narrow passage, or wall-hugging conditions as trajectory abnormality. After the comparison is completed, the trajectory changes that cannot be interpreted by the corresponding mirror state card are marked as difference segments; they are marked as velocity-stationary misalignment candidate segments, steering noise candidate segments, or vertical detuning candidate segments, and a candidate crack record set is formed.

[0037] The crack aggregation and table compilation unit rearranges the candidate crack records in chronological order according to the segment number, and determines whether adjacent candidate crack records meet the aggregation conditions of temporal continuity, near continuity, correlation of trajectory change direction, and consistency of local intervals. When two or more candidate crack records meet the conditions, they are merged into the same crack segment. When adjacent candidate crack records, although of different types, point to the same chain of conflicting interpretations, they are associated and grouped. After merging, spatial adjacency checks are performed on each crack segment, and adjacent crack adjacency relationship identifiers are written for crack segments that meet the adjacency conditions. Based on the number of continuous segments, candidate type composition, template source interval, connection tension, and adjacent crack adjacency relationship of crack segments, each crack segment is sorted by level, and the crack start and end segment number, crack type, crack spanning segment duration, template source interval, connection tension identifier, spatial adjacency identifier, and adjacent crack adjacency relationship identifier are written, and the trajectory crack table is output.

[0038] In this embodiment, a candidate state construction switching unit is used to first extract deep features and scene constraint indication groups from the task micro-screen. Then, a candidate state set is constructed based on the combination form, occurrence order, and co-occurrence relationship of each feature. This makes the state determination no longer rely solely on a single trajectory change or a single symptom fluctuation, but forms a more complete state interpretation basis around the same task segment. By introducing scene constraint action template switching conditions in the candidate state construction stage, when constraint symptoms such as low clearance, narrow passage, and close-to-wall passage occur continuously, the current task micro-screen is automatically switched to the scene constraint action template branch, and the head scan arc interpretation boundary, the stationary tail interpretation boundary, and the head position-displacement deviation interpretation boundary are rewritten simultaneously. This enables the system to distinguish between two types of situations: "action is restricted by the scene and shrinks" and "trajectory abnormality leads to inconsistency". This helps to alleviate the problem of incompatibility of fixed action templates in complex spaces.

[0039] By continuously verifying the card unit, the action continuity direction, body movement continuity, head behavior continuity, and scene constraint continuity between the current task micro-screen and the adjacent task micro-screens are incorporated into the same verification process. This makes the state freeze result no longer an instantaneous judgment of isolated segments, but rather a convergence along the continuous chain of segments. This processing method helps to reduce the interference of short-term fluctuations, local noise, or single-segment anomalies on the overall state judgment.

[0040] By performing continuous verification, compatibility screening, and target state freezing on candidate states, multiple possible interpretations of the same task segment can be compressed into a single target state, generating a structured physiological-behavioral mirror state card. As a result, subsequent modules do not read loose features, but a unified state object containing the segment's state, the range of action interpretations, the range of physical sign interpretations, coherence conditions, and template source boundaries, which helps improve the targeting of subsequent difference judgment and crack identification.

[0041] When comparing and judging segments, the trajectory crack scanning module no longer only considers the continuity of the guide trajectory itself, but also performs segment-by-segment matching and review of displacement change segments, turning change segments, and vertical change segments with the state interpretation boundaries in the corresponding physiological-behavioral mirror state cards. This transforms the identification of trajectory anomalies from "judging within the trajectory domain" to a comprehensive judgment of "whether the trajectory can be interpreted by the current physiological-behavioral state," which is beneficial to addressing the deficiency in the background section that "trajectory drift can only be handled within the position domain."

[0042] By selecting different judgment boundaries according to the template source mark, when the current task micro-screen uses a conventional action template, the conventional head scanning behavior and vital sign swing behavior are used for comparison. When the current task micro-screen uses a scene-constrained action template, the restricted observation is replaced with the comparison of signs, scene constraint continuity relationship and displacement continuity relationship. This can reduce misjudgments caused by the natural contraction of head movements under conditions of low clearance, narrow passage or close to the wall, and help make the crack recognition results closer to the working state of rescuers in real confined scenarios.

