Radar Echo Noise Reduction-Based Fall Detection System and Recognition Method

Abstract

This invention discloses a fall detection system and recognition method based on radar echo noise reduction, belonging to the field of human behavior perception technology. The system includes the following steps: In a fall detection and recognition scenario, information on changes in human distance, human movement speed, and human posture collected by millimeter-wave radar is synchronously processed and spliced ​​in chronological order to form a time-series of height changes reflecting the gradual decrease in human height. This invention, by constructing a time-series of continuous evolution of human height and introducing a set of slow-falling behaviors and spatial boundary offset features, effectively identifies the process of slowly falling over furniture. Furthermore, by combining phased judgment, time backtracking, and observation rhythm adjustment mechanisms, it continuously observes and dynamically judges high-risk fall behaviors without relying on violent movement characteristics, improving monitoring reliability and timely alarm.

Description

Technical Field

[0001] This invention relates to the field of human behavior perception technology, specifically to a fall detection system and recognition method based on radar echo noise reduction. Background Technology

[0002] Radar echo noise reduction-based fall detection and recognition refers to a technology used in indoor health and wellness monitoring scenarios. It leverages an intelligent sensing system and millimeter-wave radar to continuously collect radar echoes generated by human activity. Before fall detection, non-human interference components mixed in the echoes are specifically suppressed and separated. The fall detection and recognition process is then completed based on the noise-reduced, effective echoes. Specifically, this technology uses millimeter-wave radar as a key sensing tool in the intelligent sensing system. Combined with ceiling geometry and detection boundary parameters, it distinguishes and weakens abnormal point clouds and velocity characteristics caused by environmental or structural interference such as curtain swaying, air conditioner vibration, flushing water flow, and overlapping multiple targets, preventing such interference from entering the fall detection process. Furthermore, the intelligent sensing system continuously tracks the echo changes as a person moves from standing, walking, or squatting to a rapid or slow instability and fall. By analyzing changes in body height, echo distribution collapse patterns, and stable descent characteristics after the fall, fall detection is completed after meeting certain time duration conditions. This allows for reliable detection and alarm triggering of genuine falls while ensuring continuous monitoring in multi-person and complex environments.

[0003] The existing technology has the following shortcomings:

[0004] In existing radar echo noise reduction fall detection technologies, systems typically rely on changes in human movement speed and abrupt changes in posture to determine fall behavior. When a monitored individual experiences muscle weakness due to a sudden cardiac event, their body often instinctively clings to the edge of furniture such as a table, bed, or cabinet, slowly sliding down the contact surface until they fall to the ground. This process is characterized by continuous and gentle movement, lacking obvious speed changes or impact characteristics. The resulting radar echoes easily approximate normal sitting or lying down posture transitions in terms of temporal evolution and spatial distribution. In this situation, existing fall recognition mechanisms may continuously classify this slow, involuntary fall as a normal activity state, failing to promptly initiate the fall confirmation process. This delays alarms and rescues when real high-risk events occur, posing a significant safety hazard.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a fall detection system and identification method based on radar echo noise reduction to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a fall detection and recognition method based on radar echo noise reduction, comprising the following steps:

[0008] In the fall detection and recognition scenario, the human distance change information, human movement speed change information, and human body posture change information collected by millimeter-wave radar are synchronously organized and spliced ​​in chronological order to form a time height change sequence reflecting the process of the human body gradually decreasing in height.

[0009] By analyzing the time-height change sequence, the time segments in which the human body comes into contact with the furniture are divided. From the divided time segments, continuous height descent processes lacking impact characteristics are extracted and grouped into a set of slow descent behaviors.

[0010] Based on the set of slow descent behaviors, the descent speed of the upper body and the dwell time of the lower limbs are jointly analyzed to determine the spatial boundary offset characteristics formed when the human body's descent process coincides with the contact position with the furniture, and the spatial boundary offset characteristics are introduced into the subsequent judgment process.

[0011] A phased judgment process is set up around the characteristics of spatial boundary offset. In the first stage, the continuous height descent process corresponding to the slow descent behavior set is kept in a pending judgment state. In the second stage, a set of judgment criteria for sustainable observation is constructed by combining the subtle changes in human body movements after the descent stops.

[0012] Based on the set of judgment criteria, a time backtracking and observation rhythm adjustment mechanism is introduced for the suspected incapacitated fall process corresponding to the set of slow sliding behavior. While reviewing recent radar echo changes, the fall judgment conditions are dynamically adjusted to complete the dynamic identification and processing of high-risk fall behaviors that lack violent movement characteristics.

[0013] Preferably, the steps for forming the time-altitude change sequence are as follows:

[0014] The human distance change information, human movement speed change information, and human body posture change information continuously collected by millimeter-wave radar are synchronously processed, and a time mark is added to each sampling moment. At the same time, the human distance change information is mapped to the human height representation quantity.

[0015] The synchronized observation records are sorted according to the time stamp, and the observation records of adjacent sampling times are spliced ​​together in chronological order. The human height characterization quantity is used as the main field, and the information on changes in human movement speed and changes in human body posture are used as accompanying fields and spliced ​​together synchronously.

[0016] After completing the time sequence stitching, the human height representation obtained by stitching is continuously organized along the time axis to form a unified sequence representation containing continuous descent segments, and the information on changes in human movement speed and changes in human body posture are aligned and labeled on the same time axis.

[0017] The unified sequence representation containing continuous descent segments is organized into a time-altitude change sequence for storage, so that the time-altitude change sequence can be used to trace back the human body distance change information, human body movement speed change information, and human body posture change information at the corresponding time in chronological order.

[0018] Preferably, the formation of the set of slow descent behaviors includes the following steps:

[0019] The time-by-time sequence of height change is read, and the human height representation is combined with the corresponding human movement speed change information and human body posture change information. The furniture edge position range is introduced as a reference boundary to form candidate segments that include height change, speed change, posture change and boundary proximity relationship.

[0020] Based on the candidate segments, time segments are divided according to the proximity relationship of the reference boundary. The time range in which the human body position is continuously within the proximity range of the reference boundary and the human body height representation quantity continuously decreases is determined as the contact time segment. At the same time, the corresponding human body movement speed change information and human body posture change information are retained.

[0021] Extract the continuous descent segment with continuously decreasing human height representation from the contact time segment, and combine the information on changes in human movement speed and changes in human body posture to exclude transient changes, forming a continuous height descent process lacking impact characteristics.

