Method and system for safety protection of multi-oil-saw cluster operation based on regional collaborative perception

Abstract

The application discloses a kind of multi-oil saw cluster operation safety protection method and system based on regional collaborative perception, comprising: in each oil saw node deployment collaborative perception terminal and establish time division multiple access communication network, broadcast carries ranging confidence parameter heartbeat packet;Collaborative positioning is carried out under the error state kalman filter framework, utilize mahalanobis distance criterion to suppress multipath outlier, output node pose trajectory information;Collecting throttle shaft level and magneto pulse frequency to calculate rotating speed, mode recognition is carried out in combination with IMU vibration frequency spectrum characteristics, and generates operation state identification information;IMU angular velocity evolution is monitored and is locked to the approximate felling azimuth angle by finite state machine, and asymmetric safety region boundary is adaptively constructed according to operation state;Receive adjacent node state package and construct collaborative situation topology, personnel-mechanical conflict and tree fall chain risk discrimination are executed, and hierarchical fault-tolerant safety disposal is implemented according to risk level.Thereby avoid tree fall chain disaster, guarantee the safety of operator.

Description

Technical Field

[0001] This invention relates to the field of forestry machinery safety control technology, and in particular to a safety protection method and system for multi-saw cluster operations based on regional collaborative perception. Background Technology

[0002] Chainsaws are indispensable portable cutting tools in forestry logging, emergency clearing, and landscaping maintenance. Their exposed, high-speed chainsaws pose an extremely high risk of injury. In stand-alone operation, existing safety devices primarily rely on inertial braking mechanisms and hand guards installed on the front handle of the chainsaw. These rely on detecting sudden upward movement of the saw or forward thrust of the hand to trigger chain lock-up. This passive, mechanical, stand-alone protection is incapable of sensing the location of surrounding personnel, and it cannot provide coordinated early warning when multiple chainsaws are operating simultaneously. When multiple chainsaws are operating concurrently in the same forest area, operators face multiple fatal risks due to factors such as tree obstruction, limited visibility, high noise levels, and continuous physical exertion. These risks include accidental chain contact with others, collisions and rebounds between chainsaw guides, and unaccompanied persons remaining in the path of falling trees.

[0003] To address the collaborative safety issues in multi-machine cluster operations, existing research has attempted to introduce wireless ranging and positioning technologies to achieve proximity alarms between personnel. Some solutions use radio frequency signal strength indicators to roughly estimate the distance between nodes, but this is severely affected by multipath reflections in forest areas and non-line-of-sight obstructions, resulting in insufficient accuracy and reliability in distance estimation. Another solution combines ultra-wideband ranging with inertial measurement units, abstracting each operator as a circular or spherical protected area with a fixed radius, using the distance between points as the alarm criterion. This type of solution has three main shortcomings: First, the protected area uses a fixed, isotropic geometry, failing to reflect dynamic changes in shape such as the operator's extended saw arm, forward lean, and guide plate direction, resulting in a significant discrepancy between alarm accuracy and actual intrusion risk. Second, the positioning system relies on pre-installed fixed base stations or anchor points, making it difficult to deploy in remote forest areas without communication infrastructure or at emergency clearing sites, thus limiting the solution's applicability. Third, it only focuses on direct physical collisions between personnel, failing to incorporate the unique direction of tree fall during logging operations into the collaborative situational awareness framework, and thus cannot provide early warning of chain-reaction secondary disasters caused by two chainsaws falling in intersecting directions or by assistants being in dangerous paths of fallen trees. These shortcomings result in a high false alarm rate, insufficient deployment flexibility, and incomplete safety coverage for existing collaborative safety solutions in actual chainsaw cluster operations. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides a safety protection method and system for multi-chainsaw cluster operations based on regional collaborative perception. Its important purpose is to enhance the collaborative situational awareness capability of multi-chainsaw cluster operations, avoid secondary disasters such as tree-falling chain reactions, and protect the lives of operators.

[0005] To achieve the above objectives, the first aspect of this invention provides a safety protection method for multi-chainsaw cluster operations based on regional collaborative perception, comprising: Deploy collaborative sensing terminals at each chainsaw node, establish a time-division multiple access communication network based on beacon frame interaction and distributed clock synchronization, and broadcast heartbeat packets carrying ranging confidence parameters to form a distributed spatiotemporal synchronization communication network. Based on the distributed spatiotemporal synchronous communication network, the relative position topology is initialized by multidimensional scaling transformation, cooperative relative positioning is performed under the error state Kalman filter framework, and multipath outliers are suppressed by Mahalanobis distance criterion to obtain node pose trajectory information. The system collects the level combination of the throttle shaft position switch of the target chainsaw and the pulse frequency of the magneto charging coil, calculates the real-time engine speed and divides the working condition interval, extracts the vibration spectrum features from the IMU acceleration frame and constructs a multi-dimensional feature vector to identify the current working state and generate working state identification information. The evolution of the IMU angular velocity vector is monitored and the operator's intention is confirmed by a finite state machine to lock the azimuth angle of the proposed felling. Based on the operation status identification information, the operation status switching situation is determined, and the asymmetric safety area boundary is constructed by combining the node pose trajectory information. The system receives status packets broadcast by neighboring nodes and constructs a collaborative situation topology graph. Based on the collaborative situation topology graph, it performs personnel-machine conflict and tree-chain risk assessment to evaluate the intrusion risk situation. Based on the intrusion risk situation assessment results, it performs hierarchical fault-tolerant security measures.

[0006] In this scheme, the deployment of collaborative sensing terminals at each chainsaw node, the establishment of a time-division multiple access communication network based on beacon frame interaction and distributed clock synchronization, and the broadcasting of heartbeat packets carrying ranging confidence parameters to form a distributed spatiotemporal synchronization communication network, specifically includes: After each collaborative sensing terminal is powered on, it traverses the preset working frequency band to perform channel scanning, captures the preamble of the neighboring node's transmitted frame and parses the physical layer header to extract the device identifier and terminal type fields, adds unrecorded identifiers to the neighbor list, and updates the received timestamp of existing identifiers. After multiple complete cycle scans, a neighbor device list is generated. Based on the neighbor device list, a temporary master clock source is determined by sorting the identifier values. The master clock source broadcasts a synchronization beacon frame carrying the current clock count value in the first available time slot. After receiving the synchronization beacon frame, the neighboring node latches the reception time and extracts the transmission timestamp of the master clock source. Combined with the preset delay, the instantaneous clock offset between the local device and the master clock is calculated. The instantaneous clock offset is input into an exponentially weighted moving average filter. The smoothing factor is adaptively adjusted by the variance of multiple consecutive offset measurements. When the variance increases, the smoothing factor is reduced to suppress outlier shocks. When the variance decreases, the smoothing factor is increased to quickly track clock drift. The smoothed clock offset estimate is then output. The local voltage-controlled oscillator frequency compensation is gradually adjusted using the smoothed clock offset estimate to fine-tune the clock rate. Each node transmits the corrected clock count value to the next hop node. After multiple rounds of recursive broadcasting and feedback adjustment, the clock synchronization error of the entire network converges to within the preset error threshold, forming a unified time base for the entire network. The timeline is divided into equal-length superframes and further subdivided into several time slots. Each node is sorted by identifier and uses a distributed negotiation mechanism to complete time slot declaration and conflict resolution. After determining that there are no conflict-free time slots, each node periodically broadcasts a heartbeat packet carrying the clock count value and ranging confidence parameter within its own time slot of each superframe, thus forming a distributed spatiotemporal synchronization communication network.