[0043] By marking trajectory changes that cannot be interpreted by the mirror status card as candidate segments of velocity-stationary misalignment, candidate segments of steering noise, or candidate segments of vertical detuning, the system can express different types of trajectory conflicts in a hierarchical manner, rather than just giving a general anomaly message. In this way, subsequent processing can clarify whether the current problem is more inclined to translational anomaly, steering anomaly, or vertical anomaly, providing a clearer crack type basis for subsequent local reassessment.

[0044] Example 4 Please refer to Figure 1 Specifically: the local envelope re-evaluation module includes an envelope extraction and coupling unit and a branch re-evaluation and backfilling unit; The envelope extraction coupling unit uses the start and end segment range corresponding to each crack segment as the local re-evaluation center interval, and then extracts the continuous trajectory segments that are directly connected and not marked as crack segments from the front and back of the center interval as the reliable segments before and after the crack. The central interval, the front reliable segment, and the rear reliable segment are combined to form a local revaluation envelope; Based on the adjacency relationship of adjacent cracks in the trajectory crack table, multiple local revaluation envelopes are coupled and arranged. Locally re-estimated envelopes that are temporally continuous, spatially adjacent, and share the same preceding and following reliable segments or common boundary intervals are merged into the same adjacent envelope coupling subdomain. Perform envelope validity processing on each local revaluation envelope, including the integrity of the central crack interval, the continuity of the preceding and following reliable segments, the consistency of the template source boundary, and the attribution relationship of the coupled subdomain; after processing, output the set of local revaluation envelopes.

[0045] The branch revaluation backfill unit constructs multiple candidate path branches on the original inertial navigation trajectory based on the central crack type, template source interval, credible anchor segment location, and coupling subdomain parameters corresponding to the local revaluation envelope, and matches the candidate path branches with the physiological-behavioral mirror state cards of the corresponding segments respectively. During the matching process, static stability repulsion constraints are applied to short-term observation or statically stable transition segments, steering lock constraints are applied to continuous scanning and turning segments, stabilization closure constraints are applied to vertical displacement segments, and scene constraint boundary constraints are applied to segments using scene-constrained action templates. By applying constraints, each candidate path branch is screened and reduced segment by segment, and candidate path branches that do not meet the interpretation boundary of the mirror state card and the connection conditions of the trusted anchor segment are removed from the re-estimation sequence. After the candidate path branches are reduced, the remaining candidate path branches are subjected to envelope-based orbit determination. For locally re-evaluated envelopes located within the coupling subdomains of adjacent envelopes, the candidate path branches in multiple adjacent envelopes are converged in a coordinated manner by combining the envelope boundary inheritance relationship, the shared anchor segment relationship, and the sway suppression relationship, so that the revision trajectory between adjacent envelopes extends continuously along the same reference direction and maintains the consistency of the common boundary interval. From the retained candidate path branches, select the path branch that has the most stable connection with the preceding and following credible segments, the fewest conflicts with the physiological-behavioral mirror state card, and the smallest deviation from the orientation of adjacent envelopes within the coupled subdomain as the envelope revision trajectory; after determining the envelope revision trajectory, generate a segment-level credibility backfill table according to the task micro-screen number.

[0046] In this embodiment, the start and end segment range corresponding to each crack segment is used as the local re-estimation center interval, and directly connected reliable segments are introduced before and after it as constraint boundaries. This makes the trajectory revision no longer revolve around a single anomaly point, but around a local continuous interval of "crack interval + reliable segments before and after". This makes the re-estimation range more closely match the real anomaly segment, and also addresses the problem mentioned in the background that "anomaly segments exist discretely and are difficult to form a continuous interpretation chain".

[0047] By combining the central interval, the preceding reliable segment, and the following reliable segment to form a local reassessment envelope, the trajectory revision has clear preceding and following boundaries. Compared to uniformly revising the entire trajectory, this local envelope approach is more suitable for the segmented movement of rescue personnel in complex spaces. It helps to limit the revision to the local area that truly needs to be addressed, allowing unaffected trajectory segments to maintain their original continuity.