[0022] The continuous descent process is chronologically grouped to form a set of slow descent behaviors that includes the start and end times of the contact time segment, the altitude reduction trajectory, the speed change trajectory, and the attitude change trajectory, providing a traceable behavioral basis for subsequent judgment.

[0023] Preferably, each slow descent behavior record in the slow descent behavior set maintains a time index consistent with the time height change sequence, and retains the reference boundary proximity relationship corresponding to the contact time segment. This allows the slow descent behavior record to trace back the corresponding evolution process of the continuous height descent process and the furniture contact state along the time index, thereby providing a continuous and consistent behavioral description basis for subsequent judgment.

[0024] Preferably, the steps for determining the spatial boundary offset features are as follows:

[0025] Read the slow descent behavior records from the slow descent behavior set, extract the trajectory of the change in the upper body height representation, the trajectory of the change in the lower body position representation, and the reference boundary position corresponding to the furniture edge position range according to the time index, and form a synchronous description including the change in upper body height, the change in lower body position, and the reference boundary position.

[0026] The changes in the upper body height representation quantity are organized around the synchronous description to form a time trend description of the upper body height descent rate, and the maintenance state of the lower limb position representation quantity in the vicinity of the reference boundary position is organized to form a time trend description of the lower limb dwell time.

[0027] By combining the time-varying descriptions of the upper body's descent speed and the lower limbs' dwell time, the main descent segment of the glide, characterized by continuous upper body descent accompanied by lower limb position maintenance, is determined.

[0028] Within the main downward time segment, the relative relationship between the projection of the outer edge of the human body and the position of the reference boundary is organized to form spatial boundary offset features, which are then introduced into the subsequent judgment process.

[0029] Preferably, the spatial boundary offset features include the coincidence and offset states formed by the projection of the human body's outer edge and the reference boundary position within the main descent time segment, and the correspondence between the duration of the coincidence state, the change in the offset direction, and the dwell time of the lower limbs are used as the basis for judgment in the subsequent judgment process.

[0030] Preferably, the decision criteria set is constructed as follows:

[0031] Extract slow descent behavior records from the slow descent behavior set, align the continuous height descent process with the spatial boundary offset feature by time indexing, and set the pending state within the time range covered by the continuous height descent process.

[0032] The system reads the human height representation and human movement speed change information around the state to be judged, determines the moment when the human height changes from decreasing to maintaining and the human movement speed change information stops descending, and forms an observation time window starting from the moment of stopping descent.

[0033] Within the observation time window, information on changes in human movement speed, changes in human body posture, and changes in human height are organized. Minor changes in human movements after stopping the descent are extracted, and a set of judgment criteria is formed by combining spatial boundary offset characteristics.

[0034] By establishing a correspondence between the set of judgment criteria and the records of slow descent behavior, the continuous descent process remains in a pending state in the first stage and enters a continuous observation state in the second stage, providing continuous constraints for subsequent judgments.

[0035] Preferably, during the formation of the judgment criteria set, the minute changes in human body movement after stopping the descent are aligned with the spatial boundary offset features by time indexing, and the joint change relationship of human body height representation, human body movement speed change information and human body posture change information is continuously organized within the observation time window to constrain the continuation range of the state to be judged and limit the observation rhythm of subsequent judgments.

[0036] Preferably, based on the set of judgment criteria, the suspected incapacitated fall process corresponding to the set of slow sliding behaviors is retrospectively analyzed and the observation rhythm is adjusted. The fall judgment criteria are dynamically adjusted based on a review of recent radar echo changes, as follows:

[0037] Based on the set of judgment criteria, the set of moments when the descent stops and the range of the observation time window are determined. Taking the moment when the descent stops as the starting point for review, a range of human behavior review including spatial boundary offset characteristics is formed around the continuous descent process.

[0038] Organize recent radar echo changes around the review area and align the time index within the review area with the slow descent behavior record to form continuously readable review entries;

[0039] Based on the review of entries, the observation rhythm is adjusted by rolling the time index within the observation time window and adjusting the observation progress method according to the spatial boundary offset characteristics.

[0040] By dynamically adjusting the fall determination criteria based on the review items formed during the observation process, the suspected disability fall process corresponding to the slow sliding behavior set can be dynamically identified and processed.

[0041] The radar echo noise reduction fall detection system includes a height time sequence construction module, a slow descent recognition module, a spatial boundary feature extraction module, a phased judgment module, and a dynamic backtracking judgment module.

[0042] The height time sequence construction module, in the fall detection and recognition scenario, synchronously organizes the human distance change information, human movement speed change information, and human body posture change information collected by millimeter-wave radar, and splices them in time order to form a time height change sequence that reflects the process of the human body gradually decreasing in height.

[0043] The slow descent recognition module analyzes the time sequence of height changes, divides the time segments in which the human body comes into contact with the furniture, extracts the continuous height descent process lacking impact characteristics from the divided time segments, and groups the continuous height descent process into a set of slow descent behaviors.

[0044] The spatial boundary feature extraction module, based on the set of slow descent behaviors, jointly analyzes the descent speed of the upper body and the dwell time of the lower limbs to determine the spatial boundary offset features formed when the human body's descent process coincides with the contact position with the furniture, and introduces the spatial boundary offset features into the subsequent judgment process.

[0045] The phased judgment module sets up a phased judgment process based on the characteristics of spatial boundary offset. In the first stage, the continuous height descent process corresponding to the slow descent behavior set is kept in a pending judgment state. In the second stage, a set of judgment criteria for continuous observation is constructed by combining the subtle changes in human body movements after the descent stops.

[0046] The dynamic backtracking judgment module, based on the judgment criteria set, introduces a time backtracking and observation rhythm adjustment mechanism for suspected incapacitated fall processes corresponding to the slow descent behavior set. While reviewing recent radar echo changes, it dynamically adjusts the fall judgment conditions to complete the dynamic identification and processing of high-risk fall behaviors that lack violent movement characteristics.

[0047] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0048] This invention constructs a time-series of height changes that continuously reflects the gradual descent of a human body. Based on this, it introduces a set of slow-sliding behaviors and spatial boundary offset features. This allows fall detection to move away from instantaneous speed changes or abrupt posture changes and instead focus on the continuous behavioral evolution of the human body in a real environment. This approach effectively covers the process of a person slowly sliding down furniture due to sudden disability, avoiding the long-term classification of such high-risk processes as normal activities. Therefore, it significantly improves the reliability of identifying genuine fall behaviors in complex indoor environments and multi-person monitoring scenarios.