[0007] In this scheme, based on the distributed spatiotemporal synchronous communication network, the relative position topology is initialized using multidimensional scaling transformation, cooperative relative positioning is performed under the error state Kalman filter framework, and multipath outliers are suppressed using the Mahalanobis distance criterion to obtain unified pose trajectory information of nodes. Specifically, this includes: Each terminal summarizes the distance values ​​collected from its own node and its neighboring nodes in the most recent ranging periods into a distance matrix after confidence weighting and smoothing. Using the local node as the reference, the matrix is ​​converted into an inner product matrix through double centering and then eigenvalue decomposition is performed to extract the maximum eigenvalue and the corresponding eigenvector, thus obtaining the initial relative position topology of all nodes in the network. An error state Kalman filter framework is introduced to perform cooperative relative positioning based on the initial relative position topology of all network nodes. The error state Kalman filter framework splits the state estimation into two parallel processing links: nominal state and error state. The nominal state link uses the initial coordinates and preset static attitude provided by the initial relative position topology as the initial nominal state values, and uses IMU angular velocity and acceleration data to drive the inertial mechanical arrangement equations to recursively update the node's three-dimensional position, velocity and attitude quaternions according to the sampling period. The error state link maintains an error state vector that includes position error, velocity error, attitude misalignment angle, and gyroscope zero bias residual. In each IMU cycle, it propagates the error covariance matrix based on the state transition matrix and the process noise covariance matrix, and scales the diagonal terms of the process noise covariance matrix according to the vibration energy level of the current IMU output. Whenever this node receives the broadcast position of a neighboring node and completes a bilateral two-way ranging, it calculates the observation residual using the Euclidean distance between the nominal position of this node and the broadcast position of the neighboring node as the predicted observation value. It then calculates the innovation covariance matrix using the error state covariance matrix and the observation matrix, and performs a Mahalanobis distance test on the observation residual. If the Mahalanobis distance is below the chi-square critical value, the error state vector is updated according to the standard Kalman gain. If it exceeds the chi-square critical value but does not reach the rejection threshold, an inflation factor proportional to the degree of exceeding is applied to the observation noise covariance. If it exceeds the rejection threshold, the update is skipped directly. After the update is completed, the nominal state is corrected with the error state and the error state vector is returned to zero. When a terminal temporarily interrupts the ranging signal due to tree obstruction or exceeding the communication range, it performs open-loop inertial estimation. The nominal state is only driven by IMU data and no longer receives observation updates. During this period, the error state link continues to propagate the covariance but suspends updates, and the terminal continuously stores the nominal state of each cycle into the local cache. If the IMU acceleration and angular velocity magnitude are below the static threshold for several consecutive cycles, a virtual zero-velocity constraint is triggered to suppress position drift. After the signal is recovered, the error state smoothing recursion is performed on the historical state sequence in reverse, starting from the state after the interruption is recovered. The continuous, non-jumping full-time node pose trajectory is output, and finally the pose trajectory information of each node is obtained.

[0008] In this scheme, the calculation of real-time engine speed and division of operating condition intervals, extraction of vibration spectrum features from IMU acceleration frames and construction of multi-dimensional feature vectors for current operation status identification, and generation of operation status identification information specifically include: The throttle position is obtained by comparing the level combination of the throttle shaft position switch with the preset encoding mapping table. At the same time, the pulse jump event of the magneto charging coil is captured and the interval between adjacent pulses is measured by the hardware timer. After conversion, the real-time engine speed is obtained by averaging through a sliding window. The idle speed, medium speed transition and high speed operating conditions are divided by combining the calibration threshold. The terminal's built-in IMU acquires triaxial acceleration signals at a fixed sampling rate. After removing high-frequency noise through a digital low-pass filter, the signals are divided into frames according to a preset frame length and frame shift. A Hanning window function is applied to each frame of data, and a fast Fourier transform is performed to convert the time-domain vibration signal into a triaxial amplitude spectrum. The average value of the triaxial spectrum characteristics is taken as the spectrum representation of the current frame. Locate the main peak frequency corresponding to the maximum amplitude in the spectrum representation and record the main peak amplitude. Calculate the ratio of the energy of the first three harmonics to the total energy and the ratio of the energy of the frequency band above the preset high-frequency threshold to the energy of the entire frequency band as the high-frequency band energy ratio. At the same time, extract the zero-crossing rate and kurtosis from the time-domain acceleration signal to form a multi-dimensional feature vector. The multidimensional feature vector is input into a pre-trained offline support vector machine classifier to identify the current working state. The posterior probability of the current frame belonging to one of four modes is output: idling, high speed without load, sawing load, or shutdown. The posterior probability of multiple consecutive frames is then subjected to sliding window multi-vote weighted filtering to obtain the frame-level identification result. The frame-level identification results are cross-validated with the operating condition range determined by real-time speed and throttle position. When there is a contradiction between the identification results and the operating condition range, the results are corrected according to the prior rules, and finally the cross-validated operating status identification information is obtained.

[0009] In this scheme, the monitoring of IMU angular velocity vector evolution and the confirmation of operator intention via a finite state machine to lock the proposed felling azimuth angle, and the determination of the operation state switching situation based on the operation state identification information, combined with node pose trajectory information to construct an asymmetric safety area boundary, specifically including: After receiving the direction setting trigger signal, the terminal enters the attitude monitoring window. The built-in IMU outputs the three-axis angular velocity vector at a fixed sampling rate. The vector is filtered by the mean of the sliding window to suppress high-frequency vibration glitches. The filter window width is set according to the vibration period corresponding to the engine base frequency when the chainsaw is idling. The output is a low-frequency angular velocity component that filters out vibration interference. The low-frequency angular velocity vector is input into a three-state finite state machine for intent recognition. The state machine is initially in a waiting state. When the angular velocity magnitude exceeds the start threshold but does not exceed the upper limit of violent motion, it transitions to the action in progress state and starts an integral timer to accumulate and integrate each component of the angular velocity vector over time. If the angular velocity magnitude exceeds the upper limit of violent motion, it is reset to the waiting state and the integral value is cleared. When the angular velocity magnitude falls below the static judgment threshold and continues to reach the preset confirmation dwell time, the state machine enters the action completion confirmation state, and the projection direction of the accumulated net angular displacement vector on the horizontal plane is used as the pseudo-azimuth angle of the operator's intention, which is locked and updated in the terminal direction parameter memory. Read the current mode label in the job status identification information. When the mode label is idle or no-load high-speed mode, set the safety area shape to the limb dynamic ellipsoid mode. When the mode label is sawing load mode or the proposed felling azimuth angle is locked, switch to the logging directional asymmetric layering mode. During the shape switching, interpolate and mix the old and new boundary point sets to avoid boundary abrupt changes. When using the limb dynamic ellipsoid mode, the posture rotation matrix is ​​determined by the torso center in the pose trajectory information of this node as the origin and the body orientation quaternion. The three-axis half lengths are determined by the shoulder lateral margin, the forward extension amount of the arm extension angle and the guide plate length superimposed, and the torso height, respectively. A discretized polar coordinate point set array is generated as the boundary of the safe area. When adopting the logging directional asymmetric layered mode, a polar coordinate system is established with the root position of the guide plate as the pole and the azimuth angle of the proposed felling as the polar axis. Three layers of boundaries are generated: the core braking zone, the severe alarm zone, and the warning zone. Gaussian modulation is applied to the radius of each layer boundary so that the polar axis direction reaches its maximum value and decays as the deviation angle increases, forming an asymmetric boundary shape. Simultaneously, the velocity vector in the node position trajectory information is used to shift the center of the asymmetric shape boundary along the velocity direction, so that the safe area always leads the operator's movement trend. A random expansion amount proportional to the body sway amplitude detected by the IMU is applied to the area contour, and finally the boundary is discretized to form the boundary of the asymmetric safe area.

[0010] In this solution, the construction of a collaborative situation topology map, the assessment of intrusion risk based on personnel-machine conflict and tree-chain risk discrimination using the collaborative situation topology map, and the implementation of tiered fault-tolerant security measures based on the intrusion risk assessment results, specifically include: The status packets broadcast by neighboring nodes are received through a distributed spatiotemporal synchronous communication network. The current position coordinates, velocity vector, attitude quaternion, operation mode label, proposed azimuth angle, and safe zone boundary parameters of the neighboring nodes are parsed from the status packets. The neighbor node status table is dynamically maintained in local memory using the neighbor node identifier as an index. At the same time, tree tag information is received and parsed, and personnel node information and tree node information are combined into a collaborative situational topology map in a unified coordinate system. Traverse the current position coordinates of all neighboring nodes in the neighboring node status table, convert the neighboring node positions to the local safe area coordinate system and calculate the penetration depth relative to each layer boundary. If the neighboring node position is in the core braking zone, generate a level 1 braking command; if it is in the severe alarm zone, generate a level 2 alarm command; if it is in the warning zone, generate a level 3 warning command. Extract the directional axis vector of the neighboring node that is under sawing load or has locked the directional azimuth of the planned felling from the neighboring node status table. Project the directional axis of the machine and the directional axis of the neighboring node onto the horizontal plane and calculate the length of the common perpendicular segment. When the length of the common perpendicular segment is less than the preset cross distance threshold and the included angle of the axis is greater than the cross angle critical value, generate a two-way cross conflict alarm command. Taking the tree nodes already marked on the local machine in the collaborative situation topology map as the object, the current position of the auxiliary personnel node in the topology map is traversed. The shortest distance from the auxiliary personnel node to the directed line segment determined by the tree root point position and the backward unit vector is calculated by the vector projection method. When the shortest distance is less than the tree crown width safety radius, a chain risk alarm instruction is generated. All generated risk events are prioritized according to risk type and penetration depth to obtain an event sorting table. Based on the event sorting table, a hierarchical fault-tolerant safety handling measure is generated to protect the safety of cluster operations. The hierarchical fault-tolerant safety handling measure includes the handling strategy for any event among the first-level braking event, the second-level alarm event, and the third-level prompt event.