[0048] Based on the adjacency relationships of adjacent cracks in the trajectory crack table, multiple local revaluation envelopes are coupled and arranged. Envelopes that are temporally continuous, spatially adjacent, and share reliable segments or common boundary intervals are merged into the same adjacent envelope coupling subdomain, so that adjacent cracks are no longer split into isolated processing objects. This is beneficial to addressing the deficiency in the background section where "path oscillation occurs after multiple cracks in the same area are processed separately".

[0049] By processing the integrity of the central crack interval, the continuity of reliable segments before and after, the consistency of template source boundaries, and the attribution of coupled subdomains for each local revaluation envelope, the local envelope already possesses relatively clear structural conditions before entering the revaluation. This preprocessing method facilitates subsequent revaluations to revolve around the same boundary system, ensuring consistency in the solution basis between different envelopes and reducing the likelihood of inconsistencies in the basis for revisions before and after.

[0050] During the candidate path branch reduction process, static stability repulsion constraints, turning locking constraints, stabilization closure constraints, and scene constraint boundary constraints are applied to different segments to give each type of crack segment a different reassessment basis. This allows for the differentiation of situations such as short-term observation, continuous turning, vertical changes, and restricted passage, making local reassessment closer to the actual behavior of rescuers on site, and also helps to reduce the spillover of judgment errors caused by head contraction in confined spaces.

[0051] For locally re-evaluated envelopes located within the coupled subdomains of adjacent envelopes, the candidate path branches in multiple adjacent envelopes are converged in a coordinated manner, taking into account envelope boundary inheritance relationships, shared anchor segment relationships, and sway suppression relationships. This ensures that the revision trajectories between adjacent envelopes extend continuously along the same reference direction and maintain consistency in the common boundary intervals. This alleviates common phenomena during continuous re-evaluation, such as left-right back-and-forth deflection and inconsistent revision directions between adjacent segments, making the revised path more coherent in the local space.

[0052] Example 5 Please refer to Figure 2 Specifically: the stabilization memory membrane recharge module includes an in-membrane writing cataloging unit and a boundary recharge update unit; The cataloging unit written into the membrane performs fragmented expansion processing on the cover revision trajectory, maps the revised trajectory results to the corresponding task micro-screen segment interval, and then matches the information segments in the segment-level credibility backfill table with the fragmented revision trajectory one by one to form a revision record for single segments. Based on the sequence of the mission micro-episodes, the position of the envelope boundary, and the affiliation of the coupled subdomains, each revision record is cataloged and organized to ensure that the revision records within the same envelope are arranged continuously, that the revision records within the coupled subdomains of adjacent envelopes are kept to the same lineage, and that they are written into the stabilization memory film. During the writing process of the stabilizing memory membrane, an intra-membrane index relationship is generated, which includes a fragment index, an envelope index, a template source index, and a coupling lineage index. Through write processing, a structured in-membrane revision record set organized by fragment, envelope, template boundary, and envelope lineage is formed within the stabilizing memory membrane; Boundary trimming is performed on the structured intramembrane revision record set, and the intramembrane refill dataset is output. The boundary re-injection update unit redefines the boundaries of the displacement-breathing decoupling residual pattern interpretation interval, head stationary wake interpretation interval, breathing valley distance viscous spot interpretation interval, and heartbeat fall tail interpretation interval corresponding to the micro-screen of subsequent tasks based on the revision trajectory, fragment index, and confidence level in the intra-membrane re-injection dataset. It also synchronously corrects the trigger intervals of low clearance collision sign, lateral head swing amplitude reduction sign, continuous bending passage sign, and forward wall-hugging passage sign by combining the template source index and envelope genealogy relationship. After the correction is completed, based on the data in the intramembrane reinfusion dataset, the state candidate boundary, action template switching boundary and state chain continuation boundary of the subsequent task micro-screen are reset, so that the mirror state generation link will preferentially inherit the revised trusted behavior chain and template source boundary when generating the subsequent physiological-behavioral mirror state card. For segments within scene-constrained action template branches, maintain the extension relationship of their template continuity intervals; For segments within the influence range of adjacent envelope coupling subdomains, the envelope coupling identifier and swing suppression state are written into the state chain continuation boundary; After processing, a boundary update result set is generated, including the boundary update results of deep feature interpretation, scene constraint triggered boundary update, state candidate boundary update, action template switching boundary update, and state chain continuation boundary update.

