Real-time data processing method of dynamic dws six-surface scanning system and medium

Abstract

The application provides a real-time data processing method and medium of a dynamic DWS six-surface scanning system, relates to the technical field of DWS scanning, and comprises the following steps: constructing a unified time reference, mapping original point cloud data collected by a DWS six-surface scanner, performing observation confidence fusion of each voxel grid according to the mapping result, performing conflict voxel detection, performing local reconstruction processing according to voxel anomaly identification, updating the observation confidence fusion result by using the local reconstruction result, performing global verification identification, converting the updated observation confidence fusion result into a structured description code for output, and solving the technical problems that the existing six-surface scanning cannot handle the multi-scan surface asynchronous sampling error based on time synchronization and linear speed compensation, and that the point cloud is dislocated and the structure reconstruction is inconsistent in a dynamic conveying scene. The unified dynamic space-time reference and voxel-level multi-source confidence fusion are constructed by using an encoder pulse, and the technical effects of improving the DWS measurement accuracy and reliability are achieved.

Description

Technical Field

[0001] This invention relates to the field of DWS scanning technology, specifically to a real-time data processing method and medium for a dynamic DWS six-sided scanning system. Background Technology

[0002] In automated logistics sorting and DWS six-sided measurement systems, packages are typically in continuous motion on a conveyor belt. While the six-sided scanning units are distributed around the conveyor belt, and different scanning heads are spatially synchronized, there are inherent temporal sampling differences; that is, the time at which each scanning surface acquires point cloud data is not entirely consistent. Furthermore, the conveyor belt speed is affected by load variations, drive control fluctuations, and mechanical transmission errors, exhibiting non-constant motion characteristics that cause continuous displacement changes in the package during scanning. Since point cloud data reflects instantaneous spatial sampling results, under dynamic motion conditions, the data collected by different scanning surfaces actually correspond to different time sections of the package, thus introducing a significant temporal misalignment problem.

[0003] However, existing technologies typically use fixed timestamp synchronization or simple displacement compensation based on average velocity for correction. However, since they cannot accurately depict the instantaneous motion state and local acceleration changes of the package, these methods are difficult to eliminate the cumulative effect of dynamic errors, resulting in the inaccurate alignment of point clouds between multiple scan surfaces, which in turn affects the accuracy of package size measurement and the reliability of structure identification.

[0004] In summary, existing DWS six-sided scanning has the technical problem of being unable to handle asynchronous sampling errors of multiple scanning surfaces based on time synchronization and linear speed compensation, resulting in point cloud misalignment and inconsistent structural reconstruction in dynamic transportation scenarios. Summary of the Invention

[0005] The purpose of this application is to provide a real-time data processing method and medium for a dynamic DWS six-sided scanning system, which solves the technical problem that existing DWS six-sided scanning systems have difficulty in handling asynchronous sampling errors of multiple scanning surfaces based on time synchronization and linear speed compensation, resulting in point cloud misalignment and inconsistent structural reconstruction in dynamic transportation scenarios.

[0006] In view of the above problems, this application provides a real-time data processing method and medium for a dynamic DWS six-sided scanning system.

[0007] The first aspect of this application provides a real-time data processing method for a dynamic DWS six-sided scanning system. This method includes: constructing a unified time reference based on conveyor belt displacement encoder pulses; mapping the raw point cloud data acquired by the DWS six-sided scanner to the same dynamic coordinate system to establish a mapping result; discretizing the dynamic coordinate system into a voxel grid; performing observation confidence fusion for each voxel grid based on the mapping result, the observation confidence fusion including viewpoint priority fusion and occlusion confidence attenuation fusion; performing conflict voxel detection using the observation confidence fusion result for each voxel grid to establish voxel anomaly identifiers; performing local reconstruction processing based on the voxel anomaly identifiers to establish local reconstruction results; updating the observation confidence fusion result using the local reconstruction result; performing global verification and identification; if the global verification and identification pass, converting the updated observation confidence fusion result into a structured DWS descriptor code for output.

[0008] Optionally, the viewpoint priority fusion layer is activated. After reading the angle between the scan surface normal and the estimated normal of the corresponding voxel grid surface, a first priority is established based on the square cosine of the angle. The scan distance from the scan surface to the center of the voxel grid is read, and after exponential decay processing of the scan distance, a second priority is established. The viewpoint priority weight is output using the first priority and the second priority. The occlusion confidence attenuation fusion layer is activated to perform spatial connectivity path detection between the current scan surface and the target voxel grid. If there is at least one intermediate voxel marked as occupied on the spatial connectivity path, the target voxel grid is determined to be in an occluded observation state. The spatial distance attenuation factor is calculated according to the occlusion observation state, and the occlusion attenuation coefficient is established by progressively multiplying according to the occluded voxel sequence. The observation confidence fusion is performed using the viewpoint priority weight and the occlusion attenuation coefficient to establish the observation confidence fusion result.

[0009] Optionally, historical confidence level data is obtained, and the historical confidence level data is decayed and retained according to a recursive update mechanism to establish a confidence level update item; the confidence level update item is used to perform confidence correction processing on the observation confidence fusion result to update the observation confidence fusion result.

[0010] Optionally, segmented judgment is performed on each voxel grid in the observation confidence fusion result: if the confidence change of the same voxel grid in adjacent time windows exceeds the dynamic fluctuation threshold, the first type of anomaly detection mechanism is triggered, and the corresponding voxel grid is marked as a confidence unstable voxel; consistency comparison is performed on the observation results of the same voxel grid under different scanning planes, and when the occupancy or idle state of the same voxel grid under different scanning planes shows opposite judgments, the second type of conflict detection mechanism is triggered, and the corresponding voxel grid is marked as a multi-source observation conflict voxel; consistency scanning is performed on the confidence distribution of adjacent voxel grids based on the spatial connectivity structure of the local neighborhood where the voxel grid is located, and when the target voxel grid and the neighboring voxel grid show topological breakage characteristics in occupancy continuity, the third type of structural anomaly detection mechanism is triggered, and the target voxel grid is marked as a spatial connectivity anomalous voxel; voxel anomaly identifiers are established based on confidence unstable voxels, multi-source observation conflict voxels, and spatial connectivity anomalous voxels.

[0011] Optionally, based on the voxel anomaly identifier, anomaly back-tracing is performed on the voxel mesh to establish an anomaly causal chain; all voxel meshes stably marked as occupied within the spatially connected domain of the anomalous voxel mesh are extracted, and boundary constraints are constructed based on the extraction results; a minimum energy manifold passing through the surfaces of all boundary voxels is constructed based on the boundary constraints using local implicit surface fitting, wherein the minimum energy manifold represents the predicted real physical surface wrapped in the corresponding local region; the expected occupancy state analysis of the anomalous voxel mesh is performed using the minimum energy manifold, and the analysis results are encoded as virtual observation evidence; the virtual observation evidence is back-injected into the historical observation records corresponding to each relevant scan surface along the anomaly causal chain to perform forced correction; and a local reconstruction result is established based on the forced correction result.