[0049] This invention, by setting up a phased judgment process and introducing a time backtracking and observation rhythm adjustment mechanism, enables fall judgment to have extensibility and process constraint capabilities. It dynamically adjusts judgment conditions in the continuous state after the body stops sliding, avoiding premature conclusions or omission of crucial evolutionary information. This method can continuously observe and dynamically identify high-risk falls lacking impact characteristics without relying on violent movement characteristics, thereby reducing the probability of false judgments and shortening alarm response time in truly dangerous situations, improving the overall safety and stability of the monitoring process. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0051] Figure 1 This is a flowchart of the fall detection and recognition method based on radar echo noise reduction according to the present invention.

[0052] Figure 2 This is a schematic diagram of the radar echo noise reduction fall detection system of the present invention. Detailed Implementation

[0053] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0054] This invention provides, for example Figure 1 The radar echo noise reduction-based fall detection and recognition method shown includes the following steps:

[0055] In the fall detection and recognition scenario, the human distance change information, human movement speed change information, and human body posture change information collected by millimeter-wave radar are synchronously organized and spliced ​​in chronological order to form a time height change sequence reflecting the process of the human body gradually decreasing in height.

[0056] To facilitate a continuous description of the process of a human body transitioning from an upright to a fallen position, the multidimensional observation information generated by millimeter-wave radar is organized and stitched together in the following manner to obtain a time-height change sequence that can express the gradual decrease in human height: The specific implementation steps are as follows:

[0057] The echo observation results output by the millimeter-wave radar during continuous acquisition are uniformly collected. The human distance change information, human movement speed change information, and human body posture change information corresponding to the same sampling time are treated as three parallel fields of the same observation record and synchronously processed. During the synchronous processing, a time stamp is added to each observation record. The time stamp adopts a combination of continuously increasing sampling sequence number and actual sampling time to ensure that the human distance change information, human movement speed change information, and human body posture change information maintain a correspondence in the time dimension. At the same time, the relative height relationship between the radar reference point and the ground reference surface is recorded according to the installation position parameters of the millimeter-wave radar. The human distance change information is mapped into a human height representation quantity related to the ground reference surface. The human height representation quantity is stored in a form with the same dimension and consistent in continuous sampling time, so that the subsequent stitched sequence can express the process of gradually decreasing human height under the same coordinate meaning.

[0058] Based on the synchronously processed observation records, the records are sorted according to time stamps, and the observation records at adjacent sampling times are spliced ​​into continuous segments in chronological order. During the splicing process, the human height representation is used as the main field, while the information on changes in human movement speed and changes in human body posture are used as accompanying fields and spliced ​​synchronously with the human height representation, forming a continuous structure in which the three fields of "height-speed-posture" evolve together over time. The information on changes in human movement speed uses the speed field within the same observation record as the height representation to run through the spliced ​​segments, and the information on changes in human body posture uses the posture field within the same observation record as the height representation to run through the spliced ​​segments. This ensures that each time position contains the height expression of human distance change information, the human movement speed change information at the corresponding time, and the human body posture change information at the corresponding time, thereby providing continuously readable chronological data for subsequent identification of the gradual decrease in human height.

[0059] After completing the temporal sequence splicing, the human height representation quantities in the spliced ​​segments are continuously organized. The human height representation quantities are formed into a sequence body on a one-dimensional time axis according to the time markers. The information on changes in human movement speed and changes in human body posture are aligned and labeled on the same time axis. This results in a unified sequence representation in which changes in height, speed, and posture can be retrieved simultaneously in any continuous time segment. The expression of the gradual decrease in human height in the sequence representation is presented in the form of "continuous descent segments". That is, a continuous descent segment is formed on the time axis based on the decreasing relationship of human height representation quantities at adjacent times. At the same time, the synchronous trajectory of the information on changes in human movement speed and changes in human body posture within the same time segment as the continuous descent segment is retained. This makes the continuous descent segment not only have a directional description of the gradual decrease in height, but also a linked description of the direction of speed change and posture change that accompany the change in height. Thus, the time height change sequence can reflect the gradual decrease in human height and carry the synchronous change background of human distance change, human movement speed change, and human body posture change information in this process.

[0060] The unified sequence representation containing continuous descent segments is organized into an output format of time-height change sequence. The output format is stored and retrieved using time stamps as indexes, human height as the main value, and human movement speed change information and human body posture change information as accompanying values. The observation record corresponding to the start time of the sequence is recorded at the beginning of the sequence, and the observation record corresponding to the end time of the sequence is recorded at the end of the sequence, so that the time-height change sequence covers the complete time span from start to finish, and the human distance change information, human movement speed change information, and human body posture change information at the same moment can be traced back at any time position. The time-height change sequence formed by the above four sub-steps is structurally continuous, temporally ordered, and synchronously consistent in fields. It can continuously reflect the process of gradually decreasing human height in a unified temporal sequence splicing form, and provide a stable serialized input for subsequent behavior segmentation and judgment criterion extraction based on the process of gradually decreasing human height.

[0061] By analyzing the time-height change sequence, the time segments in which the human body comes into contact with the furniture are divided. From the divided time segments, continuous height descent processes lacking impact characteristics are extracted and grouped into a set of slow descent behaviors.

[0062] To ensure that subsequent judgments focus on the continuous process of a person sliding down the furniture support surface until falling to the ground, time segments are divided and processes are collected based on the time-height change sequence. The specific implementation steps are as follows:

[0063] The time-series of height changes is read moment by moment. The height representation of the human body at each moment is compared with the changes in human movement speed and body posture at the same moment. A unified time index is used to organize several consecutive moments into candidate segments. Within the candidate segments, the decreasing trend of the height representation, the changing trend of human movement speed, and the changing trend of body posture are recorded synchronously. At the same time, the edge position range of the furniture in the monitoring space is introduced as a reference boundary. The phenomenon of the human body staying near the reference boundary in the candidate segments is included in the description. Thus, the candidate segments have the synchronous expression of four elements: "height change, speed change, posture change, and boundary proximity". This provides a continuous and traceable time organization basis for the subsequent division of the time period of human contact with furniture.

[0064] Based on the candidate segments, contact time segments are divided. The division process revolves around the element of proximity to the reference boundary. The time range in which the human body position is continuously within the proximity range of the reference boundary and the human body height representation value continuously decreases is marked as the contact time segment. Within the contact time segment, the information on changes in human body movement speed and changes in human body posture are further retained synchronously. This makes the contact time segment not only include the spatial semantics of "contact occurs near the edge of the furniture" but also the temporal semantics of "the human body height gradually decreases when contact occurs". At the same time, the starting time of the contact time segment is defined as the moment when the human body position enters the proximity range of the reference boundary and the height decreases. The ending time of the contact time segment is defined as the moment when the human body position leaves the proximity range of the reference boundary, the human body height representation value stops decreasing, and the human body movement speed change information enters a stationary state. In this way, a set of contact time segments with clear boundary meaning, clear start and end times, and consistent field synchronization is formed.