[0053] In this embodiment, inertial navigation displacement, attitude, heartbeat, respiration, and speech activity are first organized into task micro-screens with unified temporal meaning. Then, a physiological-behavioral mirror state card is generated around the task micro-screens. Under the constraints of this state card, the inertial navigation trajectory is reviewed segment by segment and cracks are identified. This allows trajectory interpretation to no longer be limited to the location domain, but to simultaneously incorporate head movements, vital sign swings, scene constraints, and segment continuity relationships as interpretive criteria. Based on this, the system has stronger segment fitting for complex motion processes in underground spaces, indoor buildings, steel structure areas, low-clearance passages, wall-hugging passage areas, and continuous turning-back areas. It enables multi-source asynchronous data to be uniformly organized around the same action turning boundary, distinguishing action contraction in restricted scenarios such as low clearance and narrow passages from abnormal real trajectories. Furthermore, the crack identification results have clear start and end ranges, crack categories, template source boundaries, and relationships between adjacent cracks.

[0054] This invention replaces the unified whole-segment revision method with a local re-evaluation envelope. It incorporates the crack interval and the preceding and following reliable segments into the local solution boundary, and performs coupled orchestration and coordinated convergence on multiple temporally continuous and spatially adjacent envelopes. This prevents adjacent cracks from being treated as isolated anomalies, suppressing left-right oscillations and boundary fragmentation during path revision. After local re-evaluation, the envelope revision trajectory and segment-level reliability backfill table are written back to the task micro-screen generation and mirror state generation stages via a stabilization memory film backfilling module. This redefines the deep feature interpretation boundaries, scene constraint trigger boundaries, state candidate boundaries, action template switching boundaries, and state chain continuation boundaries of subsequent segments, allowing subsequent interpretations to continue along the revised trajectory and state chains. Based on the above combined relationships, this invention not only makes local revisions to the current crack segment but also allows the revision results to continue participating in the interpretation and judgment of subsequent segments, forming a closed-loop structure from segment organization, state card generation, crack identification, local re-evaluation to boundary backfilling. This ensures a continuous correspondence between personnel position judgment, vital sign status interpretation, and path revision. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. A rescue intelligent helmet vital sign positioning system, characterized in that: It includes an asynchronous time-fold alignment module, a mirror state generation module, a trajectory crack scanning module, a local envelope re-evaluation module, and a stabilization memory film recharge module; The asynchronous time-fold alignment module acquires the vital sign positioning data and the reversal point of the breathing valley of the smart helmet to construct event stakes, and generates a sequence of task micro-screens based on the data interval between adjacent event stakes; The mirror state generation module generates physiological-behavioral mirror state cards for each task micro-screen based on the deep features within the task micro-screen sequence, scene constraint indicator groups, and the continuity relationship between adjacent segments. The trajectory crack scanning module compares the physiological-mirror state card with the inertial navigation trajectory segment by segment, extracts suspicious path segments that cannot be interpreted by the mirror state card, and forms a trajectory crack table. The local envelope re-estimation module uses the suspicious path segments marked in the trajectory crack table as the scope, combines the reliable segments before and after the crack to construct a local re-estimation envelope, and generates an envelope revision trajectory and a segment-level reliability backfill table within the local re-estimation envelope. The stabilization memory membrane recharge module writes the envelope revision trajectory and fragment-level credibility backfill table into the stabilization memory membrane and recharges it into the task micro-screen generation stage and the mirror state generation stage, updating the deep feature interpretation boundary and state determination boundary of subsequent fragments.