[0011] The second aspect of the present invention provides a safety protection system for multi-chainsaw cluster operations based on regional collaborative perception. The system includes: a memory, a processor, and a communication interface. The memory contains a program for a safety protection method for multi-chainsaw cluster operations based on regional collaborative perception. When the program for a safety protection method for multi-chainsaw cluster operations based on regional collaborative perception is executed by the processor, it implements the steps of the safety protection method for multi-chainsaw cluster operations based on regional collaborative perception as described in any of the above claims. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0013] Figure 1 A flowchart of a multi-chainsaw cluster operation safety protection method based on regional collaborative perception is provided in an embodiment of the present invention; Figure 2 This is a flowchart of an asymmetric safe region boundary construction method according to an embodiment of the present invention; Figure 3 A block diagram of a multi-chainsaw cluster operation safety protection system based on regional collaborative perception is provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0015] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0016] Figure 1A flowchart of a multi-chainsaw cluster operation safety protection method based on regional collaborative perception is provided in an embodiment of the present invention; like Figure 1 As shown, the present invention provides a flowchart of a safety protection method for multi-chainsaw cluster operations based on regional collaborative perception, including: Deploy collaborative sensing terminals at each chainsaw node, establish a time-division multiple access communication network based on beacon frame interaction and distributed clock synchronization, and broadcast heartbeat packets carrying ranging confidence parameters to form a distributed spatiotemporal synchronization communication network. Based on the distributed spatiotemporal synchronous communication network, the relative position topology is initialized by multidimensional scaling transformation, cooperative relative positioning is performed under the error state Kalman filter framework, and multipath outliers are suppressed by Mahalanobis distance criterion to obtain node pose trajectory information. The system collects the level combination of the throttle shaft position switch of the target chainsaw and the pulse frequency of the magneto charging coil, calculates the real-time engine speed and divides the working condition interval, extracts the vibration spectrum features from the IMU acceleration frame and constructs a multi-dimensional feature vector to identify the current working state and generate working state identification information. The evolution of the IMU angular velocity vector is monitored and the operator's intention is confirmed by a finite state machine to lock the azimuth angle of the proposed felling. Based on the operation status identification information, the operation status switching situation is determined, and the asymmetric safety area boundary is constructed by combining the node pose trajectory information. The system receives status packets broadcast by neighboring nodes and constructs a collaborative situation topology graph. Based on the collaborative situation topology graph, it performs personnel-machine conflict and tree-chain risk assessment to evaluate the intrusion risk situation. Based on the intrusion risk situation assessment results, it performs hierarchical fault-tolerant security measures.

[0017] Furthermore, in a preferred embodiment of the present invention, the step of deploying collaborative sensing terminals at each chainsaw node, establishing a time-division multiple access communication network based on beacon frame interaction and distributed clock synchronization, and broadcasting heartbeat packets carrying ranging confidence parameters to form a distributed spatiotemporal synchronization communication network specifically includes: After each collaborative sensing terminal is powered on, it traverses the preset working frequency band to perform channel scanning, captures the preamble of the neighboring node's transmitted frame and parses the physical layer header to extract the device identifier and terminal type fields, adds unrecorded identifiers to the neighbor list, and updates the received timestamp of existing identifiers. After multiple complete cycle scans, a neighbor device list is generated. Based on the neighbor device list, a temporary master clock source is determined by sorting the identifier values. The master clock source broadcasts a synchronization beacon frame carrying the current clock count value in the first available time slot. After receiving the synchronization beacon frame, the neighboring node latches the reception time and extracts the transmission timestamp of the master clock source. Combined with the preset delay, the instantaneous clock offset between the local device and the master clock is calculated. The instantaneous clock offset is input into an exponentially weighted moving average filter. The smoothing factor is adaptively adjusted by the variance of multiple consecutive offset measurements. When the variance increases, the smoothing factor is reduced to suppress outlier shocks. When the variance decreases, the smoothing factor is increased to quickly track clock drift. The smoothed clock offset estimate is then output. The local voltage-controlled oscillator frequency compensation is gradually adjusted using the smoothed clock offset estimate to fine-tune the clock rate. Each node transmits the corrected clock count value to the next hop node. After multiple rounds of recursive broadcasting and feedback adjustment, the clock synchronization error of the entire network converges to within the preset error threshold, forming a unified time base for the entire network. The timeline is divided into equal-length superframes and further subdivided into several time slots. Each node is sorted by identifier and uses a distributed negotiation mechanism to complete time slot declaration and conflict resolution. After determining that there are no conflict-free time slots, each node periodically broadcasts a heartbeat packet carrying the clock count value and ranging confidence parameter within its own time slot of each superframe, thus forming a distributed spatiotemporal synchronization communication network.

[0018] It should be noted that forestry work environments typically lack stable external time bases and communication infrastructure. The strong vibrations at chainsaw operation sites, tree obstruction, and node mobility pose severe challenges to clock synchronization and communication coordination among multiple terminals. To address this, a collaborative sensing terminal is installed above the handle of each chainsaw via a vibration-damping bracket. This terminal integrates a UWB transceiver unit, a six-axis IMU inertial measurement unit, a throttle status detection module, an audible and visual alarm, and an ignition control interface linked to the primary coil of the high-voltage transformer. Wearable auxiliary sensing units are attached to the operator's neck or helmet in the form of miniature UWB tags; optional tags can also be added to the shoulders and wrists. After powering on, each terminal first discovers neighboring nodes and builds a neighbor device list through a full-channel scan. This process ensures that, without pre-configuration, each node can autonomously perceive its surrounding communication partners, laying the topology foundation for subsequent distributed networking. The neighbor list records the unique identifier of each neighboring node and the time of its first signal acquisition, providing the terminal with a clear communication target for subsequent synchronization negotiation and time slot allocation.

[0019] The process of electing a temporary master clock source based on a list of neighboring devices employs a deterministic rule—identifier numerical sorting—rather than a random competition mechanism. This fundamentally avoids conflicts and convergence delays caused by multiple nodes simultaneously vying for the master clock role. The node with the smallest identifier is selected as the master clock source, ensuring that all nodes in the network reach a consistent understanding of the master clock identity without central controller intervention. In the clock offset estimation stage, the terminal does not directly use the instantaneous offset measured in a single measurement to correct the local clock. Instead, it feeds the offset into an exponentially weighted moving average filter and adaptively adjusts the smoothing factor based on the variance of multiple consecutive offset measurements. When the variance increases, the smoothing factor is reduced to suppress outlier impacts introduced by multipath reflections or sudden interference; when the variance decreases, the smoothing factor is increased to quickly track the real clock drift caused by temperature changes or crystal oscillator aging. Thus, even under conditions of repeated chainsaw starts and stops and drastic engine speed fluctuations, this approach ensures both high synchronization convergence speed and avoids clock jumps caused by instantaneous interference.

[0020] Instead of directly jumping the clock count value, the frequency compensation of the local voltage-controlled oscillator is gradually adjusted using a smoothed clock offset estimate, ensuring the continuity and monotonicity of the timestamps and avoiding the problem of timestamp sequence disorder in subsequent positioning algorithms caused by clock backtracking. Each node embeds the corrected clock count value into the synchronization beacon frame and continues to transmit it to the next hop node. Through multiple rounds of recursive broadcasting and feedback adjustment, the network clock gradually converges to the preset error range. Finally, based on a unified time reference, the time axis is divided into equal-length superframes and further subdivided into time slots. Each node independently completes time slot declaration and conflict resolution using a distributed negotiation mechanism. Each node only sends heartbeat packets in its conflict-free time slots, and remains in receiving state in the remaining time slots, eliminating data collisions and channel congestion caused by concurrent transmission by multiple nodes. The ranging confidence parameters carried in the heartbeat packet are composed of the received signal strength indication, the first-path signal power ratio, and the main peak broadening of the channel impulse response. This provides a link quality-driven observation noise adaptive basis for subsequent cooperative positioning algorithms, ensuring that positioning accuracy remains stable in occlusion and multipath environments.