[0012] Optionally, the structured DWS description code includes outer contour topology coding, internal cavity structure marking, and size constraint vectorization representation.

[0013] Optionally, the pulse signal of the conveyor belt displacement encoder is read in real time, and a unified timestamp sequence is generated using the rising edge of each pulse signal as a global time synchronization reference. A nonlinear spatiotemporal mapping function with the cumulative number of encoder pulses as the independent variable is established. When the package enters the measurement area of ​​the DWS six-sided scanner, the motion phase offset is dynamically initialized according to the pulse signal and photoelectric trigger signal. Each point in the original point cloud collected by each scanning surface is transformed into a dynamic coordinate system according to the spatiotemporal mapping function. The transformed point cloud data is timestamp aligned and verified, and the aligned point cloud is output as the mapping result.

[0014] Optionally, the contour projection boundaries of the updated observation confidence fusion results in the relative scanning direction are extracted respectively to establish a set of opposing contour projections; the boundary overlap degree calculation and deviation distribution analysis between the opposing contour projection sets are performed; when the contour deviation in the corresponding direction exceeds the preset threshold, it is determined that there is a structurally inconsistent region in the current local reconstruction result; based on the structurally inconsistent region, the reverse localization of the corresponding abnormal voxel mesh is performed, and the local reconstruction process is retried.

[0015] Optionally, when all opposing contour projection sets satisfy the boundary closure consistency condition, the global verification and identification are deemed to have passed, and the updated observation confidence fusion result is converted into a structured DWS description code for output.

[0016] A second aspect of this application provides a computer-readable storage medium storing a computer program that, when executed, implements the steps of the real-time data processing method of the aforementioned dynamic DWS six-sided scanning system.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application constructs a unified time reference based on conveyor belt displacement encoder pulses, maps the raw point cloud data acquired by the DWS six-sided scanner to the same dynamic coordinate system, and establishes a mapping result. The dynamic coordinate system is discretized into a voxel grid, and observation confidence fusion is performed for each voxel grid according to the mapping result. This observation confidence fusion includes viewpoint priority fusion and occlusion confidence attenuation fusion. Conflict voxel detection is performed using the observation confidence fusion result for each voxel grid to establish voxel anomaly markers. Local reconstruction processing is performed based on the voxel anomaly markers to establish local reconstruction results. After updating the observation confidence fusion result using the local reconstruction result, global verification and identification are performed. If the global verification and identification pass, the updated observation confidence fusion result is converted into a structured DWS description code for output. This achieves the technical effect of constructing a unified dynamic spatiotemporal reference and voxel-level multi-source confidence fusion through encoder pulses, realizing accurate alignment and structural consistency reconstruction of the DWS six-sided point cloud, and improving the accuracy and reliability of DWS measurements.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 A flowchart illustrating the real-time data processing method of the dynamic DWS six-sided scanning system provided in this application.

[0021] Figure 2 A schematic diagram of the observation confidence fusion process in the real-time data processing method of the dynamic DWS six-sided scanning system provided in this application. Detailed Implementation

[0022] This application provides a real-time data processing method and medium for a dynamic DWS six-sided scanning system, addressing the technical problem of existing DWS six-sided scanning systems struggling to handle asynchronous sampling errors across multiple scanning surfaces, leading to point cloud misalignment and inconsistent structural reconstruction in dynamic transport scenarios. It achieves the technical effect of constructing a unified dynamic spatiotemporal reference through encoder pulses and voxel-level multi-source confidence fusion, enabling precise alignment of DWS six-sided point clouds and consistent structural reconstruction, thereby improving the accuracy and reliability of DWS measurements.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 , Figure 2 As shown, this application provides a real-time data processing method for a dynamic DWS six-sided scanning system, the real-time data processing method for the dynamic DWS six-sided scanning system includes: A unified time reference is constructed based on the pulses of the conveyor belt displacement encoder, and the raw point cloud data collected by the DWS six-sided scanner is mapped to the same dynamic coordinate system to establish the mapping result.

[0025] Furthermore, a unified time reference is constructed based on the conveyor belt displacement encoder pulses. The raw point cloud data collected by the DWS six-sided scanner is mapped to the same dynamic coordinate system, and the mapping result is established. This includes: real-time reading of the pulse signal of the conveyor belt displacement encoder, generating a unified timestamp sequence with the rising edge of each pulse signal as a global time synchronization reference, and establishing a nonlinear spatiotemporal mapping function with the cumulative number of encoder pulses as the independent variable; dynamically initializing the motion phase offset according to the pulse signal and photoelectric trigger signal when the package enters the measurement area of ​​the DWS six-sided scanner, and transforming each point in the raw point cloud collected by each scanning surface to the dynamic coordinate system according to the spatiotemporal mapping function; performing timestamp alignment verification on the transformed point cloud data, and outputting the aligned point cloud as the mapping result.

[0026] Specifically, a high-resolution incremental rotary encoder, or displacement encoder, is installed on the drive roller or synchronous drive shaft of the conveyor belt. This encoder outputs pulse signals corresponding to the actual displacement of the conveyor belt in real time. A high-speed interrupt acquisition module reads the pulse signals output by the conveyor belt displacement encoder in real time, defining each pulse rising edge as a global synchronous sampling event. All rising edge events are sequentially numbered to form a cumulative pulse sequence. Simultaneously, the FPGA or real-time controller synchronously records the local high-precision clock value upon detecting each rising edge, generating a unified timestamp sequence. A nonlinear spatiotemporal mapping function, with the cumulative number of encoder pulses as the independent variable, is established based on the mechanical parameters of the displacement encoder to describe the actual spatial position of the conveyor belt at any given time. For example, a mapping relationship between the cumulative pulse count and the conveying displacement is established based on the mechanical parameters of the encoder with displacement. Assuming that the encoder outputs P pulses per revolution, the radius of the active roller is R, and the reduction ratio is K, then the displacement corresponding to a single pulse is: Δs=2πR / (P×K). Based on this, the cumulative pulse count N is converted into the conveying displacement: S(N)=N×Δs. Since there are elastic deformation of the conveyor belt, acceleration and deceleration fluctuations, and slippage phenomena in the actual conveying process, a speed drift compensation term is further introduced to establish a nonlinear spatiotemporal mapping function: X(N)=S(N)+fv(N), where fv(N) is the speed fluctuation compensation function, which is derived from the statistics of adjacent pulse intervals, to realize dynamic spatiotemporal mapping with the cumulative pulse count of the encoder as the independent variable.

