VR large space mapping precision evaluation method

Abstract

The invention discloses a VR large space mapping precision evaluation method, and particularly relates to the field of VR large space, and the method comprises the steps: collecting the original data of a sensor in real time through a front end, executing feature extraction and time synchronization, and generating initial pose estimation, observation factors and key frames; the front end stores the packaged key frame data and the sensor meta-information into an annular buffer area and transmits the key frame data and the sensor meta-information to the rear end according to a timestamp; after the back end receives key frame packed data, pose, speed, bias and environment feature state nodes are constructed for each key frame in the factor graph; by detecting instantaneous jump of bright drifting in a short period and compensating sensor bias of dark drifting in a long period, the defect that short-time mutation and long-term error accumulation cannot be both considered in traditional SLAM is overcome, and synchronous suppression of short-time positioning abnormity and long-term map distortion is realized, so that the precision and robustness of a VR large space are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of VR large space, and more specifically, to a method for evaluating the spatial mapping accuracy of VR large space. Background Art

[0002] VR large space spatial mapping refers to constructing a high-precision, continuous and stable spatial coordinate system through multi-sensor data fusion and real-time calculation in a large-scale virtual environment to support accurate positioning and interaction of users in the virtual environment. This mapping usually relies on SLAM technology, combines sensors such as vision, IMU, and lidar to obtain environmental features, and constructs a globally consistent three-dimensional space model through optimization algorithms to ensure that the VR system can maintain high-precision tracking and interaction stability within a large range; In the evaluation of VR large space spatial mapping accuracy, pose estimation and environmental reconstruction are usually based on SLAM. However, traditional SLAM lacks a comprehensive drift control mechanism for short-term mutations and long-term cumulative errors, that is, it has deficiencies in the instantaneous jump detection of "visible drift" and the long-term compensation of sensor biases of "invisible drift". Therefore, in practical applications, it is easy to cause short-term positioning anomalies, long-term map distortions, and unstable global coordinate systems, affecting the accuracy and robustness of the system. Summary of the Invention

[0003] To overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for evaluating the spatial mapping accuracy of VR large space, which overcomes the deficiency that traditional SLAM cannot take into account short-term mutations and long-term error accumulation by detecting the instantaneous jump of "visible drift" in the short term and compensating the sensor bias of "invisible drift" in the long term, so as to solve the problems raised in the above background art.

[0004] To achieve the above object, the present invention provides the following technical solution: A method for evaluating the spatial mapping accuracy of VR large space, including: Real-time collect sensor raw data through the front end, perform feature extraction and time synchronization to generate preliminary pose estimation, observation factors and key frames; The front end stores the packed key frame data and sensor meta information in a circular buffer and transmits them to the back end according to the time stamp; After receiving the key frame packed data, the back end constructs pose, velocity, bias and environmental feature state nodes for each key frame in the factor graph; Monitor the observation residuals through the back end, trigger the generation of multiple hypothesis branches, independently construct branches and then compare the residuals, select the target branch and discard the data of other branches to correct the "visible drift" problem of instantaneous positioning jumps; The back end generates snapshots containing the factor graph and observation information at fixed key frame intervals for subsequent fallback mechanism calls; When the backend detects that the residual value exceeds the preset residual threshold and the pose state has not converged, the system will roll back to the pre-stored snapshot and load the corresponding observation data from the circular buffer; The backend constructs a time- and temperature-dependent bias function node for the sensor in the factor graph to correct the sensor parameters and suppress the "dark drift" problem of long-term drift; During the non-linear optimization iteration process of the backend, the sensor bias function node is updated in real time according to the temperature- and time-related observation factors, and the global state estimation is adjusted synchronously.

[0005] In a preferred embodiment, the front end obtains raw data from the sensor and uses feature extraction and time synchronization to generate a preliminary pose estimate and observation factors; the raw data includes camera images, IMU readings, and laser point clouds; The front end selects key frames based on the motion amplitude and / or feature distribution in the image sequence, and packs the key frames together with the corresponding preliminary pose estimates and observation factors; The front end stores the packed key frame data in a circular buffer and retains the timestamp index for subsequent backtracking or branching operations; The front end sends the key frame packed data and sensor meta information to the backend together as the input source for the backend global optimization.