[0021] Furthermore, in a preferred embodiment of the present invention, the step of initializing the relative position topology using multidimensional scaling transformation based on the distributed spatiotemporal synchronization communication network, performing cooperative relative positioning under the error state Kalman filter framework, and suppressing multipath outliers using the Mahalanobis distance criterion to obtain unified pose trajectory information of nodes specifically includes: Each terminal summarizes the distance values ​​collected from its own node and its neighboring nodes in the most recent ranging periods into a distance matrix after confidence weighting and smoothing. Using the local node as the reference, the matrix is ​​converted into an inner product matrix through double centering and then eigenvalue decomposition is performed to extract the maximum eigenvalue and the corresponding eigenvector, thus obtaining the initial relative position topology of all nodes in the network. An error state Kalman filter framework is introduced to perform cooperative relative positioning based on the initial relative position topology of all network nodes. The error state Kalman filter framework splits the state estimation into two parallel processing links: nominal state and error state. The nominal state link uses the initial coordinates and preset static attitude provided by the initial relative position topology as the initial nominal state values, and uses IMU angular velocity and acceleration data to drive the inertial mechanical arrangement equations to recursively update the node's three-dimensional position, velocity and attitude quaternions according to the sampling period. The error state link maintains an error state vector that includes position error, velocity error, attitude misalignment angle, and gyroscope zero bias residual. In each IMU cycle, it propagates the error covariance matrix based on the state transition matrix and the process noise covariance matrix, and scales the diagonal terms of the process noise covariance matrix according to the vibration energy level of the current IMU output. Whenever this node receives the broadcast position of a neighboring node and completes a bilateral two-way ranging, it calculates the observation residual using the Euclidean distance between the nominal position of this node and the broadcast position of the neighboring node as the predicted observation value. It then calculates the innovation covariance matrix using the error state covariance matrix and the observation matrix, and performs a Mahalanobis distance test on the observation residual. If the Mahalanobis distance is below the chi-square critical value, the error state vector is updated according to the standard Kalman gain. If it exceeds the chi-square critical value but does not reach the rejection threshold, an inflation factor proportional to the degree of exceeding is applied to the observation noise covariance. If it exceeds the rejection threshold, the update is skipped directly. After the update is completed, the nominal state is corrected with the error state and the error state vector is returned to zero. When a terminal temporarily interrupts the ranging signal due to tree obstruction or exceeding the communication range, it performs open-loop inertial estimation. The nominal state is only driven by IMU data and no longer receives observation updates. During this period, the error state link continues to propagate the covariance but suspends updates, and the terminal continuously stores the nominal state of each cycle into the local cache. If the IMU acceleration and angular velocity magnitude are below the static threshold for several consecutive cycles, a virtual zero-velocity constraint is triggered to suppress position drift. After the signal is recovered, the error state smoothing recursion is performed on the historical state sequence in reverse, starting from the state after the interruption is recovered. The continuous, non-jumping full-time node pose trajectory is output, and finally the pose trajectory information of each node is obtained.

[0022] It should be noted that in forest operation scenarios without fixed anchor points, each terminal lacks a predefined unified coordinate system. Therefore, the primary task of collaborative positioning is to establish the initial spatial relationships between nodes. Multidimensional scaling transformation is then used to convert the distance measurements between nodes into relative coordinates. Its core advantage is that initialization can be completed using only distance information, without any prior angle or position information. The distance measurements, weighted and smoothed by confidence over the most recent several periods, are summarized into a distance matrix, allowing the quality of the initial topology to directly benefit from the distance confidence parameters carried in the aforementioned heartbeat packets. The dual-centering process, which converts the distance matrix into an inner product matrix, essentially transforms distance constraints into coordinate constraints, and then extracts principal components as node coordinates through eigenvalue decomposition.

[0023] After obtaining the initial topology, an error-state Kalman filter framework is introduced for continuous recursive positioning. The ingenious aspect of splitting state estimation into two parallel links—nominal state and error state—lies in that the nominal state uses high-frequency IMU data to drive the inertial machinery orchestration equations, continuously updating the 3D position, velocity, and attitude quaternions of the nodes, ensuring the continuity of the positioning output. The error state link, on the other hand, is specifically responsible for estimating and compensating for accumulated errors during the inertial integration process. The two links perform their respective functions without interfering with each other. The error state vector is specifically designed for the characteristics of chainsaw operations, incorporating position error, velocity error, attitude misalignment angle, and gyroscope zero-bias residual into the estimation. Explicit modeling of the attitude misalignment angle and gyroscope zero-bias residual allows the filter to continuously calibrate the IMU's slowly varying zero bias online, avoiding attitude divergence after long-term operation due to pure inertial estimation. The process noise covariance matrix is ​​adaptively scaled according to the vibration energy level output by the IMU. When the chainsaw is running at high speed and causes strong vibrations, the noise terms of the accelerometer and gyroscope are dynamically amplified. This allows the filter to appropriately reduce its confidence in the inertial integration results during periods of strong vibration and instead rely more on UWB ranging observations to constrain position drift.

[0024] Trees and operators' bodies at chainsaw sites frequently obstruct wireless signals, causing some UWB ranging values ​​to have a positive bias due to non-line-of-sight propagation. Indiscriminately introducing filters would directly skew the position estimation. The Euclidean distance between the nominal position of the current node and the broadcast positions of neighboring nodes is used as the predicted observation value to calculate the residual. Then, the error state covariance matrix and the observation noise covariance matrix are used to synthesize the innovation covariance matrix. The residual is then tested using Mahalanobis distance, which essentially normalizes the residual due to its inherent uncertainty. If the residual significantly deviates from the expected distribution statistically, the ranging value is considered contaminated. A three-level tiered processing strategy allows mildly contaminated ranging values ​​to continue participating in updates with reduced weight rather than being completely discarded, achieving a fine balance between fully utilizing observational information and preventing outlier distortion. When the UWB ranging signal is interrupted due to occlusion, the filter switches to open-loop inertial estimation and continuously stores the nominal state history sequence. Simultaneously, it monitors acceleration and angular velocity magnitudes to trigger zero-velocity constraints. This design effectively suppresses the accumulation of open-loop drift by utilizing the frequent, brief pauses of the chainsaw operator during operation. After signal recovery, a smooth recursive error state is performed in reverse time, optimally distributing the drift error accumulated during the interruption along the entire trajectory. The final output is a continuous and seamless pose trajectory across all time periods, thus resolving potential abrupt changes in position estimation after occlusion recovery.

[0025] Furthermore, in a preferred embodiment of the present invention, the step of calculating the real-time engine speed and dividing the operating condition interval, extracting vibration spectrum features from the IMU acceleration frames and constructing a multi-dimensional feature vector to identify the current operating state and generate operating state identification information specifically includes: The throttle position is obtained by comparing the level combination of the throttle shaft position switch with the preset encoding mapping table. At the same time, the pulse jump event of the magneto charging coil is captured and the interval between adjacent pulses is measured by the hardware timer. After conversion, the real-time engine speed is obtained by averaging through a sliding window. The idle speed, medium speed transition and high speed operating conditions are divided by combining the calibration threshold. The terminal's built-in IMU acquires triaxial acceleration signals at a fixed sampling rate. After removing high-frequency noise through a digital low-pass filter, the signals are divided into frames according to a preset frame length and frame shift. A Hanning window function is applied to each frame of data, and a fast Fourier transform is performed to convert the time-domain vibration signal into a triaxial amplitude spectrum. The average value of the triaxial spectrum characteristics is taken as the spectrum representation of the current frame. Locate the main peak frequency corresponding to the maximum amplitude in the spectrum representation and record the main peak amplitude. Calculate the ratio of the energy of the first three harmonics to the total energy and the ratio of the energy of the frequency band above the preset high-frequency threshold to the energy of the entire frequency band as the high-frequency band energy ratio. At the same time, extract the zero-crossing rate and kurtosis from the time-domain acceleration signal to form a multi-dimensional feature vector. The multidimensional feature vector is input into a pre-trained offline support vector machine classifier to identify the current working state. The posterior probability of the current frame belonging to one of four modes is output: idling, high speed without load, sawing load, or shutdown. The posterior probability of multiple consecutive frames is then subjected to sliding window multi-vote weighted filtering to obtain the frame-level identification result. The frame-level identification results are cross-validated with the operating condition range determined by real-time speed and throttle position. When there is a contradiction between the identification results and the operating condition range, the results are corrected according to the prior rules, and finally the cross-validated operating status identification information is obtained.

[0026] It should be noted that chainsaws exhibit significantly different dynamic and kinematic characteristics at different stages of operation, directly impacting the shape of the safety zone and protection strategies. At idle, the operator is mostly moving or observing, and the safety zone should primarily focus on protecting the body from collisions. Under sawing load, the chainsaw guide plate rotates at high speed and the operator leans forward, requiring the safety zone to extend significantly along the logging direction and cover dangerous paths where trees might fall. Therefore, accurately identifying the current operating state is a prerequisite for adaptive switching of the safety zone.

[0027] This scheme employs a dual-source identification architecture that fuses signals from the fuel power system and vibration signals. These two signals are physically complementary: engine speed and throttle position reflect the engine's operating range from the power source, while the vibration spectrum reveals whether the saw chain is actually cutting into the wood from a mechanical response perspective. While relying solely on engine speed can distinguish between idling and high speed, it's difficult to effectively differentiate between unloaded high speed and sawing load. In the former, the saw chain idles and only overcomes frictional resistance, while in the latter, the saw chain undergoes periodic impacts and cutting with the wood fibers. Although their speeds may be similar, their vibration characteristics differ significantly. Specifically, at idle, the engine maintains only minimal stable operation, with vibration energy concentrated at the engine's fundamental frequency and its first few harmonics, resulting in a narrow-band concentrated spectrum. At unloaded high speed, the engine speed increases, but the load torque is minimal, causing the spectral peak to shift to higher frequencies while the harmonic structure remains regular. Under sawing load conditions, the random impacts between the saw chain and the wood excite significant continuous high-frequency burr energy in the spectrum. Simultaneously, due to the periodic disturbance of engine speed by the cutting force, characteristic sidebands appear near the main peak frequency, and the kurtosis of the time-domain signal increases significantly due to the presence of the impact component. Multidimensional feature vectors jointly encode the discriminative information in the frequency and time domains, enabling the classifier to find an effective decision boundary for segmenting different operation modes in a high-dimensional feature space.