[0027] When a package enters the measurement area of ​​the DWS six-sided scanner, a photoelectric trigger located at the transport inlet detects the arrival of the package's leading edge. When the photoelectric signal changes from high to low, the current encoder's cumulative pulse value N0 is immediately read and used as the motion reference starting point for the package entering the scanning area. Simultaneously, based on the physical installation distance L0 between the photoelectric trigger and the DWS center scanning surface, the pulse offset theoretically required for the package's leading edge to reach the scanning center is calculated: ΔN = L0 / Δs. The difference between the current encoder's cumulative pulse value N0 and the pulse offset ΔN is used to calculate the initial motion phase offset, reflecting the dynamic correction parameter of the package's actual spatial starting point relative to the encoder's zero-position reference. Since different package lengths, entry postures, and transport gaps can all cause offsets in the scanning starting point, the initial motion phase offset is recorded as the package's motion reference zero point, ensuring that each package can establish an independent and accurate spatial reference system during dynamic transport.

[0028] A DWS six-sided scanner is used to scan the package. Each point in the original point cloud collected from each scanned surface is input into a nonlinear spatiotemporal mapping function based on the encoder's cumulative pulse value at the time of acquisition, transforming it into a unified dynamic coordinate system. This allows point clouds from different times and scanned surfaces to be mapped to the same physical package surface. Each point cloud contains: local coordinates (x, y, z) and a sampling timestamp t. i The scanning surface number k is used. After coordinate mapping is completed, the converted point cloud data is timestamped and aligned. A unified time window is established according to the encoder pulse sequence, that is, a fixed pulse span is used as the time window unit. For example, every 256 encoder pulses are divided into a synchronization window. Then, the continuity of point cloud boundaries, contour overlap, and motion displacement consistency of different scanning surfaces within the same time window are detected. Boundary continuity refers to whether the spatial connection between edge points of adjacent scanning surfaces is smooth and continuous. The existence of breaks is determined by calculating the nearest neighbor distance and normal vector change rate of edge points. Contour overlap refers to the overlap ratio of the projected contours of different scanning surfaces. The IoU is calculated after projecting the contours of each scanning surface onto the same cross section. Motion displacement consistency is used to detect whether the motion compensation results of the same physical edge are consistent under different scanning surfaces. It is determined by calculating the displacement deviation of the corresponding edge points in the conveying direction.

[0029] When a boundary deviation between adjacent scanned surfaces exceeds a threshold, such as a nearest neighbor distance exceeding 3mm, a contour overlap below 85%, or a displacement deviation exceeding 2mm, a spatial breakage anomaly is identified. The pulse mapping results for the corresponding time period are automatically retrieved, and temporal interpolation correction or local remapping is performed on the anomaly points. For example, if a scanned surface is found to have a short sampling delay within a window, normal displacement data from adjacent windows are read, and cubic spline interpolation or linear time interpolation is used to reconstruct the displacement value corresponding to the anomaly moment. The corrected displacement is then used to recalculate the dynamic coordinates of the corresponding point. After verification, the aligned point cloud is output as the mapping result, ensuring the consistency and continuity of the point cloud across multiple scanned surfaces under dynamic transport conditions.

[0030] By establishing a unified time and displacement reference through encoder pulses, dynamic spatiotemporal alignment and motion compensation of multi-scan point clouds can be achieved, eliminating time misalignment and position deviation during the transportation process, and improving the accuracy and structural consistency of six-sided scan data fusion.

[0031] The dynamic coordinate system is discretized into a voxel grid, and observation confidence fusion is performed for each voxel grid based on the mapping result. The observation confidence fusion includes viewpoint priority fusion and occlusion confidence attenuation fusion.

[0032] Furthermore, based on the mapping results, observation confidence fusion is performed for each voxel grid, including: treating each voxel grid as an evidence receiving unit, and performing observation confidence fusion using a multi-view evidence dynamic confidence field model, including: activating the viewpoint priority fusion layer; after reading the angle between the scan surface normal and the estimated normal of the corresponding voxel grid surface, establishing a first priority based on the square cosine of the angle; reading the scan distance from the scan surface to the center of the voxel grid, performing exponential decay processing on the scan distance, and establishing a second priority; outputting the viewpoint priority weight using the first priority and the second priority; activating the occlusion confidence attenuation fusion layer to perform spatial connectivity path detection between the current scan surface and the target voxel grid; if there is at least one intermediate voxel marked as occupied on the spatial connectivity path, the target voxel grid is determined to be in an occluded observation state; calculating the spatial distance attenuation factor based on the occlusion observation state, and establishing an occlusion attenuation coefficient by progressively multiplying according to the occluded voxel sequence; performing observation confidence fusion using the viewpoint priority weight and the occlusion attenuation coefficient to establish the observation confidence fusion result.

[0033] Specifically, the continuous point cloud mapping results in the dynamic coordinate system are discretized using voxelization, that is, the three-dimensional space is divided into regular voxel grids according to a fixed spatial resolution. Each voxel grid is a minimum spatial sampling unit with a fixed side length, such as 2mm×2mm×2mm or 5mm×5mm×5mm. Each voxel grid corresponds to a spatial position index (i,j,k) to carry multi-view observation information. The size of the voxel grid is preset according to requirements such as scanning accuracy, transmission speed, and target size accuracy, and is used to transform continuous point cloud data into structured spatial units. Each voxel grid is used as an evidence receiving unit, that is, to receive observation evidence from different scanning surfaces. Observation confidence fusion is performed on each voxel grid based on a multi-view evidence dynamic confidence field model, which includes a viewpoint priority fusion layer and an occlusion confidence attenuation fusion layer. Observation confidence fusion includes viewpoint priority fusion and occlusion confidence attenuation fusion.

[0034] During the observation confidence fusion process, the viewpoint priority fusion layer is activated. In the DWS six-sided scanning device, each scanning plane has completed extrinsic parameter calibration during the calibration phase, and the calibration data is stored in the device calibration database in matrix form. When performing voxel confidence fusion, the corresponding extrinsic parameter matrix is ​​directly read from the calibration database according to the current scanning plane number k, and used for spatial transformation calculation. For any voxel center point P v First, it is inversely transformed from the world coordinate system to the scan surface coordinate system to obtain the surface normal v corresponding to the observation direction. o That is, the unit direction from the optical center of the scanning plane to the center of the voxel: v o =(P v -C k ) / ||P v -C k ||, where C k The optical center position of the k-th scan plane is calculated from the translation vector in the extrinsic parameter matrix. Simultaneously, the voxel mesh surface normal n is estimated by computing the voxel neighborhood points. v The process involves taking the current voxel as the center, selecting a radius r (e.g., 10 mm) around it, and constructing a covariance matrix. Principal component analysis is then performed on this matrix, and the eigenvector corresponding to the smallest eigenvalue is taken as the estimated normal to the voxel mesh surface. Alternatively, the local plane normal vector can be obtained using least-squares plane fitting, and this vector is used as the estimated normal to represent the local geometric orientation of the voxel. Then, the angle θ between the scan surface normal and the corresponding estimated normal to the voxel mesh surface is calculated using a dot product. This angle θ characterizes the consistency between the scan incident direction and the object surface normal, thus reflecting the reliability of the observation geometry. The square of the cosine of the angle is given first priority: W1 = cosθ. 2(θ), the first priority reflects the characteristic that the more perpendicular the observation angle is to the surface, the more reliable the data. The scanning distance from the scanning surface to the center of the voxel grid, i.e., the Euclidean distance d between them, is calculated. After applying exponential decay to the scanning distance d, a second priority is established: W2 = e -λd , where λ is the distance attenuation coefficient, obtained from device resolution and point cloud density calibration. Then, the first priority and the second priority are summed and averaged to output the viewpoint priority weight.