[0006] In a preferred embodiment, the backend monitors the residual of the key frame in the observation factor. If the residual value exceeds the preset residual threshold, multiple hypothesis branch generation is triggered; The backend shares the previous key frame node at the divergence for multiple hypothesis branches and independently creates new key frames and constraint factors after the divergence; The backend compares the overall residuals of each branch during the parallel or polling graph optimization process to determine the target branch solution; After selecting the target branch, the backend marks the remaining inferior branches as discarded and reclaims the storage of related nodes and factors.

[0007] In a preferred embodiment, the backend establishes a health assessment model for each feature factor to quantify its observation quality, and gradually reduces its weight and / or marks it for elimination when the cumulative observation failure occurs; The backend regularly scans the feature factors determined to be inefficient by the health assessment model in the background thread, batch clears them, and performs a local re-optimization process after the clearance; The backend configures anchor nodes in the factor graph and regularly applies global constraints based on the absolute coordinate measurement factors; After receiving the factor graph update or branch merge signal from the backend, the front end refreshes the index of the key frames in the circular buffer and their corresponding observation factor references.

[0008] In a preferred embodiment, a comprehensive cost function for multiple hypothesis branches is established based on Equation (1); by comprehensively scoring the multiple hypothesis branches to identify and correct the positioning mutations caused by "drift", when the backend detects an instantaneous jump in the system, the overall cost of each branch is calculated based on Equation (1) to compare the performance of each branch in terms of observation residuals, instantaneous jump amplitude, feature quality, and branch duration; Equation (1) is: ; where is the comprehensive cost of the branch; is the branch set of key frame indices included; is the key frame observation residual vector; is the key frame residual difference between it and its adjacent key frame; is the key frame time weight coefficient; is the residual surge factor; is the key frame corresponding feature health value; refers to the health decay function; represents the time span of the branch duration; is the branch duration penalty term.

[0009] In a preferred embodiment, a feature health and observation failure function is constructed based on Equation (2); Equation (2) provides health quantification for the feature reflecting the adaptability of the feature under different perspectives and appearances. The system filters feature factors through the value of Equation (2) and adjusts the weights of the features according to the health during backend optimization; Equation (2): ; where is the health of the feature ; is the cumulative amount of perspective differences; is used to describe the sub-deviation amount; ,[[]]END]] are the weight coefficients of the perspective and appearance deviations respectively; , is the non-linear exponent; An observation failure function is constructed based on Equation (3); during the operation of the system, if the feature continues to be unable to match the observation, its observation failure degree Increases, indicating that the backend reduces the feature weight or eliminates its factor; Equation 3 amplifies or reduces the "failure trend" by combining time with health and matching confidence, maintaining the dynamic cleanliness of the optimized graph; Equation 3 is: ; Where is the observed failure degree; is the feature The length of time of continuous failure since the last successful observation; is the time factor function; is the feature 's health, which is based on Equation 2; is the confidence at the time of the last match; is the comprehensive amplification function.

[0010] In a preferred embodiment, a local re-optimization and snapshot fallback residual model is constructed based on Equation 4; when performing "local re-optimization" through Equation 4, the total residual of the subgraph or key frame subset is measured , after switching or falling back to the previous snapshot in the multiple hypothesis branches, the system re-projects and compares the affected key frames and features to repair the "apparent drift" deviation that has occurred; Equation 4 is expressed as: ; Where is the total local reprojection residual; represents the set of feature or key frame indices covered by this local re-optimization; is the feature or key frame corresponding state; is the time sensor bias; represents the true measurement in the observation space; is the projection function; represents the robust norm with a Huber kernel, is the kernel threshold; is the residual power; A time-temperature bias evolution equation is constructed based on Equation 5; where Equation 5 is used to counteract "dark drift" by combining with , and to continuously estimate and update the sensor bias; in addition, Equation 5 and the local re-optimization in Equation 4 operate simultaneously, enabling the system to resist both instantaneous drift and long-term drift; Equation 5 is: ; Where is the time The sensor bias vector; is the time derivative of the bias vector; is the moment temperature; represents at the moment The residual vector obtained by comprehensive statistics; is the instantaneous residual change; is the temperature coupling matrix; is the residual amplification function; is the bias attenuation function.