[0028] The posterior probability output by the classifier needs to be processed by a sliding window multi-vote weighted filter. This is significant in suppressing classification oscillations that may occur during transient changes in operating conditions within a single frame's spectrum. Each frame within the window is assigned a different voting weight based on its freshness, with more recent frames receiving a higher weight to balance recognition response speed and stability. Cross-validating the vibration identification results with the operating condition range determined jointly by engine speed and throttle position utilizes the independence of the two signal physical sources to achieve error detection and fault-tolerant correction. For example, if the engine speed is clearly in the idle range but the spectrum classifier misclassifies it as a sawing load due to occasional vibration spikes, the prior rules can immediately correct the contradictory results. The operating state identification information generated through this process provides a reliable state basis for the morphological switching of the safe zone and the dynamic adjustment of the graded response threshold.

[0029] Furthermore, in a preferred embodiment of the present invention, the construction of a collaborative situation topology map, the assessment of intrusion risk situation based on personnel-machine conflict and tree-chain risk discrimination based on the collaborative situation topology map, and the implementation of graded fault-tolerant security measures based on the intrusion risk situation assessment results, specifically include: The status packets broadcast by neighboring nodes are received through a distributed spatiotemporal synchronous communication network. The current position coordinates, velocity vector, attitude quaternion, operation mode label, proposed azimuth angle, and safe zone boundary parameters of the neighboring nodes are parsed from the status packets. The neighbor node status table is dynamically maintained in local memory using the neighbor node identifier as an index. At the same time, tree tag information is received and parsed, and personnel node information and tree node information are combined into a collaborative situational topology map in a unified coordinate system. Traverse the current position coordinates of all neighboring nodes in the neighboring node status table, convert the neighboring node positions to the local safe area coordinate system and calculate the penetration depth relative to each layer boundary. If the neighboring node position is in the core braking zone, generate a level 1 braking command; if it is in the severe alarm zone, generate a level 2 alarm command; if it is in the warning zone, generate a level 3 warning command. Extract the directional axis vector of the neighboring node that is under sawing load or has locked the directional azimuth of the planned felling from the neighboring node status table. Project the directional axis of the machine and the directional axis of the neighboring node onto the horizontal plane and calculate the length of the common perpendicular segment. When the length of the common perpendicular segment is less than the preset cross distance threshold and the included angle of the axis is greater than the cross angle critical value, generate a two-way cross conflict alarm command. Taking the tree nodes already marked on the local machine in the collaborative situation topology map as the object, the current position of the auxiliary personnel node in the topology map is traversed. The shortest distance from the auxiliary personnel node to the directed line segment determined by the tree root point position and the backward unit vector is calculated by the vector projection method. When the shortest distance is less than the tree crown width safety radius, a chain risk alarm instruction is generated. All generated risk events are prioritized according to risk type and penetration depth to obtain an event sorting table. Based on the event sorting table, a hierarchical fault-tolerant safety handling measure is generated to protect the safety of cluster operations. The hierarchical fault-tolerant safety handling measure includes the handling strategy for any event among the first-level braking event, the second-level alarm event, and the third-level prompt event.

[0030] It should be noted that the safety risks of multi-chain saw operations in forest areas are diverse and interconnected. These risks include not only direct physical collisions between operators, but also conflicts caused by improper logging direction planning leading to overlapping and collisions, and the risk of chain-like secondary disasters involving support personnel along the path of fallen trees. Existing solutions only focus on a single threshold judgment of the distance between personnel, which cannot cover the aforementioned multiple types of risks. This solution constructs a collaborative situational topology map, relating personnel nodes and tree nodes in a unified coordinate system, enabling the unified detection and tiered handling of various risks within the same framework.

[0031] Specifically, the neighbor node information parsed from the status packet by the terminal covers multiple dimensions such as position, speed, attitude, operation mode, and planned felling azimuth, rather than just obtaining position coordinates. A neighbor node status table is dynamically maintained in local memory using neighbor node identifiers as indexes, updated once per superframe cycle to ensure the situational topology map remains synchronized with the physical scene. Tree marking information is triggered by the operator bringing the guide plate root close to the tree trunk base, binding the tree root coordinates to the locked falling direction, so that the risk path of falling trees can be accurately represented as directed line segments in the topology map, rather than a vague azimuth estimate. This is followed by progressive conflict detection. The first layer, direct spatial conflict detection, uses the local safe zone as the core, calculating the penetration depth of neighbor nodes to each layer boundary. The greater the penetration depth, the deeper the intrusion, and the higher the corresponding handling level. The second layer, bidirectional falling direction intersection conflict detection, addresses the problem of two chainsaws planning to fall in directions that intersect. This intersection means that the tree falling paths planned by the operators overlap spatially; if both are sawing simultaneously, one's tree may fall towards the other's operating position. The length of the common perpendicular segment is calculated by projecting the falling axis onto a horizontal plane, and the intersection angle is used as an auxiliary judgment condition to avoid parallel but closely spaced axes being misjudged as intersecting conflicts. The third-level tree chain-link risk detection uses vector projection to calculate the shortest distance from the assistant to the directed line segment determined by the tree root and the falling direction. The projected scalar is used to determine whether the assistant is within the effective range of the falling direction. If within the range, the perpendicular distance from the point to the line is taken; otherwise, the smaller distance from the point to the two endpoints is taken. This calculation method accurately delineates the danger zone within the tree canopy area.

[0032] The core of graded fault-tolerant safety handling lies in the differentiated intervention intensity corresponding to different risk levels. Level 1 braking directly acts on the ignition control interface to forcibly reduce the engine speed to below the clutch engagement speed, representing the highest level of non-shutdown emergency braking. This ensures the saw chain stops rotating without completely shutting down the engine, allowing the operator to quickly resume work. Level 2 alarms utilize a two-way linkage between the local audible and visual alarms and coordinated alarms or evacuation guidance commands sent to adjacent nodes at risk, ensuring both parties receive warning information simultaneously and reducing the delay in manual communication. Level 3 alerts trigger only low-frequency warning signals, avoiding operator fatigue and neglect due to frequent high-intensity alarms. Before executing the power cut-off command, a mandatory multi-sensor consistency verification is performed, confirming the authenticity of the risk from three independent dimensions: distance measurement confidence, vibration intensity, and intrusion duration. This effectively prevents false braking caused by momentary sensor anomalies or adjacent nodes briefly crossing the boundary, ensuring a balance between safety protection and operational continuity.

[0033] Figure 2 This is a flowchart of an asymmetric safe region boundary construction method according to an embodiment of the present invention; like Figure 2 As shown in the flowchart, this invention provides a method for constructing asymmetric security region boundaries, including: After receiving the direction setting trigger signal, the terminal enters the attitude monitoring window. The built-in IMU outputs the three-axis angular velocity vector at a fixed sampling rate. The vector is filtered by the mean of the sliding window to suppress high-frequency vibration glitches. The filter window width is set according to the vibration period corresponding to the engine base frequency when the chainsaw is idling. The output is a low-frequency angular velocity component that filters out vibration interference. The low-frequency angular velocity vector is input into a three-state finite state machine for intent recognition. The state machine is initially in a waiting state. When the angular velocity magnitude exceeds the start threshold but does not exceed the upper limit of violent motion, it transitions to the action in progress state and starts an integral timer to accumulate and integrate each component of the angular velocity vector over time. If the angular velocity magnitude exceeds the upper limit of violent motion, it is reset to the waiting state and the integral value is cleared. When the angular velocity magnitude falls below the static judgment threshold and continues to reach the preset confirmation dwell time, the state machine enters the action completion confirmation state, and the projection direction of the accumulated net angular displacement vector on the horizontal plane is used as the pseudo-azimuth angle of the operator's intention, which is locked and updated in the terminal direction parameter memory. Read the current mode label in the job status identification information. When the mode label is idle or no-load high-speed mode, set the safety area shape to the limb dynamic ellipsoid mode. When the mode label is sawing load mode or the proposed felling azimuth angle is locked, switch to the logging directional asymmetric layering mode. During the shape switching, interpolate and mix the old and new boundary point sets to avoid boundary abrupt changes. When using the limb dynamic ellipsoid mode, the posture rotation matrix is ​​determined by the torso center in the pose trajectory information of this node as the origin and the body orientation quaternion. The three-axis half lengths are determined by the shoulder lateral margin, the forward extension amount of the arm extension angle and the guide plate length superimposed, and the torso height, respectively. A discretized polar coordinate point set array is generated as the boundary of the safe area. When adopting the logging directional asymmetric layered mode, a polar coordinate system is established with the root position of the guide plate as the pole and the azimuth angle of the proposed felling as the polar axis. Three layers of boundaries are generated: the core braking zone, the severe alarm zone, and the warning zone. Gaussian modulation is applied to the radius of each layer boundary so that the polar axis direction reaches its maximum value and decays as the deviation angle increases, forming an asymmetric boundary shape. Simultaneously, the velocity vector in the node position trajectory information is used to shift the center of the asymmetric shape boundary along the velocity direction, so that the safe area always leads the operator's movement trend. A random expansion amount proportional to the body sway amplitude detected by the IMU is applied to the area contour, and finally the boundary is discretized to form the boundary of the asymmetric safe area.