[0065] The process extracts continuous descent processes lacking impact characteristics from the set of contact time segments. The extraction process uses the continuous decrease in human height representation as the main thread, and human movement speed and posture changes as auxiliary constraints. The changes in human height representation within the contact time segment are organized into several continuous descent segments. For each continuous descent segment, the human movement speed change is checked to ensure it maintains a smooth trend and to exclude transient jumps. Simultaneously, the human posture change is checked to ensure it exhibits a continuous change trend and to exclude transient abrupt changes. Thus, continuous descent segments possessing "continuous height decrease, smooth speed change, and continuous posture change" are identified as continuous descent processes lacking impact characteristics. During this continuous descent process, the duration of the descent, the starting and ending height of the descent, the range and trend of human movement speed changes during the descent, and the range and trend of human posture changes during the descent are further recorded, forming process description entries that correspond one-to-one with the time index. This allows the continuous descent process to be directly referenced in time series form and continuously invoked by subsequent steps.

[0066] The extracted continuous descent processes lacking impact characteristics are aggregated into a slow descent behavior set. The aggregation process uses process description entries of the continuous descent process as the basic unit. Multiple continuous descent processes occurring on the same monitored object within a continuous time period are sequentially linked into a slow descent behavior record. This slow descent behavior record retains the start and end times of the contact time segment, the description of the human body's position within the vicinity of the reference boundary, the gradual decrease in the human body's height representation from the start to the end height, the smooth change trajectory of the human body's movement speed throughout the entire process, and the information on changes in the human body posture. The continuous transformation trajectory throughout the entire process; after collection, the slow descent behavior set is stored in the form of multiple slow descent behavior records. Each slow descent behavior record is consistent with the time index of the time height change sequence, and retains the reference relationship that can be traced back to the original contact time segment and the original continuous height descent segment. Thus, the slow descent behavior set not only reflects the time segment division results of human contact with furniture, but also fully carries the time evolution details of the continuous height descent process without impact characteristics. This provides a continuous, synchronous, and traceable textual basis for subsequent spatial feature extraction and phased judgment based on the slow descent behavior set.

[0067] Based on the set of slow descent behaviors, the descent speed of the upper body and the dwell time of the lower limbs are jointly analyzed to determine the spatial boundary offset characteristics formed when the human body's descent process coincides with the contact position with the furniture, and the spatial boundary offset characteristics are introduced into the subsequent judgment process.

[0068] To obtain spatial semantic information that can distinguish between slow descent and autonomous attitude changes, spatial boundary offset features are formed based on the set of slow descent behaviors and incorporated into the subsequent judgment process. The specific implementation steps are as follows:

[0069] The system reads each slow descent behavior record sequentially from the collection of slow descent behaviors. Based on the time index within each record, it extracts the time-varying trajectories of human height, human movement speed, and human posture. Simultaneous descriptions of upper body height and lower limb position are established under the same time index. The upper body height reference is formed by the upper vertical coverage of the human point cloud, while the lower limb position reference is formed by the lower vertical coverage of the human point cloud. Simultaneously, the reference boundary positions corresponding to the furniture edge positions are continuously recorded between the start and end times of the slow descent behavior record. These reference boundary positions serve as spatial alignment benchmarks throughout the entire slow descent behavior record. This ensures that each slow descent behavior record simultaneously possesses three types of information in the time dimension: descriptions of upper body height changes, lower limb position changes, and reference boundary positions. This provides a unified data format for the joint analysis of upper body descent speed and lower limb dwell time.

[0070] A joint analysis was conducted on the descent velocity of the upper body and the dwell time of the lower limbs, based on the same slow descent behavior record. The joint analysis used a time index as the main thread, continuously organizing the differences in the upper body height representation at adjacent time points to form a temporal description of the descent velocity of the upper body. This description included the start time, duration, and end time of the descent. Simultaneously, the dwell time of the lower limbs within the vicinity of the reference boundary was continuously organized to form a temporal description of the dwell time of the lower limbs. During the joint analysis, the temporal description of the descent velocity of the upper body was... By synchronously comparing the description of the time trajectory of the upper body descent with the time of lower limb position maintenance, a combination relationship was obtained, and the time range corresponding to the combination relationship was limited to the main descent time segment. Within the main descent time segment, the changes in human movement speed and the changes in human body posture were further combined to form a complete process context. This ensures that the joint analysis results reflect both the continuous change in the descent speed of the upper body and the continuous maintenance of the lower limb position, while also having a time index correspondence consistent with the slow descent behavior record. This avoids mixing height fluctuations caused by short-term posture adjustments into the main descent time segment.

[0071] Within the main descent time segment, the spatial boundary offset features formed when the human body's descent process coincides with the furniture contact position are determined. Using the reference boundary position as the alignment benchmark, the spatial boundary offset features continuously describe the relative relationship between the human body's point cloud's outer edge projection in the horizontal direction and the reference boundary position. This continuous description includes the temporal evolution sequence of three spatial states: closeness, coincidence, and deviation between the human body's outer edge projection and the reference boundary position. The time range of the coincidence state is then overlapped with the main descent time segment, and the overlap segment is recorded. Within the overlap segment, the displacement trajectory of the human body's outer edge projection along the reference boundary direction and the displacement trajectory of the human body's outer edge projection perpendicular to the reference boundary direction are further recorded. These displacement trajectories are then compared with the upper part of the human body... The temporal trajectory description of the descent speed of the body is time-aligned to obtain a spatial-temporal coupled description of "boundary overlap accompanied by upper body descent". In the spatial-temporal coupled description, the deviation of the reference boundary position during the overlap state of the projection of the human body's outer edge is taken as the spatial boundary offset. The starting value, ending value, direction of change, and duration of change of the spatial boundary offset are taken as components of the spatial boundary offset feature. At the same time, the temporal trajectory description of the lower limb's dwell time is incorporated into the components, so that the spatial boundary offset feature can simultaneously express the spatial offset evolution when the furniture contact position overlaps and the lower limb position holding state. Thus, the spatial boundary offset feature can reflect the boundary slip semantics formed by the constraint of furniture support during the slow descent process.