[0011] In a preferred embodiment, an absolute anchor global constraint is formed based on Equation VI; Equation VI is expressed as an absolute anchor residual, and Equation VI is used to introduce known external reference coordinates into the system to form a global constraint, thereby correcting the global coordinate system offset caused by "dark drift". By continuously minimizing , the system is kept aligned with the real-world reference in a large space or over a long time; Equation VI is: ; where is the anchor factor residual vector; is the global state; b is the current sensor bias; is the external reference observation; is the anchor residual calculation function.

[0012] In a preferred embodiment, a final state fusion update equation is constructed based on Equation VII; through Equation VII, the calculation results of Equations I to VI are combined into a unified state update process to generate the global state at the next moment ; Equation VII is expressed as: ; where , respectively represent the global states of the th frame and th frame; is the state synthesis operator; is the fusion function; is the local incremental solution; is the value of the sensor bias at the moment corresponding to the key frame .

[0013] The technical effects and advantages of the present invention: By detecting the instantaneous jump of "bright drift" in the short term and compensating for the sensor bias of "dark drift" in the long term, the deficiencies of traditional SLAM that cannot balance short-term mutations and long-term error accumulation are overcome, and the synchronous suppression of short-term positioning anomalies and long-term map distortions is achieved, thereby improving the accuracy and robustness of large VR spaces; With the help of the multiple hypothesis branch generation and local reprojection residual optimization mechanism, when the observation residual suddenly increases, a reliable branch can be quickly selected and local correction can be performed to avoid the error from spreading to the overall factor graph, and the global consistency is maintained by distinguishing reliable key frames from abnormal key frames; By establishing health and failure functions for feature factors and batch eliminating failed features in the background thread, the interference of invalid or abnormal features on the optimization process is reduced, and the overall optimization efficiency and state estimation quality are improved; Based on the bias evolution equation coupled with time-temperature, adaptive correction is performed for the drift caused by temperature fluctuations and long-term use. By dynamically updating the bias nodes in the non-linear optimization iteration, the sensor parameters are continuously calibrated, so that the system can have high pose estimation accuracy after long-term operation; By introducing the global constraint of absolute anchor points and combining with bias correction, the system can maintain alignment with the true coordinate system within a large range, and relatively avoid large-scale drift in the application of large VR spaces. Brief Description of the Drawings

[0014] Figure 1 It is a flowchart of the present invention. Detailed Embodiments

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] Refer to the attached Figure 1 For a method for evaluating the spatial mapping accuracy of a large VR space according to an embodiment of the present invention, it includes: By collecting sensor raw data in real time at the front end, performing feature extraction and time synchronization, and generating preliminary pose estimation, observation factors, and key frames; The front end stores the packed key frame data and sensor meta information in a circular buffer and transmits them to the back end according to the time stamp; After receiving the key frame packed data, the back end constructs pose, velocity, bias, and environmental feature state nodes for each key frame in the factor graph; By monitoring the observation residuals at the backend, trigger the generation of multiple hypothesis branches, independently construct the branches and then compare the residuals, select the target branch and discard the data of other branches, which is used to correct the "apparent drift" problem of instantaneous positioning jumps; The backend generates snapshots containing factor graphs and observation information at fixed key-frame intervals for subsequent invocation of the fallback mechanism; When the backend detects that the residual value exceeds the preset residual threshold and the pose state has not converged, the system will fallback to the pre-stored snapshot and load the corresponding observation data from the circular buffer; The backend constructs time- and temperature-dependent bias function nodes for the sensors in the factor graph to correct the sensor parameters and suppress the "dark drift" problem of long-term drift; During the non-linear optimization iteration process of the backend, the sensor bias function nodes are updated in real time according to the temperature- and time-related observation factors, and the global state estimation is adjusted synchronously.