[0035] Simultaneously, the occlusion confidence attenuation fusion layer is activated to establish a spatial connectivity path detection between the current scan plane and the target voxel mesh. This involves traversing the voxel grid point-by-point along the scan ray direction. If at least one intermediate voxel is marked as occupied in the spatial connectivity path, the target voxel mesh is determined to be in an occlusion observation state. An occupied state means that the voxel's confidence level exceeds a set threshold (e.g., 0.6) in historical fusion, and its existence as a solid structure is confirmed. Based on the occlusion observation state, an occlusion voxel sequence is constructed, and a distance attenuation factor e is applied to each level of occlusion voxel. -μdi The occlusion attenuation coefficient is formed by progressively multiplying the values ​​at each level. Here, *di* represents the distance contribution between the *i*-th occlusion voxel and the observation path, calculated based on the difference between the perpendicular distance from the center point of the occlusion voxel to the ideal ray path or the projected distance along the ray. *μ* is the occlusion attenuation sensitivity coefficient, used to control the sensitivity of occlusion to the attenuation of observation confidence. The value of *μ* is limited to 0.1~1 according to the calibration results, used to cover observation scenarios with different attenuation intensities from weak to strong occlusion. Then, observation confidence fusion is performed based on the viewpoint priority weight and the occlusion attenuation coefficient, weighting and accumulating observations from different scanning planes to form the observation confidence fusion result. The weighting coefficients can be set based on actual needs and historical experience.

[0036] By discretizing the continuous point cloud mapping results into voxel evidence receiving units and introducing a viewpoint geometric reliability and occlusion propagation attenuation mechanism, a reliable quantitative fusion of observation information from multiple scanning surfaces is achieved. This enables the stable construction of the spatial structure of real objects under complex occlusion and multi-view conflict conditions, thereby improving the accuracy and reliability of voxel-level reconstruction.

[0037] Furthermore, establishing the observation confidence fusion result also includes: acquiring historical confidence state data, performing attenuation retention on the historical confidence state data according to a recursive update mechanism to establish a confidence update term; and using the confidence update term to perform confidence correction processing on the observation confidence fusion result to update the observation confidence fusion result.

[0038] Specifically, historical confidence status data is obtained through voxel-level temporal storage and index backtracking. After each frame of observation confidence fusion is completed, the corresponding observation confidence fusion result, the source scan plane number, and the time window identifier are written to a continuous temporal cache structure, such as a circular buffer or a key-value database, using the voxel grid index (i,j,k) as the key. When historical confidence status data is needed, the record within the previous time window or a specified backtracking depth Δt is directly accessed based on the current voxel index. The location is determined by timestamp matching, and the corresponding observation confidence fusion result, i.e., the confidence status data, is read from the cache.

[0039] A time decay retention strategy is introduced to exponentially decay historical confidence state data, in order to avoid old data having an excessive impact on the current dynamic scenario. A confidence update term Cnew=C is established. t-1 ×e -αΔt , where C t-1 The historical confidence level data is used, where Δt is the difference between the current time window number t and the historical time window t-1, and α is the time decay coefficient, calibrated by the conveyor belt speed fluctuation rate and the system frame rate. This coefficient controls the strength of historical information retention; a smaller value (e.g., 0.1–0.3) is used in high-speed, stable conveyor scenarios to enhance historical stability, while a larger value (e.g., 0.5–1.0) is used in scenarios with large speed fluctuations or strong dynamic disturbances to reduce historical dependence. This allows the historical confidence level to adaptively reflect the motion stability and observation reliability changes of the DWS six-sided scanning system. The confidence update terms for all observation sources are normalized to obtain corrected weights. These corrected weights are then used to perform weighted fusion of the observation confidence fusion results for each observation source, achieving confidence correction processing. Confidence correction processing uses the time-decayed historical confidence information to dynamically adjust the contribution ratio of each observation source in the current fusion result. This automatically reduces the weight of observation sources with poor long-term performance and increases the weight of observation sources with stable recent performance, thus making the fusion result more closely reflect the actual reliability status of each observation source.

[0040] By introducing a historical confidence decay mechanism, the system no longer relies on single-frame observation results but integrates stability information in the time dimension. This effectively suppresses confidence fluctuations caused by instantaneous noise, false detections, and local occlusion, improves the temporal consistency and structural stability of voxel occupancy judgment, and further enhances the reliability and accuracy of data processing in the dynamic DWS six-sided scanning system.

[0041] Conflicting voxel detection is performed using the observation confidence fusion results of each voxel grid to establish voxel anomaly markers.

[0042] Furthermore, conflict voxel detection is performed using the observation confidence fusion results of each voxel grid to establish voxel anomaly identifiers. This includes: performing segmented judgment on each voxel grid in the observation confidence fusion results; if the confidence change of the same voxel grid in adjacent time windows exceeds the dynamic fluctuation threshold, triggering the first type of anomaly detection mechanism and marking the corresponding voxel grid as a confidence unstable voxel; performing consistency comparison of observation results of the same voxel grid under different scan planes; when the occupancy or idle state of the same voxel grid under different scan planes shows opposite judgments, triggering the second type of conflict detection mechanism and marking the corresponding voxel grid as a multi-source observation conflict voxel; performing consistency scanning of the confidence distribution of adjacent voxel grids based on the spatial connectivity structure of the local neighborhood where the voxel grid is located; when the target voxel grid and the neighboring voxel grid show topological breakage features in occupancy continuity, triggering the third type of structural anomaly detection mechanism and marking the target voxel grid as a spatial connectivity anomalous voxel; and establishing voxel anomaly identifiers based on confidence unstable voxels, multi-source observation conflict voxels, and spatial connectivity anomalous voxels.