[0072] The spatial boundary offset feature is introduced into the subsequent judgment process. Specifically, a spatial boundary offset feature field is added to each slow descent behavior record in the slow descent behavior set. The added field maintains the same time index, descent main segment time zone marker, and overlapping segment marker as the slow descent behavior record. This allows the subsequent judgment process to simultaneously obtain the time trend description of the descent speed of the upper body, the time trend description of the lower limb dwell time, and the components of the spatial boundary offset feature when reading the slow descent behavior record. At the same time, the spatial boundary offset feature is used as a judgment basis at the same level as the human height representation in the input organization of the subsequent judgment process, forming a joint input structure of height evolution description, lower limb maintenance description, and boundary offset description. The overlapping segment is used as a key observation window to undertake the staged processing in the subsequent judgment process. This allows the subsequent judgment process to continuously observe and combine conditions around the spatial boundary offset feature caused by the overlap of furniture contact positions, thereby establishing a more distinguishable spatial constraint judgment basis in scenarios where the appearance of the slow descent process is similar to that of the daily sitting process.

[0073] A phased judgment process is set up around the characteristics of spatial boundary offset. In the first stage, the continuous height descent process corresponding to the slow descent behavior set is kept in a pending judgment state. In the second stage, a set of judgment criteria for sustainable observation is constructed by combining the subtle changes in human body movements after the descent stops.

[0074] To ensure that the slow descent behavior is adequately constrained before reaching the conclusion of a fall, spatial boundary offset characteristics are incorporated into the staged processing to form a phased judgment process. The specific implementation method is as follows:

[0075] Slow descent behavior records are extracted one by one from the set of slow descent behavior records. The continuous height descent process within the slow descent behavior records is aligned with the spatial boundary offset features by time indexing to form a joint description entry under the same time axis. The joint description entry includes the start time, end time, duration, and human height representation trajectory of the continuous height descent process, as well as the start offset, end offset, offset direction, offset duration, and overlap time range with the furniture edge position range of the spatial boundary offset features. A pending judgment mark field is added to the joint description entry and placed in all time index positions covered by the continuous height descent process. This ensures that the continuous height descent process does not produce a fall conclusion or a non-fall conclusion in the first stage, but is continuously recorded only in a pending state. During the formation of the pending state, the human movement speed change information and human body posture change information corresponding to the slow descent behavior records are simultaneously retained. The human movement speed change information, human body posture change information, and spatial boundary offset features are bound together to the joint description entry to ensure that the four types of information—height, speed, posture, and boundary offset—can be referenced under the same time index in subsequent stages without field disconnection.

[0076] A phase switching condition is set around the pending state of the first stage. The phase switching condition is triggered by the process representation of the human body stopping its descent. The evolution trajectory of the human height representation and the evolution trajectory of the human movement speed change information are read along the time index in the joint description entry. The time position where the human height representation changes from a decreasing trend to a stable trend near the end of the continuous height descent process is marked as a candidate moment for stopping the descent. Simultaneously, the time position where the human movement speed change information changes from a continuously changing trend to a stationary trend at the same time position is marked as a candidate moment for stopping the descent. These candidate moments for stopping the descent are then merged with the offset endpoint time of the spatial boundary offset feature to form a set of moments for stopping the descent. After the set of moments for stopping the descent is formed, each stopping moment in the set is... The observation time window is formed by extending the time window after the moment of descent stops. The observation time window is composed of a continuous time index of fixed length. The starting point of the observation time window is the moment of descent stops, and the ending point is a preset time position after the moment of descent stops. At the same time, the judgment mark field is extended from the time index position covered by the continuous height descent process to the time index position covered by the observation time window, so that the judgment state in the first stage can be naturally transitioned to the continuous observation state in the second stage. In the process of organizing the observation time window, the consistency of the time index of the slow descent behavior record, the continuous height descent process, and the spatial boundary offset features is maintained. The joint description entries within the observation time window are spliced ​​into observation segments in chronological order to provide continuous support for the extraction of the subtle changes in human body movements after descent stops in the second stage.

[0077] In the second stage, a set of criteria for sustainable observation is constructed by combining the subtle changes in human movement after the descent stops. These subtle changes are quantified using the minute fluctuations observed in radar echoes within the observation window. Specifically, the information on changes in human movement speed within the observation segment is decomposed into short-term change segments, and the frequency, duration, and direction of these segments are recorded. Simultaneously, the information on changes in human body posture within the observation segment is decomposed into posture adjustment segments, and the frequency, duration, and direction of these adjustments are recorded. Furthermore, the evolution trajectory of human height representation within the observation segment is decomposed into height micro-motion segments, and the frequency, duration, and direction of these micro-motion segments are recorded. During the summarization of these subtle changes, spatial boundary offset characteristics are used as a consistent spatial constraint clue. Specifically, the position of the human outer edge projection relative to the furniture edge is continuously recorded within the observation window. The relative relationships within the scope are organized into three types of time evolution segments: boundary close-maintaining segments, boundary deviation and retreat segments, and boundary re-close segments. These three types of time evolution segments are then aligned with short-term change segments, attitude adjustment segments, and altitude micro-motion segments using time indexing to form joint evidence items of "velocity micro-motion—attitude micro-motion—altitude micro-motion—boundary relationship evolution." Subsequently, these joint evidence items are categorized into a judgment basis set in chronological order. The judgment basis set includes at least the set of moments when descent stops, the observation time window range, statistical descriptions of short-term change segments, attitude adjustment segments, altitude micro-motion segments, boundary relationship evolution descriptions, and spatial boundary offset characteristics. This ensures that the judgment basis set covers both the final state of the continuous altitude descent process and the ongoing state after descent stops, and reflects the constraining influence of spatial boundary offset characteristics on subtle changes in human body movements within the same structure.

[0078] The judgment criteria set is introduced into subsequent processing in the form of sustainable observation. Specifically, a one-to-one correspondence is established between the judgment criteria set and the slow descent behavior records in the slow descent behavior set. An index field of the judgment criteria set is added to the slow descent behavior records, and the set of stopping descent times and the observation time window range are written in, so that subsequent processing can directly locate the continuous observation time range when accessing the slow descent behavior records. At the same time, the pending judgment mark field is retained in the slow descent behavior records as a state field that runs through both stages. The pending state covers the continuous height descent process in the first stage and the observation time window in the second stage. Subsequent processing can use this to distinguish the time segments in the pending state from the time segments in the non-pending state. Furthermore, in the input organization of subsequent processing, the spatial boundary offset feature is placed side by side with the judgment criteria set to ensure that the spatial boundary offset feature always participates in the interpretation of subtle movement changes. This makes the phased judgment process maintain the restraint of the conclusion output in the scenario where there are no impact characteristics in the continuous height descent process. Within the observation time window after the human stops descent, the judgment criteria set is formed by the continuous organization of subtle movement changes and sustainable observation is achieved. This lays a unified, continuous and traceable textual foundation for the process-oriented identification of slow disability and fall.