[0017] The front end obtains the raw data from the sensors and uses feature extraction and time synchronization to generate preliminary pose estimates and observation factors; the raw data includes camera images, IMU readings, and laser point clouds; The front end selects key frames in the image sequence based on the motion amplitude and / or feature distribution, and packs the key frames together with the corresponding preliminary pose estimates and observation factors; The front end uses a circular buffer to store the packed key-frame data and retains the timestamp index for subsequent backtracking or branching operations; The front end sends the key-frame packed data and the sensor meta-information to the backend together as the input source for the backend global optimization.

[0018] By monitoring the residuals of the key frames in the observation factors at the backend, if the residual value exceeds the preset residual threshold, trigger the generation of multiple hypothesis branches; The backend shares the previous key-frame nodes at the divergence for multiple hypothesis branches and independently creates new key frames and constraint factors after the divergence; The backend compares the overall residuals of each branch in the parallel or polling graph optimization process to determine the target branch solution; After selecting the target branch, the backend marks the remaining inferior branches as discarded and reclaims the storage of related nodes and factors.

[0019] The backend establishes a health assessment model for each feature factor to quantify its observation quality, and gradually reduces its weight and / or marks it for rejection when the cumulative observation failures occur; The backend regularly scans the feature factors determined to be inefficient by the health assessment model in the background thread, clears them in batches, and performs a local re-optimization process after the clearing, which is used to reduce the local reprojection error and improve the state estimation accuracy; The backend configures anchor nodes within the factor graph and periodically applies global constraints based on absolute coordinate measurement factors to prevent cumulative pose drift; After receiving the backend factor graph update or branch merge signal, the frontend refreshes the indices of the key frames and their corresponding reference to the observation factors in the circular buffer, which is used to ensure the real-time consistency of the overall system.

[0020] It should be noted that in the formula structure involved in this solution, dimensionless terms can be used as proportional or structural adjustment factors. When combined with quantities with units, they only play a role in numerical scaling and do not introduce new physical dimensions. Therefore, they do not change or confuse the unit system of the overall expression. Such combinations of "dimensionless terms and terms with units" can be understood as the composite structure expression forms commonly used in mathematical and physical modeling, which conform to the principle of dimensional consistency and have a clear physical interpretation basis; Secondly, in the formula structure of this solution, if there are multiple variable terms with different physical units, including but not limited to time, mass, or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can be composed into a unified structure through function mapping, ratio combination, or normalization adjustment, with clear units and meanings. The overall expression conforms to the principle of dimensional consistency and the common norms of engineering modeling; In this solution, if constants, weights, adjustment factors, threshold parameters, proportional coefficients, etc. are designed, they all belong to adjustable control parameters for different application environments. Their values depend on the target device configuration, data input characteristics, and performance optimization goals, and converge and are set within a reasonable range through model verification, performance constraints, or engineering calibration during the implementation stage. Although these parameters do not have a preset unique value, they have a clear adjustment logic and calculation path, which belongs to the deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution has both general adaptability and reproducibility and operability, without affecting its technical clarity and implementability; This solution includes instantaneous and long-term drift compensation in the VR large-space environment mapping; the drift compensation includes drift processing, dark drift processing, and fusion update; In the bright drift processing, multiple hypothesis branch cost evaluation, local reprojection residual minimization, and snapshot rollback mechanisms are used to detect and correct instantaneous positioning jump problems; In the dark drift processing, the sensor bias is dynamically corrected through the time-temperature coupled bias evolution equation and absolute anchor global constraint, and all correction information is unified and integrated into the global state estimation through fusion update; A multiple hypothesis branch comprehensive cost function is established based on Equation (1); by comprehensively scoring the multiple hypothesis branches, to identify and correct the positioning mutations caused by "apparent drift". When the backend detects an instantaneous jump in the system, it calculates the overall cost of each branch based on Equation (1) to compare the performance of each branch in terms of observation residuals, instantaneous jump amplitude, feature quality, and branch duration; in addition, the optimal branch, that is, the one with the minimum cost, is regarded as the current credible solution, and the remaining branches are marked as discarded or rolled back, so as to stabilize the system position estimation in the shortest time and reduce the impact of large jumps on global mapping and tracking accuracy; Equation (1) is: ; where is the branch comprehensive cost, the larger the value of the branch contains the set of key frame indices; is the key frame the observation residual vector of, which is used to measure the gap between the measurement and estimation of this frame; is the key frame the residual difference between it and its adjacent key frames, used to capture the short-term drift amount; is the key frame the time weight coefficient of, and in practical applications, relatively high weights can be assigned to recent key frames; is the residual surge factor, which nonlinearly amplifies the impact of instantaneous jumps on the cost; is the key frame the corresponding feature health value; refers to the health decay function, and the lower the health of the key frame, the greater its contribution to the cost; represents the time span that the branch lasts; is the branch duration penalty term, and if the branch lasts too long, the penalty value of this term will increase; Regarding the formation of Equation (1), it includes: Weighted integration of key frames: Accumulate the residuals of all key frames within the branch and use to distinguish the importance of early frames and recent frames; Instantaneous jump detection: Pass to transmit the short-term residual change. If the jump amplitude is large, then pass to strengthen the cost of this key frame; Feature quality correction: If the health in the key frame is low, then correspondingly increase the penalty for this key frame; Duration constraint: Pass through Restrict the infinite delay of branches; Comprehensive assignment: Add the above factors to obtain , which is used for multi-branch comparison and selection of the optimal one.