[0043] Specifically, after obtaining the observation confidence fusion result, each voxel grid is segmented and judged within the result: the confidence level of the same voxel in adjacent time windows is extracted, and the difference is calculated to determine the confidence change amplitude. The confidence level represents the probability of the voxel being occupied, as indicated by the observation confidence fusion result for each voxel grid. The confidence change amplitude is compared with a dynamic fluctuation threshold. When the confidence change amplitude of the same voxel grid in adjacent time windows exceeds the dynamic fluctuation threshold, the first type of anomaly detection mechanism is triggered, and the corresponding voxel grid is marked as a confidence unstable voxel. The dynamic fluctuation threshold is statistically obtained based on the conveyor belt speed fluctuation rate and the point cloud inter-frame error, and is typically set to 0.15–0.25. This threshold is used to filter the difference between normal observation fluctuations and anomalous sudden changes. In areas where the confidence level itself fluctuates significantly, such as target edges or noisy areas, a larger threshold is set to avoid misjudging normal jitter as anomalies. In areas with stable confidence levels, such as purely occupied or purely idle areas, a smaller threshold is set to improve anomaly detection sensitivity.

[0044] Consistency comparisons are performed on observations of the same voxel grid across different scan surfaces, such as the top, left and right sides, and front and back. First, the voxel grid confidence score for each scan surface is binarized. If the confidence score is greater than 0.6, the voxel grid on that scan surface is considered occupied; if the confidence score is less than 0.4, it is considered idle; and if it is between 0.6 and 0.4, it is considered uncertain and not included in the conflict statistics. The number of clearly occupied and clearly idle voxels across all scan surfaces is counted. When the number of occupied and idle voxels on the same voxel grid is greater than 0 across different scan surfaces, it indicates that the same voxel has been given opposite occupancy or idle status assessments across different scan surfaces, triggering the second type of conflict detection mechanism, and the corresponding voxel grid is marked as a multi-source observation conflict voxel.

[0045] Based on voxel neighborhood, a spatial connectivity structure of local neighborhoods is constructed. A consistency scan is performed on the confidence distribution of the neighborhood voxel mesh set surrounding the target voxel. The spatial connectivity structure refers to whether continuous occupied paths are formed between voxels, used to describe the topological integrity of object surfaces or solid structures. First, a local neighborhood of each voxel is predefined, such as a 3×3×3 cube neighborhood centered on each voxel, with a total of 26 neighboring voxels. The confidence values ​​of all voxels within the neighborhood are extracted under the current time window. Connectivity analysis is then performed on the target voxel itself and its neighborhood. Occupation continuity means that in three-dimensional space, if a voxel is determined to be occupied, at least one of its neighboring voxels should also be determined to be occupied, thus forming a continuous occupied region.

[0046] Graph theory connectivity is used for determination: The system checks whether there is a connected path between the occupied state of a target voxel and its neighboring occupied voxel set. If the target voxel's confidence score is greater than 0.6, it is determined to be occupied, but if the number of occupied voxels in its 26 neighbors is 0, it indicates that the target voxel is an isolated occupied point. Alternatively, if the target voxel is determined to be idle, but the number of occupied voxels in its 26 neighbors is greater than or equal to 3, it indicates that there is a clear occupied cluster around it, but the central voxel is determined to be idle, forming a void. This indicates a topological break in the occupancy continuity between the target voxel and its neighbors, triggering the third type of structural anomaly detection mechanism. The target voxel mesh is then marked as a spatially connected anomaly voxel, and the break type is recorded, such as isolated occupation or a void surrounded by occupied voxels. Furthermore, voxels with unstable confidence, multi-source conflict voxels, and spatially connected anomaly voxels are summarized to form voxel anomaly labels, corresponding to unstable confidence, multi-source conflict, and spatial topological anomalies, respectively.

[0047] By employing triple constraints of temporal consistency, cross-view consistency, and spatial topological consistency, conflict identification is performed on voxel-level observation results, thereby accurately locating abnormal regions generated during the fusion process, avoiding the spread of erroneous confidence, and improving the spatial consistency and physical rationality of the output results of the DWS six-sided scanning system.

[0048] Perform local reconstruction processing based on the voxel anomaly identifiers to establish local reconstruction results.

[0049] Furthermore, local reconstruction processing is performed based on the voxel anomaly identifiers to establish local reconstruction results, including: performing anomaly reverse tracing on the voxel mesh based on the voxel anomaly identifiers to establish anomaly causal chains; extracting all voxel meshes stably marked as occupied within the spatially connected domains of the anomalous voxel meshes, and constructing boundary constraints based on the extraction results; using local implicit surface fitting to construct a minimum energy manifold passing through the surfaces of all boundary voxels based on the boundary constraints, wherein the minimum energy manifold represents the predicted real physical surface wrapped in the corresponding local region; performing expected occupancy state analysis of the anomalous voxel meshes using the minimum energy manifold, and encoding the analysis results as virtual observation evidence; injecting the virtual observation evidence back along the anomaly causal chain into the corresponding historical observation records of each relevant scan surface to perform forced correction; and establishing local reconstruction results based on the forced correction results.

[0050] Specifically, the target anomalous voxel is located based on the voxel anomaly identifier, and the complete observation record of the voxel in the historical time-series cache is read, including the confidence change value under each time window, the scan surface number participating in the fusion, the corresponding observation timestamp, the viewpoint priority weight, and the occlusion attenuation coefficient. Starting from the current anomalous voxel identifier, anomaly reverse tracing analysis is performed on the target voxels marked as confidence unstable voxels, multi-source observation conflict voxels, and spatial connectivity anomalous voxels. The analysis is performed by tracing back along the historical fusion path in reverse chronological order, analyzing the contribution ratio of each observation source to the final state of the voxel in each confidence update. The contribution record includes the scan surface number, the original observation confidence of the scan surface, the viewpoint priority weight, the occlusion attenuation coefficient, and the normalized contribution ratio when finally participating in the fusion.

[0051] The contribution percentage of each observation source to the final confidence result is calculated in reverse using the fusion formula. For example, the contribution value of a certain scan surface can be calculated as follows: , where R k W represents the proportion of the k-th scan plane's contribution to the final fusion result of the current target voxel, i.e., the relative influence of this scan plane in the current voxel confidence update. k C represents the comprehensive fusion weight of the k-th scan surface. This weight is calculated by considering factors such as viewpoint priority weight and occlusion attenuation coefficient, and is used to characterize the reliability of the current observation of this scan surface. k This represents the observation confidence of the k-th scan face towards the current target voxel, i.e., the degree of confidence that the scan face determines that the voxel is in an occupied or idle state. N represents the total number of scan faces participating in the fusion calculation of the current target voxel. For example, in a six-faceted scanning system, N can be up to 6. n C represents the overall fusion weight of the nth scan surface. nThis represents the observation confidence level of the nth scan surface.

[0052] Continuing back to the previous time window, the contribution trend of the same observation source in consecutive time windows is statistically analyzed. If an observation source continuously introduces high conflict weights, abnormal fluctuations, or inconsistencies with the neighborhood topology in multiple consecutive time windows, the corresponding scan plane, time window, and associated voxel node of the observation source are recorded as an abnormal propagation node. For example, if the top scan plane continuously classifies the same voxel as occupied for 3 consecutive frames, while the side scan plane remains idle, and this observation causes a local structural break, then the top scan plane, the corresponding time window, and the associated voxel node are jointly recorded as an abnormal propagation node and written into the abnormal propagation table.