[0079] Based on the set of judgment criteria, a time backtracking and observation rhythm adjustment mechanism is introduced for the suspected incapacitated fall process corresponding to the set of slow sliding behavior. While reviewing recent radar echo changes, the fall judgment conditions are dynamically adjusted to complete the dynamic identification and processing of high-risk fall behaviors that lack violent movement characteristics.

[0080] To ensure that the set of slow descent behaviors has a reviewable, extensible, and adjustable procedural constraint before reaching the conclusion output, a time backtracking and observation rhythm adjustment mechanism can be introduced based on the set of judgment criteria and integrated into the dynamic adjustment of the fall judgment conditions. The specific implementation steps are as follows:

[0081] Based on the set of judgment criteria, a review scope and review anchor points are established for suspected incapacitated fall processes. The review anchor points are selected from the set of times of cessation of descent recorded in the judgment criteria set and the observation time window range. Each time of cessation of descent in the set of times of cessation of descent is used as a review starting point. Records of slow descent behavior within the set of slow descent behavior that match the time index of the review starting point are extracted to form review entries. Each review entry simultaneously carries the time index of the continuous descent process, spatial boundary offset characteristics, time-varying descriptions of changes in human movement speed, time-varying descriptions of changes in human body posture, statistical descriptions of height micro-motion segments, statistical descriptions of posture adjustment segments, and speed... The system employs statistical descriptions of micro-motion segments and boundary relationship evolution descriptions to ensure that each review entry possesses a complete process context on the same timeline, encompassing the initial descent phase, the main descent phase, cessation of descent, and continuous observation. Regarding the review scope, a recent echo review window is formed by extending forward from the observation time window. The starting point of the recent echo review window is a preset time position before the start of the continuous altitude descent process, and the ending point is a preset time position after the end of the observation time window. The time index covered by the recent echo review window is consistent with the time-altitude change sequence, enabling continuous reading of radar echo changes within the recent echo review window and alignment with each review entry moment by moment.

[0082] A mechanism for adjusting the observation rhythm is introduced around the recent echo review window, binding the observation rhythm to review entries. This mechanism uses the sampling density of the time index and the step size of the observation window as adjustable factors. The time index within the observation time window is divided into multiple observation sub-windows, each containing a continuous time index segment that scrolls along the time axis. Simultaneously, the step size of the observation sub-window is linked to the pending marker field of the slow descent behavior set. When the pending marker field covers the continuous height descent process and the observation time window range, the observation sub-window uses a fine-grained step size to improve the temporal resolution of process tracking. When the observation sub-window crosses the end of the observation time window range and enters the tail end of the recent echo review window, it uses a coarse-grained step size to extend the observation coverage and maintain the stability of resource usage. This mechanism adjusts the observation rhythm... In the entire mechanism, the overlapping segment markers of spatial boundary offset features are used as rhythm switching cues. When the observation sub-window enters the time range covered by the overlapping segment marker, the observation sub-window advance step size is adjusted to a shorter step size and the recording density of minute movement changes is increased. This allows the boundary relationship evolution description and the velocity micro-motion segment, attitude adjustment segment, and height micro-motion segment to be arranged in a more detailed process under the same time index. When the observation sub-window leaves the time range covered by the overlapping segment marker, the observation sub-window advance step size returns to the normal step size and the continuous recording of echo changes is maintained. Through the above rhythm organization, the observation rhythm adjustment mechanism forms a rolling observation mode in the time dimension of fine following of key segments and steady advancement of non-key segments. It also ensures that each observation sub-window can be associated with the same slow descent behavior record in the slow descent behavior set with the support of the review entries.

[0083] During the observation of the sub-window's scrolling progress, the fall determination criteria are dynamically adjusted. The fall determination criteria are based on a set of joint evidence items collected in the judgment criteria set. This set includes descriptions of the end state of the continuous descent process, the set of moments when the descent stops, descriptions of boundary relationship evolution, statistical descriptions of velocity micro-motion segments, statistical descriptions of attitude adjustment segments, and statistical descriptions of height micro-motion segments, maintaining the same terminology. Spatial boundary offset characteristics are also used as spatial constraints throughout the condition set. The dynamic adjustment method is as follows: when the time index segment covered by the observation sub-window exhibits a combination of continuous boundary proximity segments, missing boundary deviation and retreat segments, decreasing frequency of velocity micro-motion segments, decreasing frequency of attitude adjustment segments, and extended duration of height micro-motion segments, the fall determination criteria are weighted in terms of time continuity requirements. This is manifested by extending the coverage time of the pending marker field and expanding the time range of the recent echo review window's backward extension. Simultaneously, the determination of recoverable activity after the descent stops is also adjusted. The threshold is raised, which is reflected in the stricter continuity requirements for the continuous existence of the posture adjustment segment and the speed micro-motion segment. When the observation sub-window covers a time index segment with a combination of boundary deviation and pullback segment appearing and continuing, posture adjustment segment and speed micro-motion segment appearing repeatedly in multiple observation sub-windows, and height micro-motion segment showing an upward trend, the fall judgment condition is adjusted in the output rhythm of the conclusion. This is reflected in shortening the number of subsequent observation sub-windows and reducing the time length of the recent echo review window extension, while keeping the pending judgment mark field removed in a shorter time and transferring the suspected disability fall process to a low-risk tracking state. Throughout the dynamic adjustment process, the fall judgment condition never introduces violent action features as a premise. The condition set is combined around three core semantic categories: continuous height descent process, spatial boundary offset features, and minor action changes after stopping the descent. This allows the process lacking violent action features to still form a stable judgment path under the combined effect of time continuity and spatial constraints.