[0034] It should be noted that high-frequency vibrations at the chainsaw operation site are the primary interference factor hindering the accurate identification of the operator's intentions. When idling, the engine typically operates continuously at a fundamental frequency of tens of hertz. The vibration acceleration transmitted through the crankshaft and engine block to the handle manifests as periodic spikes in the IMU's angular velocity output at the same frequency band. The amplitude of these spikes is sufficient to drown out the low-frequency angular velocity signal generated by the operator's slow body movements. This solution sets the window width of the sliding window mean filter based on the vibration period corresponding to the engine's fundamental frequency when the chainsaw is idling. The design intent is to ensure that the window precisely covers the duration of a complete vibration cycle. The positive and negative half-cycles of the vibration spikes within this cycle are weighted and then canceled out. This filters out vibration interference while preserving, to the greatest extent possible, the low-frequency components of the operator's intentional body movements. This parametric design ensures good compatibility of the filter across different chainsaw models.

[0035] Chainsaw operators exhibit diverse movements at work sites, such as slowly turning their bodies to point at trees, quickly waving their arms to signal colleagues, bending down to pick up tools, or unconsciously swaying due to uneven ground. If only a single threshold is used to judge based on angular velocity amplitude, non-directional movements can easily be misidentified as directional settings. The state machine distinguishes between slow directional turns and rapid arm swings through a dual-layer threshold system of an initiation threshold and a maximum threshold for violent movement. It also ensures the operator is stably positioned in the desired direction before locking the azimuth angle by confirming the dwell time condition. This three-stage logic of "action initiation—continuous monitoring—stability confirmation" effectively eliminates interference from momentary swaying and rapid movements. The reliability of the identification results directly affects the consistency between the subsequent safe zone direction and the actual felling hazard zone. When the chainsaw is idling or operating at high speed without load, before cutting into the wood, the risk primarily comes from direct physical contact between personnel. The limb dynamic ellipsoid incorporates the arm extension range, guide plate length, and body orientation into the model, using anisotropic ellipsoids to accurately match the actual spatial occupancy contour of the operator while sawing. Once the sawing load or the falling direction is locked, the risk center shifts to a chain reaction of disasters along the felling direction. At this point, the system switches to an asymmetric layered region with the root of the guide plate as the pole and the azimuth angle of the planned fall as the polar axis. Gaussian modulation is used to maximize the boundary along the polar axis and exponentially decrease it with the deviation angle, forming a shape that extends forward along the felling direction and contracts laterally. During the shape switching, the old and new boundary point sets are interpolated and mixed to prevent the safe zone boundary from abruptly changing at the moment of switching, which could trigger instantaneous false alarms or false braking. The speed look-ahead offset ensures that the safe zone always leads the operator's movement trend, while the body sway expansion accommodates the natural micro-movements when standing on rugged terrain. Together, these factors enhance the protection margin of the safe zone from static geometric coverage to dynamic motion compensation, effectively reducing the risk of delayed intrusion caused by the operator's movement inertia without increasing the base radius.

[0036] Furthermore, the multi-chainsaw cluster operation safety protection method based on regional collaborative perception provided by the present invention also includes the following steps: During each period of collaborative situational update, each terminal encapsulates the current time slot's position coordinates, velocity vector, operation mode label, triggered risk event type, and corresponding neighbor node identifier into a log record, attaches a unified timestamp across the entire network, and writes it into the terminal's built-in memory in a cyclic overwrite manner. After the day's work is completed, each terminal exports the stored log records in batches to the portable management terminal via wired data interface or short-range wireless transmission. The management terminal performs multi-source fusion and alignment of the log records of all participating nodes according to the unified timestamp of the entire network, and reconstructs the continuous movement trajectory and risk event sequence of each node throughout the entire time period. The work area is divided into uniform grid cells according to a preset spatial resolution. The number of triggers of risk events at all levels in each grid cell is counted throughout the time period and weighted according to the event level. The risk heat value of each grid cell is obtained by weighted summation and mapped into a two-dimensional risk heat map with color gradient. The operation mode duration distribution, risk event trigger frequency and response type of each node are statistically analyzed to generate a single node behavior profile and mark high-risk operation habits that exceed a preset frequency threshold. The high-risk operation habits include entering the sawing load mode without locking the direction or failing to adjust the position when the neighboring node is constantly approaching. The high-risk hotspots marked on the two-dimensional risk heat map are used as spatial references for the planning of the next day's work passage and the setting of safety distances. Risk operation clustering is performed based on single-node behavior profiles, and enhanced training strategies are generated based on the clustering results to assist in the skill assessment and safety training of operators.

[0037] It should be noted that during chainsaw cluster operations, the collaborative situational data generated by each terminal each cycle is not only used for real-time conflict detection, but the spatiotemporal behavioral information it contains also has significant offline analysis value. Existing safety solutions do not archive or backtrack process data after the end of the day's work, resulting in the direct discarding of a large amount of historical information that can be used for risk prevention. This solution writes key fields such as position coordinates, velocity vectors, work mode labels, and types of triggered risk events for each cycle into the built-in storage of each terminal in a cyclic overwrite manner. The addition of a unified timestamp across the entire network allows asynchronous logs across nodes to be accurately aligned along the timeline during post-event fusion without relying on additional synchronization operations.

[0038] After the operation is completed, logs are exported in batches to a portable management terminal via wired or wireless means. Then, multi-source fusion and alignment are performed using a unified timestamp across the entire network to reconstruct the continuous movement trajectories and complete risk event sequences of each node throughout the entire time period. This archiving mechanism aggregates distributed multi-node data into a unified spatiotemporal database, allowing global risk patterns that might otherwise be imperceptible due to the limited field of view of a single node to emerge during the operation. Based on the reconstructed data, the operation area is divided into uniform grid units, and the number of triggers for each level of risk event is counted. Different weights are assigned to events according to their severity, and a weighted summation is performed. The resulting two-dimensional risk heatmap visually marks high-risk hotspots such as narrow passages, blind spots, or areas where operation paths frequently intersect using color gradients. Single-node behavior profiling automatically identifies high-risk operating habits such as entering sawing loads without locking the direction of the fall by statistically analyzing the distribution of operation mode duration and the frequency of risk event triggers. Heat map results directly guide the planning of the next day's work passage and the setting of safety distances, while behavioral profiles drive the targeted development of enhanced training strategies, transforming process data from a single operation into a quantitative basis for continuous improvement of safety management, forming a complete closed loop from real-time intervention during the operation to post-event analysis and pre-event prevention.

[0039] Furthermore, the multi-chainsaw cluster operation safety protection method based on regional collaborative perception provided by the present invention also includes the following steps: Each terminal continuously monitors the triaxial acceleration modulus and angular velocity modulus output by the IMU. When the acceleration modulus of a node continuously exceeds the fall judgment threshold within a preset time and the angular velocity modulus shows a rotation pattern that matches the rolling characteristics of the human body, it automatically determines that the operator has suffered a serious injury and fallen to the ground. It then broadcasts the highest priority emergency event message carrying the location coordinates of the fallen node and the event type identifier through the collaborative communication network. After receiving the emergency event message, the neighboring node enters the collaborative evacuation mode. Using the current position of each node in the collaborative situation topology map, the root point of the marked tree and the azimuth angle of the planned felling, the node takes its current position as the starting point and defines the location of the emergency event source and the root point and canopy coverage of each marked tree to be felled as the repulsive potential source. The safe open direction that has been explored in the work area is defined as the attractive potential source. The direction of the resultant force is calculated by superimposing the repulsive potential field and the attractive potential field to generate a personalized evacuation direction vector for this node. The strength of the repulsive potential field is inversely proportional to the distance from the source and directly proportional to the risk level. The strength of the attractive potential field is determined by weighting based on the accessibility of the safe and open direction, so that the evacuation path avoids both the direct threat of emergency events and the fan-shaped danger zone caused by fallen trees. The evacuation direction vector is converted into a steering angle command with reference to the current orientation of the node. The terminal LED arrow array indicates to the operator to evacuate to the left, right, or straight ahead. At the same time, the piezoelectric buzzer is driven to emit a special evacuation frequency conversion tone that is different from the conventional alarm tone, so as to ensure that the operator can accurately identify the evacuation signal in a high-noise environment.