[0021] Construct the feature health degree and observation failure degree functions based on Equation 2; Equation 2 is the feature Provide health degree quantification, reflecting the adaptability of the feature under different perspectives and appearances. The system screens stable feature factors through the value of Equation 2 and adjusts the weights of the features according to the health degree during backend optimization; in addition, features with lower health degrees are vulnerable to the influence of light or perspective changes. If they experience observation failures subsequently, they can also be excluded at the backend; Equation 2: ; Where is the health degree of the feature , and the value of is in the interval (0, 1]; is used as a feature identifier to indicate a certain map point or key frame feature; is the cumulative amount of perspective differences, and the cumulative amount of perspective differences is used to statistically analyze the feature in the difference degree of shooting angles in multiple frames; is used to describe the sub-deviation amount, and the sub-deviation amount is also a measure of the feature in the difference degree of appearance descriptors; , are the weight coefficients of perspective and appearance deviations respectively. In addition, in the above formula , are all positive values; , is the non-linear exponent, , is used to amplify or suppress the influence of the corresponding deviation on the health degree; Regarding the formation of Equation 2, it should be noted that it includes: Perspective data aggregation: Record the observation angles of the feature under multiple camera poses, and accumulate the differences to obtain ; Descriptor comparison: Compare the descriptors of the feature in different frames and calculate to quantify the appearance change; Deviation superposition: Weightedly superpose and and add 1; Reciprocal mapping: Take the reciprocal so that when both deviations are very low, is close to 1, and vice versa, the value drops sharply; Health degree output: Finally is provided for backend optimization and Equation 1 to call; Construct an observation failure degree function based on Equation Three; during the operation of the system, if a feature continually fails to match the observation, its observation failure degree increases, indicating that the backend should reduce the weight of this feature or eliminate its factor; by combining time with health and matching confidence, Equation Three amplifies or reduces the "failure trend" to maintain the dynamic cleanliness of the optimization graph; Equation Three is: ; where is the observation failure degree, and the larger the value of the observation failure degree, the higher the failure probability of the feature; is the time length of continuous failure of feature since the last successful observation; is the time factor function, which increases as increases; is the health of feature according to Equation Two; is the confidence at the last match, is associated with the similarity of the feature descriptor; is the comprehensive amplification function. When is low and is not high either, it will increase ; Regarding the formation of Equation Three, it includes: Continuous failure tracking: The system accumulatively counts and amplifies time through ; Health interaction: If is originally low, the failure degree will rise faster through ; Matching confidence association: The lower its , the larger, exacerbating the accumulation of the failure degree; Failure marking: When breaks through the preset threshold, the backend can eliminate the feature node or lower its optimization weight.