[0084] Driven by dynamically adjusted fall detection conditions, the system dynamically identifies and processes high-risk fall behaviors. Dynamic identification is based on a time series observed through rolling sub-windows. When the fall detection conditions meet the time continuity requirement across multiple consecutive observation sub-windows and present a combination of results in the review entries ("continuous descent process closure, spatial boundary offset characteristics maintained, and minor movement changes after stopping the descent entering a low-activity state"), the corresponding slow descent behavior record in the slow descent behavior set is marked as a high-risk fall behavior, and the corresponding time index, spatial boundary offset characteristic components, set of stopping descent times, and recent echo review window range are output as processing record content. When the fall detection conditions transition to a low-risk tracking state during the observation sub-window progression, the corresponding slow descent behavior record in the slow descent behavior set remains pending release. The system marks and outputs the corresponding observation time window range, boundary relationship evolution description, and attitude adjustment segment statistical description as processing record content. The processing actions revolve around recording, alarm triggering, alarm cancellation, and continuous tracking, and are bound to the time index. Alarm triggering is executed when a high-risk fall behavior marker is formed, alarm cancellation is executed after the low-risk tracking state continuously covers the preset time range, and continuous tracking is maintained during the coverage of the tail segment of the recent echo review window. By integrating the time backtracking and observation rhythm adjustment mechanism into the continuous reference of the judgment basis set and the dynamic adjustment of the fall judgment conditions, the suspected disability fall process corresponding to the slow descent behavior set can obtain an extensible observation organization and condition combination path while reviewing recent radar echo changes, thereby completing the dynamic identification and processing of high-risk fall behaviors that lack violent movement characteristics.

[0085] This invention constructs a time-series of height changes that continuously reflects the gradual descent of a human body. Based on this, it introduces a set of slow-sliding behaviors and spatial boundary offset features. This allows fall detection to move away from instantaneous speed changes or abrupt posture changes and instead focus on the continuous behavioral evolution of the human body in a real environment. This approach effectively covers the process of a person slowly sliding down furniture due to sudden disability, avoiding the long-term classification of such high-risk processes as normal activities. Therefore, it significantly improves the reliability of identifying genuine fall behaviors in complex indoor environments and multi-person monitoring scenarios.

[0086] This invention, by setting up a phased judgment process and introducing a time backtracking and observation rhythm adjustment mechanism, enables fall judgment to have extensibility and process constraint capabilities. It dynamically adjusts judgment conditions in the continuous state after the body stops sliding, avoiding premature conclusions or omission of crucial evolutionary information. This method can continuously observe and dynamically identify high-risk falls lacking impact characteristics without relying on violent movement characteristics, thereby reducing the probability of false judgments and shortening alarm response time in truly dangerous situations, improving the overall safety and stability of the monitoring process.

[0087] This invention provides, for example Figure 2 The radar echo noise reduction-based fall detection system shown includes a height time sequence construction module, a slow descent recognition module, a spatial boundary feature extraction module, a phased judgment module, and a dynamic backtracking judgment module.

[0088] The height time sequence construction module, in the fall detection and recognition scenario, synchronously organizes the human distance change information, human movement speed change information, and human body posture change information collected by millimeter-wave radar, and splices them in time order to form a time height change sequence that reflects the process of the human body gradually decreasing in height.

[0089] The slow descent recognition module analyzes the time sequence of height changes, divides the time segments in which the human body comes into contact with the furniture, extracts the continuous height descent process lacking impact characteristics from the divided time segments, and groups the continuous height descent process into a set of slow descent behaviors.

[0090] The spatial boundary feature extraction module, based on the set of slow descent behaviors, jointly analyzes the descent speed of the upper body and the dwell time of the lower limbs to determine the spatial boundary offset features formed when the human body's descent process coincides with the contact position with the furniture, and introduces the spatial boundary offset features into the subsequent judgment process.

[0091] The phased judgment module sets up a phased judgment process based on the characteristics of spatial boundary offset. In the first stage, the continuous height descent process corresponding to the slow descent behavior set is kept in a pending judgment state. In the second stage, a set of judgment criteria for continuous observation is constructed by combining the subtle changes in human body movements after the descent stops.

[0092] The dynamic backtracking judgment module, based on the judgment criteria set, introduces a time backtracking and observation rhythm adjustment mechanism for suspected incapacitated fall processes corresponding to the slow descent behavior set. While reviewing recent radar echo changes, it dynamically adjusts the fall judgment conditions to complete the dynamic identification and processing of high-risk fall behaviors that lack violent movement characteristics.

[0093] The fall detection and recognition method based on radar echo noise reduction provided in this embodiment of the invention is implemented through the aforementioned fall detection system based on radar echo noise reduction. For details of the specific methods and processes of the fall detection system based on radar echo noise reduction, please refer to the embodiments of the fall detection and recognition method based on radar echo noise reduction described above, which will not be repeated here.

[0094] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A fall detection and recognition method based on radar echo noise reduction, characterized in that, Includes the following steps: In the fall detection and recognition scenario, the human distance change information, human movement speed change information, and human body posture change information collected by millimeter-wave radar are synchronously processed and spliced ​​in chronological order to form a height change sequence. By analyzing the time-height change sequence, the time segments in which the human body comes into contact with the furniture are divided. From the divided time segments, continuous height descent processes lacking impact characteristics are extracted and grouped into a set of slow descent behaviors. By combining the descent speed of the upper body and the dwell time of the lower limbs in the context of slow descent behavior, the spatial boundary offset characteristics formed when the descent of the human body coincides with the contact position with the furniture are determined. A phased judgment process is set up around the spatial boundary offset characteristics. In the first stage, the continuous height descent process corresponding to the slow descent behavior set is kept in a pending judgment state. In the second stage, a set of judgment criteria is constructed by combining the subtle changes in human body movements after the descent stops. Based on the set of judgment criteria, a time retrospective and observation rhythm adjustment mechanism is introduced for the suspected incapacitated fall process corresponding to the set of slow descent behaviors. While reviewing recent radar echo changes, the fall judgment conditions are dynamically adjusted.

2. The fall detection and recognition method based on radar echo noise reduction according to claim 1, characterized in that, The steps for forming a time-altitude change sequence are as follows: The human distance change information, human movement speed change information, and human body posture change information continuously collected by millimeter-wave radar are synchronously processed, and a time mark is added to each sampling moment. At the same time, the human distance change information is mapped into a human height representation quantity. The synchronized observation records are sorted according to the time stamp, and the observation records of adjacent sampling times are spliced ​​together in chronological order. The human height characterization quantity is used as the main field, and the information on changes in human movement speed and changes in human body posture are used as accompanying fields and spliced ​​together synchronously. After completing the time sequence stitching, the human height representation obtained by stitching is continuously organized along the time axis to form a unified sequence representation containing continuous descent segments, and the information on changes in human movement speed and changes in human body posture are aligned and labeled on the same time axis. The unified sequence representation containing continuous descent segments is organized into a time-altitude change sequence for storage, so that the time-altitude change sequence can be used to trace back the human body distance change information, human body movement speed change information, and human body posture change information at the corresponding time in chronological order.