[0053] Using voxel observation events as graph nodes, each node represents a specific observation update event for a voxel within a certain time window. The node content includes the voxel index, time window number, observation source number, and updated confidence state. Dependency relationships between nodes are established based on historical fusion records. That is, if the current voxel state is obtained by fusing multiple observations from the previous time step, directed edges are established from these historical observation nodes to the current node, constructing a directed dependency graph. Through graph traversal methods, such as Depth-First Search (DFS) or Breadth-First Search (BFS), the entire directed dependency graph is traversed backward from the current anomalous node, gradually finding key contributing nodes on the anomaly propagation path, forming a chain structure from the original scan observation to the intermediate fusion state to the current anomalous voxel—the anomaly causal chain.

[0054] After constructing the causal chain, stable voxel mesh extraction is performed on the spatial connected domains containing anomalous voxel meshes. This involves using anomalous voxel meshes as seed nodes and employing a 3D connected domain search, such as 26-neighbor expansion, to obtain their local structural regions. From these local structural regions, voxel meshes with a confidence level consistently above a stability threshold and no anomalous markers (e.g., a confidence level above 0.6 and no anomalous markers) are selected and defined as stable, occupied voxel meshes, representing reliable real structural support points. Boundary constraints are then constructed based on the extracted stable, occupied voxel meshes to limit the spatial range that subsequent surface fitting must traverse or approach.

[0055] A local implicit surface fitting method is introduced to reconstruct boundary constraints. The local implicit surface refers to a surface whose positional relationships are represented by an implicit function F(x,y,z)=0, rather than explicitly representing mesh connectivity. Using a stable voxel set as input, the optimal implicit function satisfying the boundary constraints is solved through least-squares fitting. This optimal implicit function minimizes its error at all boundary voxel points, thus constructing a minimum energy manifold. The minimum energy manifold represents the continuous geometric structure that minimizes the overall bending energy of the surface, such as the curvature integral or gradient change, under the boundary constraints. It characterizes the predicted real physical surface enveloping the corresponding local region, essentially providing an optimal estimate of the real physical surface, used to fill in structural gaps caused by occlusion or collisions.

[0056] Using a minimum energy manifold, the expected occupancy state of the anomalous voxel mesh is analyzed. By determining the sign value F(x,y,z) of the anomalous voxel in the implicit surface function, if F<0, it is determined to be internally occupied; if F>0, it is determined to be in an idle state, thus obtaining the predicted true state of the anomalous region. This predicted true state is encoded as virtual observation evidence, whose data structure is consistent with the real scan observation, including voxel index, occupancy state, and confidence weight. The virtual observation evidence is injected in reverse along the aforementioned anomalous causal chain, i.e., written back to the historical observation records corresponding to each relevant scan surface, forcibly correcting the original observation data. By locating the scan surface data entry corresponding to the anomalous causal chain in the historical cache, the virtual observation evidence is inserted as a new observation item, and the original anomalous observation value is overwritten or corrected with a high confidence weight, so that the subsequent confidence fusion process is recalculated based on the corrected observation sequence. The corrected voxel set is then locally fused and updated to generate a local reconstruction result, which achieves structural completion and topology restoration in the anomalous region.

[0057] For example, if three anomalous voxels appear in a local area, connected component analysis determines that the region contains 120 voxels, of which 90 are stable high-confidence voxels with a confidence level > 0.6 and no other anomaly markers. Based on this, a boundary constraint point set is constructed. After obtaining the implicit function through Poisson reconstruction, two points among the anomalous voxels satisfy F(x,y,z) < 0 and are determined to be actually occupied areas, while one point with F > 0 is corrected to be empty. This analysis result is used to generate virtual observation evidence and written back to the corresponding top and side scan history records, unifying the previously conflicting occupation determinations into a consistent structure, thereby completing the local reconstruction and repair of the region.

[0058] Through a closed-loop mechanism of anomaly tracing, stable boundary extraction, implicit surface reconstruction, and virtual observation back-injection, the incomplete voxel structure originally caused by occlusion, multi-source conflict, or sampling loss is physically and consistently reconstructed, thereby restoring the true geometric shape in the local area and simultaneously correcting historical observation data in reverse. This not only corrects anomalies at the current moment but also eliminates the cumulative impact of anomalies on the fusion of subsequent moments, further improving the long-term stability and reconstruction accuracy of the dynamic DWS six-sided scanning system's multi-source observation fusion in complex scenarios such as occlusion, multipath, and sensor failure.

[0059] After updating the observation confidence fusion result using the local reconstruction result, a global verification and identification is performed. If the global verification and identification pass, the updated observation confidence fusion result is converted into a structured DWS description code for output.

[0060] Furthermore, a global verification and identification process is performed, including: extracting the contour projection boundaries of the updated observation confidence fusion results in the relative scanning direction and establishing a set of opposing contour projections; performing boundary overlap calculation and deviation distribution analysis between the opposing contour projection sets; when the contour deviation in the corresponding direction exceeds a preset threshold, determining that there is a structurally inconsistent region in the current local reconstruction result; performing reverse localization of the corresponding abnormal voxel mesh based on the structurally inconsistent region, and re-triggering the local reconstruction process.

[0061] Specifically, after completing local reconstruction and updating the observation confidence fusion results, a global verification and identification process is further performed to verify the geometric consistency and closure integrity of the entire enclosed 3D structure in multiple directions. First, a set of stable occupying voxels is extracted from the updated observation confidence fusion results; voxels with a confidence level higher than a set threshold, such as 0.6, are selected as stable occupying voxels. Contour projection boundaries are then generated based on these stable occupying voxels along their relative scanning directions. The relative scanning directions refer to the scanning directions relative to each other in the DWS six-sided scanning system, such as top-bottom, left-right, and front-rear directions. The contour projection boundary refers to the outermost boundary curve formed after projecting the stable occupying voxels along a specified direction onto a two-dimensional plane, representing the outer contour shape of the object in that direction. An orthogonal projection method is used to project the set of stable occupying voxels onto the corresponding two-dimensional projection planes. For example, for the left-right relative scanning directions, voxels are projected along the X-axis to generate a two-dimensional contour projection image on the YZ plane; for the front-rear direction, projection is onto the XZ plane.