2. The rescue intelligent helmet vital sign positioning system according to claim 1, characterized in that: The asynchronous time-fold alignment module includes a stake point extraction and merging unit and an interval slice arrangement unit; The pile extraction and merging unit acquires the inertial navigation displacement increment sequence, attitude sequence, heartbeat interval sequence, respiratory rhythm sequence and voice activity sequence output by the smart helmet, and maps each sequence to the same time reference for alignment processing. Gait discontinuity points are identified based on inertial navigation displacement increments and attitude changes; Identify the head turning inflection point based on changes in head yaw and pitch; Identify the start and end boundaries of speech based on the rise and fall boundaries of speech activity energy; Identify the reversal point of the respiratory trough based on the change in the direction of the respiratory rhythm trough; The identified candidate points are sorted according to time sequence. Candidate points in the same short time neighborhood are merged into the same cluster. Isolated candidate points that only appear in a single sequence and do not meet the neighborhood continuity condition are removed. Candidate points that fall into the same action transition interval across sequences are combined into the same event stump. Time location identifier, source identifier, and adjacency relationship identifier are written for each event stump to obtain an ordered set of event stumps.

3. The rescue intelligent helmet vital sign positioning system according to claim 2, characterized in that: The interval slicing unit reads the ordered set of event stakes and uses the time interval between adjacent event stakes as the original slice boundary to extract the corresponding data segments of the inertial navigation displacement increment sequence, attitude sequence, heartbeat interval sequence, respiratory rhythm sequence and speech activity sequence within the time interval to form candidate task micro-screens. Perform interval legality checks on each candidate task micro-screen, including interval span checks, sequence coverage checks, and boundary coherence checks; When the span of the candidate task micro-screen is less than the preset lower limit and the adjacent intervals have continuous action directions, the current interval is spliced ​​with the previous or next interval. When the span of the candidate task micro-screen is greater than the preset upper limit, a secondary segmentation point is inserted within the current interval based on the displacement turning point, posture change, or local reversal of breathing rhythm, and then the current interval is re-segmented. When a sequence has a missing segment in the current interval, write a missing segment identifier to the candidate task micro-screen and retain the remaining sequence data; After all candidate task micro-screens have been spliced, re-segmented, and segmented, the task micro-screen sequence is output in chronological order.

4. The rescue intelligent helmet vital sign positioning system according to claim 3, characterized in that: The mirror state generation module includes a candidate state construction and switching unit and a continuous verification and card generation unit; The candidate state construction and switching unit extracts deep features and scene constraint indication groups for each task micro-screen sequence; Based on the combination form, occurrence order and co-occurrence relationship of each deep feature in the current task micro-screen, construct the candidate state set corresponding to the task micro-screen; and detect whether the current task micro-screen satisfies the scene-constrained action template switching condition. When at least two signs in the scene constraint indication group appear consecutively in the current task micro-screen and the adjacent task micro-screen, the current task micro-screen is switched from the regular action template branch to the scene constraint action template branch, and the head scan arc interpretation boundary, the stationary wake interpretation boundary, and the head position-displacement deviation interpretation boundary are rewritten simultaneously to generate the state candidate record corresponding to the current task micro-screen.

5. The rescue intelligent helmet vital sign positioning system according to claim 4, characterized in that: The continuous verification card unit is based on the status candidate record and combines the continuity relationship between the current task micro-screen and the adjacent task micro-screen to perform continuous verification, compatibility screening and target status freezing processing on each candidate status in sequence. Continuity relationships include the direction of action continuation, the continuation of bodily signs, the continuation of head behavior, and the continuation of scene constraints; First, determine whether the candidate states of the current task's micro-screen meet the continuous transition conditions with the frozen state of the previous task's micro-screen. Then, determine whether the state trend of the next task's micro-screen remains consistent. When a candidate state satisfies the requirement of continuous transition between the preceding and following segments, the candidate state is retained as the priority state. If a candidate state of the current task micro-screen conflicts with the state chain between adjacent task micro-screens, or does not satisfy at least one of the continuous relationships, then the candidate state is downgraded or eliminated. After the screening is completed, the state that is most coherent with the preceding and following segments is determined from the remaining candidate states and used as the target state for the current task's micro-screen. After determining the target state, a physiological-behavioral mirror state card corresponding to the current task micro-screen is generated.