3. The fall detection and recognition method based on radar echo noise reduction according to claim 2, characterized in that, The formation of a set of slow descent behaviors involves the following steps: The time-by-time sequence of height change is read, and the human height representation is combined with the corresponding human movement speed change information and human body posture change information. The furniture edge position range is introduced as a reference boundary to form candidate segments. Based on the candidate segments, time segments are divided according to the proximity relationship of the reference boundary. The time range in which the human body position is continuously within the proximity range of the reference boundary and the human body height representation quantity continuously decreases is determined as the contact time segment. At the same time, the corresponding human body movement speed change information and human body posture change information are retained. Extract the continuous descent segment with continuously decreasing human height representation from the contact time segment, and combine the information on changes in human movement speed and changes in human body posture to exclude transient changes, forming a continuous height descent process lacking impact characteristics. The continuous descent process is categorized chronologically to form a set of slow descent behaviors that include the start and end times of the contact time segment, the altitude decrease trajectory, the speed change trajectory, and the attitude change trajectory.

4. The fall detection and recognition method based on radar echo noise reduction according to claim 3, characterized in that, Each slow descent behavior record in the slow descent behavior set maintains a time index consistent with the time height change sequence and preserves the reference boundary proximity relationship corresponding to the contact time segment, so that the slow descent behavior record can trace back the corresponding evolution process of continuous height descent process and furniture contact state along the time index.

5. The fall detection and recognition method based on radar echo noise reduction according to claim 3, characterized in that, The steps for determining spatial boundary offset features are as follows: Read slow descent behavior records from the slow descent behavior set, extract the trajectory of the change in the upper body height representation, the trajectory of the change in the lower limb position representation, and the reference boundary position corresponding to the furniture edge position range according to the time index, and form a synchronous description; The changes in the upper body height representation quantity are organized around the synchronous description to form a time trend description of the upper body height descent rate, and the maintenance state of the lower limb position representation quantity in the vicinity of the reference boundary position is organized to form a time trend description of the lower limb dwell time. By combining the time-varying descriptions of the upper body's descent speed and the lower limbs' dwell time, the main descent segment of the glide, characterized by continuous upper body descent accompanied by lower limb position maintenance, is determined. Within the main downward time segment, the relative relationship between the projection of the outer edge of the human body and the position of the reference boundary is organized to form spatial boundary offset features, which are then introduced into the subsequent judgment process.

6. The fall detection and recognition method based on radar echo noise reduction according to claim 5, characterized in that, The spatial boundary offset characteristics include the coincidence and offset states formed by the projection of the human body's outer edge and the reference boundary position within the main descent time segment. The correspondence between the duration of the coincidence state, the change in the offset direction, and the dwell time of the lower limbs is used as the basis for judgment in the subsequent judgment process.

7. The fall detection and recognition method based on radar echo noise reduction according to claim 5, characterized in that, The process of constructing the set of criteria for judgment is as follows: Extract slow descent behavior records from the slow descent behavior set, align the continuous height descent process with the spatial boundary offset feature by time indexing, and set the pending state within the time range covered by the continuous height descent process. The system reads the human height representation and human movement speed change information around the state to be judged, determines the moment when the human height changes from decreasing to maintaining and the human movement speed change information stops descending, and forms an observation time window starting from the moment of stopping descent. Within the observation time window, information on changes in human movement speed, changes in human body posture, and changes in human height are organized. Minor changes in human movements after stopping the descent are extracted, and combined with spatial boundary offset characteristics, a set of judgment criteria is formed. By establishing a correspondence between the set of judgment criteria and the records of slow descent behavior, the continuous descent process remains in a pending state in the first stage and enters a continuous observation state in the second stage, providing continuous constraints for subsequent judgments.

8. The fall detection and recognition method based on radar echo noise reduction according to claim 7, characterized in that, In the process of forming the judgment criteria set, the minute changes in human body movement after stopping the descent are aligned with the spatial boundary offset features by time indexing, and the joint change relationship of human body height representation, human body movement speed change information and human body posture change information is continuously organized within the observation time window.

9. The fall detection and recognition method based on radar echo noise reduction according to claim 7, characterized in that, Based on the set of judgment criteria, time retrospection and observation rhythm adjustment are introduced for suspected incapacitated falls corresponding to the set of slow descent behaviors. The fall judgment criteria are dynamically adjusted based on a review of recent radar echo changes, as follows: Based on the set of judgment criteria, the set of moments when the descent stops and the range of the observation time window are determined. Taking the moment when the descent stops as the starting point for review, a range of human behavior review including spatial boundary offset characteristics is formed around the continuous descent process. Organize recent radar echo changes around the review area and align the time index within the review area with the slow descent behavior record to form continuously readable review entries; Based on the review of entries, the observation rhythm is adjusted by rolling the time index within the observation time window and adjusting the observation progress method according to the spatial boundary offset characteristics. By dynamically adjusting the fall determination criteria based on the review items formed during the observation process, the suspected disability fall process corresponding to the slow sliding behavior set can be dynamically identified and processed.

10. A radar echo noise reduction fall detection system, used to implement the radar echo noise reduction fall detection and recognition method according to any one of claims 1-9, characterized in that, It includes a height-time sequence construction module, a slow descent recognition module, a spatial boundary feature extraction module, a phased judgment module, and a dynamic backtracking judgment module: The height time sequence construction module, in the fall detection and recognition scenario, synchronously organizes the human distance change information, human movement speed change information, and human body posture change information collected by millimeter-wave radar, and splices them in time order to form a time height change sequence; The slow descent recognition module analyzes the time sequence of height changes to divide the time period when the human body comes into contact with the furniture, and extracts the continuous height descent process lacking impact characteristics from the divided time period. The spatial boundary feature extraction module, based on the set of slow descent behaviors, jointly analyzes the descent speed of the upper body and the dwell time of the lower limbs to determine the spatial boundary offset features formed when the human body's descent process coincides with the contact position with the furniture, and introduces the spatial boundary offset features into the subsequent judgment process. The phased judgment module sets up a phased judgment process based on the spatial boundary offset characteristics. In the first stage, the continuous height descent process corresponding to the slow descent behavior set is kept in a pending judgment state. In the second stage, a set of judgment criteria is constructed by combining the subtle changes in human body movements after the descent stops. The dynamic backtracking judgment module, based on the judgment criteria set, introduces a time backtracking and observation rhythm adjustment mechanism for the suspected incapacitated fall process corresponding to the slow downward behavior set. While reviewing recent radar echo changes, it dynamically adjusts the fall judgment conditions.

Images