[0022] Construct a local re-optimization and snapshot rollback residual model based on Equation Four; when performing "local re-optimization" through Equation Four, the total residual of the subgraph or key frame subset is measured . After switching multiple hypothesis branches or rolling back to the previous snapshot, the system re-projects and compares the affected key frames and features to repair the "apparent drift" deviation that has occurred. In practical applications, by minimizing , the local state is restored to a credible position and attitude as soon as possible; Equation Four is expressed as: ; where is the total local reprojection residual; represents the set of feature or key - frame indices covered by this local re - optimization; is the state of the feature or key - frame corresponding thereto, which includes pose, velocity, bias and environmental feature state; is the time of the sensor bias, which is dynamically updated by Equation (5); represents the true measurement in the observation space, and the observation space includes image coordinates or laser ranging; is the projection function used to map to the observation domain; represents the robust norm with Huber kernel, is the kernel threshold; in Equation (4), is the residual power, which is used to non - linearly strengthen or suppress error values within a certain range; Regarding the formation of Equation (4), it includes: Local index determination: It is formed according to the set of key - frames covered by multiple hypothesis branches or fallback snapshots ; Projection error calculation: For each , estimate the observation value with the projection function and compare it with ; Robust kernel correction: Suppress the interference of extreme anomalies through ; Non - linear degree: Moderately amplify medium errors or weaken small noises through ; Local minimization: Obtain the updated state and bias by iteratively solving , and repair the influence of obvious drift on the local map; Construct a time - temperature bias evolution equation based on Equation (5); where Equation (5) is used to counter "dark drift" by combining with , and to continuously estimate and update the sensor bias, reducing the accumulation of long - term errors at the hardware level; in addition, Equation (5) and the local re - optimization in Equation (4) operate simultaneously, enabling the system to resist both instantaneous drift and long - term drift; Equation (5) is: ; where is the sensor bias vector at time ; is the time derivative of the bias vector; is the moment temperature; represents at the moment the residual vector obtained by comprehensive statistics; is the instantaneous residual change, which is used to capture the instantaneous information coupled with the "apparent drift"; is the temperature coupling matrix, which represents the influence intensity of temperature deviation on bias change; is the residual amplification function, which combines the long-term and short-term to moderately adjust the bias; is the bias attenuation function, which is used to prevent the unbounded growth of the bias vector; Regarding the formation of Equation Five, it includes: Temperature coupling: From map the temperature difference to the influence on the bias; Residual injection: Through simultaneously absorb and information, so that the bias can sense the overall error and instantaneous change; Self-attenuation mechanism: prevent the bias from developing towards infinity, and once the bias value is too high, it will be offset by the negative feedback term; Discrete update: In the optimization iteration at the backend, is discretized, and the bias value is gradually corrected, thereby suppressing the dark drift in the long term; represents the time step in the discrete optimization iteration.

[0023] Based on Equation Six, an absolute anchor global constraint is formed; Equation Six represents the absolute anchor residual, and Equation Six is used to introduce the known external reference coordinates into the system to form a global constraint, thereby correcting the global coordinate system offset caused by the "dark drift". By continuously minimizing , the system can maintain alignment with the real-world reference in a large space or over a long time, and avoid the long-term pose from getting out of control; Equation Six is: ; where is the anchor factor residual vector, which is used to quantify the deviation between the current estimate and the absolute reference coordinates; is the global state, which includes pose, velocity, bias, and environmental feature state; b is the current sensor bias, which is given by Equation Five; is the external reference observation; is the anchor residual calculation function, by mapping to the global coordinate system and comparing with Alignment; Regarding the formation of Equation VI, it includes: Anchor point data acquisition: Measuring the absolute position / attitude at a specified location or marker ; System projection: Using to transform the current system estimate into the same world coordinate system; Comparing to obtain the residual: Through to reflect the difference between the system estimate and the actual anchor point value; Global correction: In the backend optimization, minimize to correct the deviation caused by the accumulation of dark drift.