6. The rescue intelligent helmet vital sign positioning system according to claim 5, characterized in that: The trajectory crack scanning module includes a fragment comparison and discrimination unit and a crack aggregation and tabulation unit; The segment comparison and discrimination unit extracts displacement change segments, turning change segments, and vertical change segments that are consistent with the time interval of the task micro-screen film from the inertial navigation trajectory. Then, the displacement change segments, turning change segments, and vertical change segments are matched and reviewed with the segment status, head movement interpretation range, respiratory rhythm interpretation range, heartbeat swing interpretation range, adjacent segment coherence conditions, and template source markers in the corresponding physiological-behavioral mirror status card. After completing the matching review, select the corresponding difference boundary according to the template source mark; When the template source marker indicates that the current task micro-screen uses a conventional action template, conventional head scanning behavior, vital sign swinging behavior, and displacement continuation behavior are used as the primary comparison criteria. When the template source marker indicates that the current task micro-screen film uses a scene-constrained action template, the primary comparison criteria are limited observation instead of symptoms, scene constraint continuity, and displacement continuity, to avoid directly judging head movement contraction under low headroom, narrow passage, or wall-hugging conditions as trajectory abnormality. After the comparison is completed, the trajectory changes that cannot be interpreted by the corresponding mirror state card are marked as difference segments; they are marked as velocity-stationary misalignment candidate segments, steering noise candidate segments, or vertical detuning candidate segments, and a candidate crack record set is formed.

7. The rescue intelligent helmet vital sign positioning system according to claim 6, characterized in that: The crack aggregation and table compilation unit rearranges the candidate crack records in chronological order according to the segment number, and determines whether adjacent candidate crack records meet the aggregation conditions of temporal continuity, near continuity, correlation of trajectory change direction, and consistency of local intervals. When two or more candidate crack records meet the conditions, they are merged into the same crack segment. When adjacent candidate crack records, although of different types, point to the same chain of conflicting interpretations, they are associated and grouped. After merging, spatial adjacency checks are performed on each crack segment, and adjacent crack adjacency relationship identifiers are written for crack segments that meet the adjacency conditions. Based on the number of continuous segments, candidate type composition, template source interval, connection tension, and adjacent crack adjacency relationship of crack segments, each crack segment is sorted by level, and the crack start and end segment number, crack type, crack spanning segment duration, template source interval, connection tension identifier, spatial adjacency identifier, and adjacent crack adjacency relationship identifier are written, and the trajectory crack table is output.

8. The rescue intelligent helmet vital sign positioning system according to claim 7, characterized in that: The local envelope re-estimation module includes an envelope extraction and coupling unit and a branch re-estimation and backfilling unit; The envelope extraction coupling unit uses the start and end segment range corresponding to each crack segment as the local re-evaluation center interval, and then extracts the continuous trajectory segments that are directly connected and not marked as crack segments from the front and back of the center interval as the reliable segments before and after the crack. The central interval, the preceding trusted segment, and the following trusted segment are combined to form a local revaluation envelope; Based on the adjacency relationship of adjacent cracks in the trajectory crack table, multiple local revaluation envelopes are coupled and arranged. Locally re-estimated envelopes that are temporally continuous, spatially adjacent, and share the same preceding and following reliable segments or common boundary intervals are merged into the same adjacent envelope coupling subdomain. Perform envelope validity processing on each local revaluation envelope, including the integrity of the central crack interval, the continuity of the preceding and following reliable segments, the consistency of the template source boundary, and the attribution relationship of the coupled subdomain; after processing, output the set of local revaluation envelopes.