[0062] Boundary extraction algorithms, such as convex hull extraction or edge tracking algorithms, are used to extract the set of boundary points of the projected contour, and a set of opposing contour projections is established: P pair ={P a ,P b}, where P a With P bThese represent two projected contours in the same relative direction. After establishing the opposing contour projection set, boundary overlap calculation and deviation distribution analysis are performed on the opposing contour projection set. Boundary overlap is used to measure the consistency of opposing contours in spatial structure. The coordinates of the two opposing contours are normalized and aligned with the center through rigid body transformation, and then their intersection-over-union ratio (IoU) is calculated. When the IoU value is low, it indicates that there is inconsistency in the structure in the corresponding direction after local reconstruction. At the same time, the spatial deviation distribution between contour boundary points is calculated, that is, the nearest neighbor distance between corresponding boundary points is statistically analyzed, and the overall deviation mean is calculated as the contour deviation. When the contour deviation in a certain relative direction exceeds a preset threshold, it is determined that there is a structurally inconsistent region in the current local reconstruction result. The deviation threshold is set according to the DWS dimensional accuracy requirements and device resolution. For example, for a system with a dimensional error requirement of ±5mm, the local contour deviation threshold is set to 2~3mm, and the IoU threshold is set to below 0.85 as structural inconsistency.

[0063] If inconsistencies are detected in the current local reconstruction results, the location of the deviation is back-mapped to the corresponding 3D voxel region. Using the spatial mapping between projected coordinates and voxel indices, the corresponding anomalous voxel mesh set is located, and the local reconstruction process is re-triggered. This means the anomalous region is then used again as the input region for local implicit surface fitting and virtual observation correction. Boundary constraint extraction, minimum energy manifold reconstruction, and virtual observation back-injection are re-executed, thus achieving secondary repair under global closed-loop verification. For example, after projecting the left and right contours, two 2D contour sets are obtained with an IoU value of 0.78, lower than the set threshold of 0.85. Simultaneously, the maximum deviation of the corresponding boundary points reaches 4.2mm, exceeding the 3mm tolerance error. Back-locating the corresponding voxel region using projected coordinates reveals 35 anomalous voxels. Local implicit surface reconstruction is then re-executed in this region, and after back-injecting virtual observation evidence, verification is performed again. Ultimately, the IoU improves to 0.91, the maximum boundary deviation decreases to 1.4mm, and the global verification is deemed successful.

[0064] By verifying the consistency of multi-directional contour projection, the overall geometric structure after local reconstruction is verified globally. This avoids the problem that although local repair is valid in a local area, contour misalignment, boundary non-closure or structural distortion may occur in the overall spatial structure. This further ensures the integrity and geometric consistency of the final 3D reconstruction result at the global scale and improves the accuracy of DWS package size measurement and the reliability of structure recognition.

[0065] Furthermore, when all opposing contour projection sets satisfy the boundary closure consistency condition, the global verification and identification are deemed successful, and the updated observation confidence fusion result is converted into a structured DWS description code for output.

[0066] Furthermore, the structured DWS description code includes outer contour topology encoding, internal cavity structure marking, and size constraint vectorization representation.

[0067] Specifically, when all opposing contour projection sets satisfy the boundary closure consistency condition, that is, when the contour deviation and IoU of all opposing contour projection sets are within the deviation threshold range, it indicates that the boundary closure consistency condition is met, the global verification and recognition are deemed successful, and the updated observation confidence fusion result is converted into a structured DWS description code for output. The structured DWS description code includes an outer contour topology code, an internal void structure marker, and a vectorized representation of size constraints. The outer contour topology code is generated by reconstructing the three-dimensional surface of the occupying voxel, such as by using the marching cubes algorithm to linearly interpolate the 8 vertices of each voxel to determine the intersection of the isosurface and the 12 edges of the voxel. Then, the intersections of adjacent voxels are connected into triangular patches according to 256 predefined topological configurations to generate the triangular mesh surface of the occupied area. The connectivity of the triangular mesh is then encoded using a graph structure, where nodes represent mesh vertices and edges represent topological connections, thus forming a topological graph code that can be used for machine parsing. The topological graph code not only records the geometric shape but also the connectivity and boundary structure between the surfaces, which is used to describe the overall shape contour of the enclosure.

[0068] Internal cavity structure labeling refers to detecting unoccupied but externally surrounded closed cavity regions through connected domain inversion analysis of voxel space. This involves using 3D flooding or ray projection methods to determine closed cavity structures. For 3D flooding, all external voxels are selected as seed nodes from the external boundary of the voxel space (i.e., the six faces of the enclosing cube). A 3D BFS or DFS is used to traverse all externally connected unoccupied voxels, marking them as reachable regions. Unoccupied voxel regions not accessible from external connections are defined as candidate closed cavity regions; these regions, although unoccupied, are completely surrounded by solid voxels and cannot be entered from the outside, thus forming an internal cavity structure. For ray projection, multi-directional rays are emitted along the x, y, and z principal axes, penetrating from the outside inwards. The number of times rays enter and leave occupied voxels is counted. If a region exhibits a completely enclosed structure in all multi-directional ray projections (i.e., rays cannot re-enter the external boundary after entering), then the region is determined to be a closed cavity. The set of voxels in the closed region obtained by any method is uniformly labeled as the set of void voxels, and its voxel index, spatial connectivity range and volume statistics are recorded to characterize the non-physical spatial structure inside the package.

[0069] The vectorized representation of size constraints is calculated based on the extreme value projection of the occupied voxel boundary along the three principal axes (x, y, z). First, the coordinates of the center points (x, y, z) of all voxels are extracted from the final set of occupied voxels. i,y i ,z i The extreme value statistical projection process is performed along the three principal axes of x, y, and z, that is, a global scan comparison is performed on the values ​​of all voxel points on the corresponding coordinate axes. In the x-axis direction, the minimum and maximum coordinate values ​​are calculated through point-by-point comparison: x min =min(x i ), x max =max(x i This is used to describe the spatial boundary range of the wrapping along the length direction. Similarly, the same statistical operations are performed along the y-axis and z-axis to describe the spatial boundary range of the wrapping along the width and height directions.

[0070] Then, the differences are calculated to obtain the geometric dimensions L, W, and H of the package in three directions, where L represents the length dimension, W represents the width dimension, and H represents the height dimension. Based on this, the squares of the three dimensions are summed and then the square root is taken using the Euclidean distance formula to represent the maximum diagonal span of the package in three-dimensional space, which is used as the overall envelope scale D. The dimensional information is organized into a vectorized expression: D=[L,W,H,D]. The three types of information—outer contour topology encoding, internal cavity structure marking, and dimensional constraint vectorized expression—are structured and encapsulated to generate a structured DWS description code. This description code not only contains geometric dimension information but also topological structure and internal spatial distribution information, which can be directly used for logistics sorting, volumetric billing, warehouse modeling, and automated grasping decisions.

[0071] By employing multi-round observation confidence fusion, conflict voxel detection, local geometric reconstruction, and global consistency verification, stable six-sided fusion is achieved in high-speed dynamic transportation scenarios, improving the accuracy and reliability of DWS measurements in such environments. The updated fused observation confidence fusion results are then structurally encoded and converted into a standardized DWS description, elevating the original discrete point cloud data into a unified semantic description encompassing outer contour topology, internal cavity structure, and dimensional constraints. This enables the mapping from low-level geometric perception data to high-level logistics semantic information, thereby improving the structural integrity of package identification, the accuracy of dimensional measurement, and the reliability of automated sorting decisions.