[0024] Construct the final state fusion update equation based on Equation VII; Merge the calculation results of Equations I to VI into a unified state update process through Equation VII to generate the global state at the next moment , which is equivalent to the "master fusion" step of the system, integrating the correction information of both "bright drift" and "dark drift", so that the overall pose and map structure always remain highly consistent in each key frame iteration; Equation VII is expressed as:

[0025] where , respectively represent the global states of the th frame and th frame; is the state synthesis operator, including the pose superposition in quaternion or Lie group; is the fusion function, which synthesizes local increment, bias correction, anchor point constraint, local residual, and branch cost; is the local increment solution, which comes from the pose or velocity update after local re-optimization; is the value of the sensor bias at the moment corresponding to the key frame , dynamically updated by Equation V and estimating its value; obtained from Equation VI; obtained from Equation IV; obtained from Equation I; Regarding the formation of Equation VII, it includes: Input integration: The backend inputs , , , and into the fusion function ; Fusion strategy: According to Select the optimal branch and use to make local corrections to the pose, and then use to improve the global constraints while correcting the influence on the observation residuals; State synthesis: Through operations, accumulate the above fusion results to to generate a new state ; Loop iteration: The system repeats this process to achieve dynamic detection and correction of "bright drift" and "dark drift", enabling continuous environmental mapping and positioning in a large VR space.

[0026] In addition, it needs to be further explained that; In the processing of bright drift, that is, instantaneous jump, the multiple hypothesis branch cost of Equation 1 is relied on to screen the sudden increase in residuals, and combined with the local residuals of Equation 4 for rapid re-optimization after snapshot rollback or branch switching; in addition, the feature health (Equation 2) and failure degree (Equation 3) dynamically eliminate or downweight unreliable features, further reducing the interference of "instantaneous anomalies" on the global system; In the processing of dark drift, that is, long-term accumulation processing, the differential continuous tracking of the sensor bias with time and temperature is carried out through Equation 5 to avoid the long-term accumulation of internal drift. According to Equation 6, an external anchor point is introduced for absolute constraint to correct the offset of the global coordinate system, and relative accuracy can still be maintained in large spaces or long-time scenarios; In the fusion update, to maintain overall consistency, the multi-branch cost, local residuals, bias correction, anchor point constraint, and local increment are unified into the fusion function through Equation 7 to form the final global update of the system state, so as to achieve both timely repair of "bright drift" and continuous suppression of "dark drift", ensuring the relative stability and accuracy of mapping and positioning in a large VR space environment.

[0027] As a further embodiment of the above solution, it includes: When the difference in observation residuals between adjacent key frames exceeds a preset threshold and the health of the same candidate feature factor is lower than the corresponding preset threshold, the system triggers a multiple hypothesis branch evaluation based on the combined judgment result and decides whether to perform local re-optimization according to the threshold comparison to correct the instantaneous positioning jump in a timely manner; When the same candidate feature factor fails to be successfully matched continuously within a preset time period and its deviation from the absolute anchor point exceeds the preset limit, the system regards this feature factor as invalid according to the combined judgment result and corrects the overall pose through the absolute anchor point global constraint. At the same time, update the sensor bias node in the subsequent non-linear optimization iteration to suppress long-term drift; For the above two groups of combined judgments, they are used to deal with short-term mutations and long-term degradations respectively; The first set of determination conditions is that "the residual difference between adjacent key frames is greater than the threshold" and "the health of the same candidate feature factor is lower than the threshold" coexist. When both occur simultaneously, it indicates that the feature factor may have an instantaneous jump within a short period of time. The system will trigger a multiple hypothesis branch evaluation and decide whether to perform local re-optimization based on the threshold comparison, so as to quickly correct the "apparent drift". The second set of determination conditions is that "the same candidate feature factor continuously fails to be successfully matched within a preset duration" and "its deviation from the absolute anchor point exceeds the preset limit" coexist. This situation means that the feature factor has suffered irreparable degradation or deviation during long-term use. At this time, the system will determine it to be invalid and re-correct the overall pose based on the global constraint of the absolute anchor point. At the same time, the sensor bias node will be updated during subsequent non-linear optimization iterations to suppress the further accumulation of long-term drift. By distinguishing these two sets of combined determinations, the algorithm can not only promptly capture the instantaneous jump situation but also detect the long-term drift situation, thereby taking into account the synchronous control of apparent drift and hidden drift in the large VR space.