9. A rescue intelligent helmet vital sign positioning system according to claim 8, characterized in that: The branch revaluation backfill unit constructs multiple candidate path branches on the original inertial navigation trajectory based on the central crack type, template source interval, credible anchor segment location, and coupling subdomain parameters corresponding to the local revaluation envelope, and matches the candidate path branches with the physiological-behavioral mirror state cards of the corresponding segments respectively. During the matching process, static stability repulsion constraints are applied to short-term observation or statically stable transition segments, steering lock constraints are applied to continuous scanning and turning segments, stabilization closure constraints are applied to vertical displacement segments, and scene constraint boundary constraints are applied to segments using scene-constrained action templates. By applying constraints, each candidate path branch is screened and reduced segment by segment, and candidate path branches that do not meet the interpretation boundary of the mirror state card and the connection conditions of the trusted anchor segment are removed from the re-estimation sequence. After the candidate path branches are reduced, the remaining candidate path branches are subjected to envelope-based orbit determination. For locally re-evaluated envelopes located within the coupling subdomains of adjacent envelopes, the candidate path branches in multiple adjacent envelopes are converged in a coordinated manner by combining the envelope boundary inheritance relationship, the shared anchor segment relationship, and the sway suppression relationship, so that the revision trajectory between adjacent envelopes extends continuously along the same reference direction and maintains the consistency of the common boundary interval. From the retained candidate path branches, select the path branch that has the most stable connection with the preceding and following credible segments, the fewest conflicts with the physiological-behavioral mirror state card, and the smallest deviation from the orientation of adjacent envelopes within the coupled subdomain as the envelope revision trajectory; after determining the envelope revision trajectory, generate a segment-level credibility backfill table according to the task micro-screen number.

10. A rescue intelligent helmet vital sign positioning system according to claim 9, characterized in that: The stabilization memory membrane recharge module includes an in-membrane writing cataloging unit and a boundary recharge update unit; The cataloging unit written into the membrane performs fragmented expansion processing on the cover revision trajectory, maps the revised trajectory results to the corresponding task micro-screen segment interval, and then matches the information segments in the segment-level credibility backfill table with the fragmented revision trajectory one by one to form a revision record for single segments. Based on the sequence of the mission micro-episodes, the position of the envelope boundary, and the affiliation of the coupled subdomains, each revision record is cataloged and organized to ensure that the revision records within the same envelope are arranged continuously, that the revision records within the coupled subdomains of adjacent envelopes are kept to the same lineage, and that they are written into the stabilization memory film. During the writing process of the stabilizing memory membrane, an intra-membrane index relationship is generated, which includes a fragment index, an envelope index, a template source index, and a coupling lineage index. Through write processing, a structured in-membrane revision record set organized by fragment, envelope, template boundary, and envelope lineage is formed within the stabilizing memory membrane; Boundary trimming is performed on the structured intramembrane revision record set to output the intramembrane refill dataset; The boundary re-injection update unit redefines the boundaries of the displacement-respiration decoupling residual pattern interpretation interval, head stationary wake interpretation interval, respiratory valley distance viscous spot interpretation interval, and heartbeat fall tail interpretation interval corresponding to the micro-screen film of subsequent tasks, based on the revision trajectory, fragment index, and confidence level in the intra-membrane re-injection dataset. It also synchronously corrects the trigger intervals of low clearance collision sign, lateral head swing amplitude reduction sign, continuous bending passage sign, and forward wall-hugging passage sign, based on the template source index and envelope genealogy relationship. After the correction is completed, based on the data in the intramembrane reinfusion dataset, the state candidate boundary, action template switching boundary and state chain continuation boundary of the subsequent task micro-screen are reset, so that the mirror state generation link will preferentially inherit the revised trusted behavior chain and template source boundary when generating the subsequent physiological-behavioral mirror state card. For segments within scene-constrained action template branches, maintain the extension relationship of their template continuity intervals; For segments within the influence range of adjacent envelope coupling subdomains, the envelope coupling identifier and swing suppression state are written into the state chain continuation boundary; After processing, a boundary update result set is generated, including the boundary update results of deep feature interpretation, scene constraint triggered boundary update, state candidate boundary update, action template switching boundary update, and state chain continuation boundary update.

Images