[0040] It should be noted that, in addition to the usual collision risks, forest work sites also present extreme situations where operators may suffer serious injuries and fall due to unexpected tree falls, sudden strong gusts of wind, or missteps in terrain. In such cases, the injured person loses the ability to evacuate on their own, and nearby colleagues often cannot detect or locate the injured person's position immediately due to obstruction by trees and noise interference, which can easily delay rescue opportunities and trigger secondary accidents. Existing chainsaw safety devices are completely incapable of sensing and responding to such emergencies.

[0041] This solution continuously monitors the three-axis acceleration and angular velocity moduli output by the IMU during the regular collaborative situational update cycle. When the acceleration modulus continuously exceeds the fall detection threshold for a preset period of time, and the angular velocity modulus exhibits a rotational pattern consistent with a person rolling or tipping after losing consciousness, the terminal automatically determines that the operator has suffered a serious injury and fallen to the ground, without the injured person needing to actively trigger any buttons or call for help. After the determination is confirmed, the terminal immediately broadcasts the highest priority emergency event message. This message is sent first in a preemptive manner, and the coordinates of the fall point location carried in the message are provided by unified pose trajectory information, avoiding the problem of being unable to locate the injured person due to the lack of GPS signal.

[0042] Upon receiving an emergency event message, a neighboring node utilizes the current positions of all nodes in the constructed collaborative situational topology map, along with the root points and orientation information of marked trees, to generate a personalized evacuation direction vector for that node using a virtual potential field method. The core advantage of the virtual potential field lies in simultaneously incorporating multiple hazard sources into a unified computational framework. The location of the emergency event source, the root point of each marked tree to be felled, and its canopy coverage all serve as repulsive potential field sources. The intensity of these repulsive potential field sources is inversely proportional to distance, ensuring the evacuation path naturally avoids all hazardous areas. Meanwhile, the identified safe and open directions within the work area act as attractive potential field sources, guiding operators towards safety. The combined force of these two sources serves both obstacle avoidance and guidance functions. Personalized evacuation directions naturally differ based on the location of each node and the distribution of surrounding risks, preventing secondary disasters caused by crowding and scrambling in the same direction during panicked evacuations. The entire collaborative evacuation process is completed autonomously within a distributed architecture, sharing the same communication network and situational topology information with real-time collision detection, without incurring any additional hardware costs. This extends the safety protection system from ongoing collision warnings to ensuring the survival of the entire group during extreme emergencies.

[0043] Figure 3 A multi-chainsaw cluster operation safety protection system 3 based on regional collaborative perception is provided in an embodiment of the present invention. The system includes: a memory 301, a processor 302, and a communication interface 303. The memory 301 contains a multi-chainsaw cluster operation safety protection method program based on regional collaborative perception. When the multi-chainsaw cluster operation safety protection method program based on regional collaborative perception is executed by the processor 302, it implements the steps of the multi-chainsaw cluster operation safety protection method based on regional collaborative perception as described above.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A safety protection method for multi-chainsaw cluster operations based on regional collaborative perception, characterized in that, include: Deploy collaborative sensing terminals at each chainsaw node, establish a time-division multiple access communication network based on beacon frame interaction and distributed clock synchronization, and broadcast heartbeat packets carrying ranging confidence parameters to form a distributed spatiotemporal synchronization communication network. Based on the distributed spatiotemporal synchronous communication network, the relative position topology is initialized by multidimensional scaling transformation, cooperative relative positioning is performed under the error state Kalman filter framework, and multipath outliers are suppressed by Mahalanobis distance criterion to obtain node pose trajectory information. The system collects the level combination of the throttle shaft position switch of the target chainsaw and the pulse frequency of the magneto charging coil, calculates the real-time engine speed and divides the working condition interval, extracts the vibration spectrum features from the IMU acceleration frame and constructs a multi-dimensional feature vector to identify the current working state and generate working state identification information. The evolution of the IMU angular velocity vector is monitored and the operator's intention is confirmed by a finite state machine to lock the azimuth angle of the proposed felling. Based on the operation status identification information, the operation status switching situation is determined, and the asymmetric safety area boundary is constructed by combining the node pose trajectory information. The system receives status packets broadcast by neighboring nodes and constructs a collaborative situation topology graph. Based on the collaborative situation topology graph, it performs personnel-machine conflict and tree-chain risk assessment to evaluate the intrusion risk situation. Based on the intrusion risk situation assessment results, it performs hierarchical fault-tolerant security measures.

2. The method for safety protection of multi-chainsaw cluster operations based on regional collaborative perception according to claim 1, characterized in that, The deployment of collaborative sensing terminals at each chainsaw node, establishing a time-division multiple access communication network based on beacon frame interaction and distributed clock synchronization, and broadcasting heartbeat packets carrying ranging confidence parameters, forms a distributed spatiotemporal synchronization communication network, specifically including: After each collaborative sensing terminal is powered on, it traverses the preset working frequency band to perform channel scanning, captures the preamble of the neighboring node's transmitted frame and parses the physical layer header to extract the device identifier and terminal type fields, adds unrecorded identifiers to the neighbor list, and updates the received timestamp of existing identifiers. After multiple complete cycle scans, a neighbor device list is generated. Based on the neighbor device list, a temporary master clock source is determined by sorting the identifier values. The master clock source broadcasts a synchronization beacon frame carrying the current clock count value in the first available time slot. After receiving the synchronization beacon frame, the neighboring node latches the reception time and extracts the transmission timestamp of the master clock source. Combined with the preset delay, the instantaneous clock offset between the local device and the master clock is calculated. The instantaneous clock offset is input into an exponentially weighted moving average filter. The smoothing factor is adaptively adjusted by the variance of multiple consecutive offset measurements. When the variance increases, the smoothing factor is reduced to suppress outlier shocks. When the variance decreases, the smoothing factor is increased to quickly track clock drift. The smoothed clock offset estimate is then output. The local voltage-controlled oscillator frequency compensation is gradually adjusted using the smoothed clock offset estimate to fine-tune the clock rate. Each node transmits the corrected clock count value to the next hop node. After multiple rounds of recursive broadcasting and feedback adjustment, the clock synchronization error of the entire network converges to within the preset error threshold, forming a unified time base for the entire network. The timeline is divided into equal-length superframes and further subdivided into several time slots. Each node is sorted by identifier and uses a distributed negotiation mechanism to complete time slot declaration and conflict resolution. After determining that there are no conflict-free time slots, each node periodically broadcasts a heartbeat packet carrying the clock count value and ranging confidence parameter within its own time slot of each superframe, thus forming a distributed spatiotemporal synchronization communication network.

3. The safety protection method for multi-chainsaw cluster operations based on regional collaborative perception according to claim 1, characterized in that, The process, based on the distributed spatiotemporal synchronous communication network, initializes the relative position topology using multidimensional scaling transformation, performs cooperative relative positioning within an error-state Kalman filter framework, and suppresses multipath outliers using the Mahalanobis distance criterion to obtain unified pose trajectory information for nodes. Specifically, this includes: Each terminal summarizes the distance values ​​collected from its own node and its neighboring nodes in the most recent ranging periods into a distance matrix after confidence weighting and smoothing. Using the local node as the reference, the matrix is ​​converted into an inner product matrix through double centering and then eigenvalue decomposition is performed to extract the maximum eigenvalue and the corresponding eigenvector, thus obtaining the initial relative position topology of all nodes in the network. An error state Kalman filter framework is introduced to perform cooperative relative positioning based on the initial relative position topology of all network nodes. The error state Kalman filter framework splits the state estimation into two parallel processing links: nominal state and error state. The nominal state link uses the initial coordinates and preset static attitude provided by the initial relative position topology as the initial nominal state values, and uses IMU angular velocity and acceleration data to drive the inertial mechanical arrangement equations to recursively update the node's three-dimensional position, velocity and attitude quaternions according to the sampling period. The error state link maintains an error state vector that includes position error, velocity error, attitude misalignment angle, and gyroscope zero bias residual. In each IMU cycle, it propagates the error covariance matrix based on the state transition matrix and the process noise covariance matrix, and scales the diagonal terms of the process noise covariance matrix according to the vibration energy level of the current IMU output. Whenever this node receives the broadcast position of a neighboring node and completes a bilateral two-way ranging, it calculates the observation residual using the Euclidean distance between the nominal position of this node and the broadcast position of the neighboring node as the predicted observation value. It then calculates the innovation covariance matrix using the error state covariance matrix and the observation matrix, and performs a Mahalanobis distance test on the observation residual. If the Mahalanobis distance is below the chi-square critical value, the error state vector is updated according to the standard Kalman gain. If it exceeds the chi-square critical value but does not reach the rejection threshold, an inflation factor proportional to the degree of exceeding is applied to the observation noise covariance. If it exceeds the rejection threshold, the update is skipped directly. After the update is completed, the nominal state is corrected with the error state and the error state vector is returned to zero. When a terminal temporarily interrupts the ranging signal due to tree obstruction or exceeding the communication range, it performs open-loop inertial estimation. The nominal state is only driven by IMU data and no longer receives observation updates. During this period, the error state link continues to propagate the covariance but suspends updates, and the terminal continuously stores the nominal state of each cycle into the local cache. If the IMU acceleration and angular velocity magnitude are below the static threshold for several consecutive cycles, a virtual zero-velocity constraint is triggered to suppress position drift. After the signal is recovered, the error state smoothing recursion is performed on the historical state sequence in reverse, starting from the state after the interruption is recovered. The continuous, non-jumping full-time node pose trajectory is output, and finally the pose trajectory information of each node is obtained.