[0072] Example 2: Based on the same inventive concept as the real-time data processing method of the dynamic DWS six-sided scanning system in the foregoing embodiments, this application also provides a computer-readable storage medium storing a computer program, which, when executed, implements the steps of the real-time data processing method of the dynamic DWS six-sided scanning system described in any one of the above embodiments.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0074] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A real-time data processing method for a dynamic DWS six-sided scanning system, characterized in that, The method includes: A unified time reference is constructed based on the pulses of the conveyor belt displacement encoder. The raw point cloud data collected by the DWS six-sided scanner is mapped to the same dynamic coordinate system to establish the mapping result. The dynamic coordinate system is discretized into a voxel grid, and observation confidence fusion is performed for each voxel grid based on the mapping result. The observation confidence fusion includes viewpoint priority fusion and occlusion confidence attenuation fusion. Conflicting voxel detection is performed using the observation confidence fusion results of each voxel grid to establish voxel anomaly markers; Perform local reconstruction processing based on the voxel anomaly markers to establish local reconstruction results; After updating the observation confidence fusion result using the local reconstruction result, a global verification and identification is performed. If the global verification and identification pass, the updated observation confidence fusion result is converted into a structured DWS description code for output.

2. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 1, characterized in that, Based on the mapping results, perform observation confidence fusion for each voxel grid, including: Each voxel grid is treated as an evidence receiving unit, and observation confidence fusion is performed using a multi-view evidence dynamic confidence field model, including: Activate the view priority fusion layer. After reading the angle between the scan surface normal and the estimated normal of the corresponding voxel mesh surface, establish the first priority based on the square cosine value of the angle. The scanning distance from the scanning surface to the center of the voxel grid is read. After the scanning distance is subjected to exponential decay processing, a second priority is established. The first priority and the second priority are used to output the viewpoint priority weight. Activate the occlusion confidence attenuation fusion layer to perform spatial connectivity path detection between the current scan plane and the target voxel mesh. If there is at least one intermediate voxel marked as occupied on the spatial connectivity path, the target voxel mesh is determined to be in an occluded observation state. The spatial distance attenuation factor is calculated based on the occlusion observation status, and the occlusion attenuation coefficient is established by progressively multiplying the occlusion voxel sequence. Observation confidence fusion is performed using viewpoint priority weights and occlusion attenuation coefficients to establish the observation confidence fusion result.

3. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 2, characterized in that, Establishing observation confidence fusion results also includes: Obtain historical confidence status data, perform attenuation retention on the historical confidence status data according to the recursive update mechanism, and establish a confidence update item; The confidence correction process of the observation confidence fusion result is performed using the confidence update term to update the observation confidence fusion result.

4. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 3, characterized in that, Conflicting voxel detection is performed using the observation confidence fusion results for each voxel grid, and voxel anomaly markers are established, including: Perform segmentation judgment for each voxel grid in the observation confidence fusion results: If the confidence level of the same voxel grid changes more than the dynamic fluctuation threshold in adjacent time windows, the first type of anomaly detection mechanism is triggered, and the corresponding voxel grid is marked as a confidence unstable voxel. Consistency comparison is performed on the observation results of the same voxel grid under different scanning planes. When the occupied or idle states of the same voxel grid under different scanning planes are opposite, the second type of conflict detection mechanism is triggered, and the corresponding voxel grid is marked as a multi-source observation conflict voxel. Based on the spatial connectivity structure of the local neighborhood of the voxel grid, a consistency scan is performed on the confidence distribution of the adjacent voxel grids. When the target voxel grid and the neighboring voxel grids show topological breakage characteristics in terms of occupancy continuity, the third type of structural anomaly detection mechanism is triggered, and the target voxel grid is marked as a spatial connectivity anomalous voxel. Voxel anomaly markers are established based on confidence-unstable voxels, multi-source observation conflict voxels, and spatial connectivity anomaly voxels.

5. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 1, characterized in that, Local reconstruction processing is performed based on the voxel anomaly markers to establish local reconstruction results, including: Based on the voxel anomaly identifier, anomaly reverse tracing is performed on the voxel mesh to establish an anomaly causal chain. Extract all stable voxel grids marked as occupied within the spatially connected domain of the anomalous voxel grid, and construct boundary constraints based on the extraction results; Local implicit surface fitting is used to construct a minimum energy manifold that passes through all boundary voxel surfaces based on boundary constraints. The minimum energy manifold represents the predicted real physical surface wrapped in the corresponding local region. The expected occupancy state analysis of the anomalous voxel mesh is performed using the minimum energy manifold, and the analysis results are encoded as virtual observation evidence. The virtual observation evidence is injected backward along the abnormal causal chain into the historical observation record corresponding to each relevant scan surface to perform forced correction. The local reconstruction results are established based on the forced correction results.

6. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 1, characterized in that, The structured DWS descriptor code includes outer contour topology encoding, internal cavity structure marking, and size constraint vectorization.

7. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 1, characterized in that, A unified time reference is constructed based on the pulses of the conveyor belt displacement encoder. The raw point cloud data acquired by the DWS six-sided scanner is mapped to the same dynamic coordinate system, and the mapping results are established, including: The pulse signal of the conveyor belt displacement encoder is read in real time. A unified timestamp sequence is generated using the rising edge of each pulse signal as a global time synchronization reference. A nonlinear spatiotemporal mapping function with the cumulative number of encoder pulses as the independent variable is established. When the package enters the measurement area of ​​the DWS six-sided scanner, the motion phase offset is dynamically initialized according to the pulse signal and photoelectric trigger signal, and each point in the original point cloud collected by each scanning surface is transformed into the dynamic coordinate system according to the spatiotemporal mapping function. The converted point cloud data is timestamped for alignment and the aligned point cloud is output as the mapping result.

8. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 1, characterized in that, Perform global verification and identification, including: Extract the contour projection boundaries of the updated observation confidence fusion results in the relative scanning direction, and establish a set of opposing contour projections; Perform boundary overlap calculation and deviation distribution analysis between opposing contour projection sets. When the contour deviation in the corresponding direction exceeds the preset threshold, it is determined that there is a structurally inconsistent region in the current local reconstruction result. Based on the structurally inconsistent regions, reverse localization of the corresponding anomalous voxel mesh is performed, and the local reconstruction process is retried.

9. The real-time data processing method of the dynamic DWS six-sided scanning system as described in claim 8, characterized in that, When all opposing contour projection sets satisfy the boundary closure consistency condition, the global verification and identification are deemed successful, and the updated observation confidence fusion result is converted into a structured DWS description code for output.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the real-time data processing method of the dynamic DWS six-sided scanning system according to any one of claims 1 to 9.

Images