4. The safety protection method for multi-chainsaw cluster operations based on regional collaborative perception according to claim 1, characterized in that, The process of calculating real-time engine speed and dividing operating condition intervals, extracting vibration spectrum features from IMU acceleration frames and constructing multi-dimensional feature vectors for current operation status identification, and generating operation status identification information specifically includes: The throttle position is obtained by comparing the level combination of the throttle shaft position switch with the preset encoding mapping table. At the same time, the pulse jump event of the magneto charging coil is captured and the interval between adjacent pulses is measured by the hardware timer. After conversion, the real-time engine speed is obtained by averaging through a sliding window. The idle speed, medium speed transition and high speed operating conditions are divided by combining the calibration threshold. The terminal's built-in IMU acquires triaxial acceleration signals at a fixed sampling rate. After removing high-frequency noise through a digital low-pass filter, the signals are divided into frames according to a preset frame length and frame shift. A Hanning window function is applied to each frame of data, and a fast Fourier transform is performed to convert the time-domain vibration signal into a triaxial amplitude spectrum. The average value of the triaxial spectrum characteristics is taken as the spectrum representation of the current frame. Locate the main peak frequency corresponding to the maximum amplitude in the spectrum representation and record the main peak amplitude. Calculate the ratio of the energy of the first three harmonics to the total energy and the ratio of the energy of the frequency band above the preset high-frequency threshold to the energy of the entire frequency band as the high-frequency band energy ratio. At the same time, extract the zero-crossing rate and kurtosis from the time-domain acceleration signal to form a multi-dimensional feature vector. The multidimensional feature vector is input into a pre-trained offline support vector machine classifier to identify the current working state. The posterior probability of the current frame belonging to one of four modes is output: idling, high speed without load, sawing load, or shutdown. The posterior probability of multiple consecutive frames is then subjected to sliding window multi-vote weighted filtering to obtain the frame-level identification result. The frame-level identification results are cross-validated with the operating condition range determined by real-time speed and throttle position. When there is a contradiction between the identification results and the operating condition range, the results are corrected according to the prior rules, and finally the cross-validated operating status identification information is obtained.

5. A safety protection method for multi-chainsaw cluster operations based on regional collaborative perception according to claim 1, characterized in that, The process involves monitoring the evolution of the IMU angular velocity vector and confirming the operator's intention via a finite state machine to lock the proposed azimuth angle. Based on the operational status identification information, the operational status switching situation is determined. Combined with node pose trajectory information, an asymmetric safety zone boundary is constructed. Specifically, this includes: After receiving the direction setting trigger signal, the terminal enters the attitude monitoring window. The built-in IMU outputs the three-axis angular velocity vector at a fixed sampling rate. The vector is filtered by the mean of the sliding window to suppress high-frequency vibration glitches. The filter window width is set according to the vibration period corresponding to the engine base frequency when the chainsaw is idling. The output is a low-frequency angular velocity component that filters out vibration interference. The low-frequency angular velocity vector is input into a three-state finite state machine for intent recognition. The state machine is initially in a waiting state. When the angular velocity magnitude exceeds the start threshold but does not exceed the upper limit of violent motion, it transitions to the action in progress state and starts an integral timer to accumulate and integrate each component of the angular velocity vector over time. If the angular velocity magnitude exceeds the upper limit of violent motion, it is reset to the waiting state and the integral value is cleared. When the angular velocity magnitude falls below the static judgment threshold and continues to reach the preset confirmation dwell time, the state machine enters the action completion confirmation state, and the projection direction of the accumulated net angular displacement vector on the horizontal plane is used as the pseudo-azimuth angle of the operator's intention, which is locked and updated in the terminal direction parameter memory. Read the current mode label in the job status identification information. When the mode label is idle or no-load high-speed mode, set the safety area shape to the limb dynamic ellipsoid mode. When the mode label is sawing load mode or the proposed felling azimuth angle is locked, switch to the logging directional asymmetric layering mode. During the shape switching, interpolate and mix the old and new boundary point sets to avoid boundary abrupt changes. When using the limb dynamic ellipsoid mode, the posture rotation matrix is ​​determined by the torso center in the pose trajectory information of this node as the origin and the body orientation quaternion. The three-axis half lengths are determined by the shoulder lateral margin, the forward extension amount of the arm extension angle and the guide plate length superimposed, and the torso height, respectively. A discretized polar coordinate point set array is generated as the boundary of the safe area. When adopting the logging directional asymmetric layered mode, a polar coordinate system is established with the root position of the guide plate as the pole and the azimuth angle of the proposed felling as the polar axis. Three layers of boundaries are generated: the core braking zone, the severe alarm zone, and the warning zone. Gaussian modulation is applied to the radius of each layer boundary so that the polar axis direction reaches its maximum value and decays as the deviation angle increases, forming an asymmetric boundary shape. Simultaneously, the velocity vector in the node position trajectory information is used to shift the center of the asymmetric shape boundary along the velocity direction, so that the safe area always leads the operator's movement trend. A random expansion amount proportional to the body sway amplitude detected by the IMU is applied to the area contour, and finally the boundary is discretized to form the boundary of the asymmetric safe area.

6. A safety protection method for multi-chainsaw cluster operations based on regional collaborative perception according to claim 1, characterized in that, The construction of a collaborative situation topology map, based on which personnel-machine conflict and tree-chain risk identification are used to assess intrusion risk situation, and hierarchical fault-tolerant security measures are implemented according to the intrusion risk situation assessment results, specifically including: The status packets broadcast by neighboring nodes are received through a distributed spatiotemporal synchronous communication network. The current position coordinates, velocity vector, attitude quaternion, operation mode label, proposed azimuth angle, and safe zone boundary parameters of the neighboring nodes are parsed from the status packets. The neighbor node status table is dynamically maintained in local memory using the neighbor node identifier as an index. At the same time, tree tag information is received and parsed, and personnel node information and tree node information are combined into a collaborative situational topology map in a unified coordinate system. Traverse the current position coordinates of all neighboring nodes in the neighboring node status table, convert the neighboring node positions to the local safe area coordinate system and calculate the penetration depth relative to each layer boundary. If the neighboring node position is in the core braking zone, generate a level 1 braking command; if it is in the severe alarm zone, generate a level 2 alarm command; if it is in the warning zone, generate a level 3 warning command. Extract the directional axis vector of the neighboring node that is under sawing load or has locked the directional azimuth of the planned felling from the neighboring node status table. Project the directional axis of the machine and the directional axis of the neighboring node onto the horizontal plane and calculate the length of the common perpendicular segment. When the length of the common perpendicular segment is less than the preset cross distance threshold and the included angle of the axis is greater than the cross angle critical value, generate a two-way cross conflict alarm command. Taking the tree nodes already marked on the local machine in the collaborative situation topology map as the object, the current position of the auxiliary personnel node in the topology map is traversed. The shortest distance from the auxiliary personnel node to the directed line segment determined by the tree root point position and the backward unit vector is calculated by the vector projection method. When the shortest distance is less than the tree crown width safety radius, a chain risk alarm instruction is generated. All generated risk events are prioritized according to risk type and penetration depth to obtain an event sorting table. Based on the event sorting table, a hierarchical fault-tolerant safety handling measure is generated to protect the safety of cluster operations. The hierarchical fault-tolerant safety handling measure includes the handling strategy for any event among the first-level braking event, the second-level alarm event, and the third-level prompt event.

7. A safety protection system for multi-chainsaw cluster operations based on regional collaborative perception, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory contains a program for a multi-chain saw cluster operation safety protection method based on regional collaborative perception. When the program for the multi-chain saw cluster operation safety protection method based on regional collaborative perception is executed by the processor, it implements the steps of the multi-chain saw cluster operation safety protection method based on regional collaborative perception as described in any one of claims 1-6.

Images