[0028] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating the accuracy of large-scale VR space mapping, comprising: The front end collects raw sensor data in real time, performs feature extraction and time synchronization, and generates preliminary pose estimation, observation factors, and key frames; The front end stores the packaged key frame data and sensor metadata into a ring buffer and transmits it to the back end according to the timestamp; Features: After receiving the keyframe packaged data, the backend constructs the pose, velocity, bias and environmental feature state nodes for each keyframe in the factor graph; Through the back-end monitoring observation residuals, multiple hypothesis branches are generated, and the residuals are compared after the branches are independently constructed. The target branch is selected and the data of other branches are discarded to correct the "bright drift" problem of instantaneous positioning jumps; The backend generates snapshots containing factor graphs and observation information at fixed keyframe intervals for subsequent fallback mechanism calls; When the backend detects that the residual value exceeds the preset residual threshold and the pose state has not converged, the system will fall back to the pre-stored snapshot and load the corresponding observation data from the ring buffer; The backend constructs time- and temperature-dependent bias function nodes for the sensor in the factor graph to correct sensor parameters and suppress the "dark drift" problem of long-term drift; During the back-end nonlinear optimization iteration process, the sensor bias function nodes are updated in real time according to the temperature and time-related observation factors, and the global state estimation is adjusted synchronously.

2. A VR large space spatial mapping accuracy assessment method according to claim 1, characterized in that: The front end obtains raw data from sensors and generates preliminary pose estimates and observation factors using feature extraction and time synchronization; the raw data includes camera images, IMU readings, and laser point clouds; The front end selects key frames in the image sequence based on motion amplitude and / or feature distribution, and packages the key frames with the corresponding preliminary pose estimation and observation factors; The front end uses a ring buffer to store the packed keyframe data and retain the timestamp index for subsequent backtracking or branching operations; The front end sends the key frame packaged data and sensor metadata to the back end as the input source for the back end global optimization.

3. A VR large space spatial mapping accuracy assessment method according to claim 2, characterized in that: The residual of the key frame in the observation factor is monitored through the backend. If the residual value exceeds the preset residual threshold, the generation of multiple hypothesis branches is triggered; The backend is that multiple hypothesis branches share previous keyframe nodes at divergence, and independently create new keyframes and constraint factors after divergence; The backend compares the overall residuals of each branch in a parallel or round-robin graph optimization process to determine the target branch solution; After selecting the target branch, the backend discards the remaining inferior branches and recycles the related node and factor storage.

4. A VR large space spatial mapping accuracy assessment method according to claim 3, characterized in that: A health assessment model is established for each feature factor through the backend to quantify its observation quality, and its weight is gradually reduced and / or marked for removal when observation failures accumulate; The backend periodically scans the feature factors that are determined to be inefficient by the health assessment model in the background thread, and clears them in batches, and executes the local re-optimization process after the clearing; The backend configures anchor nodes in the factor graph and periodically applies global constraints based on the absolute coordinate measurement factors; After receiving the backend factor graph update or branch merge signal, the front end refreshes the index of the key frame and its corresponding observation factor reference in the ring buffer.

5. A VR large space spatial mapping accuracy assessment method according to claim 4, characterized in that: Based on Formula 1, a comprehensive cost function of multiple hypothesis branches is established; by comprehensively scoring multiple hypothesis branches, the positioning mutation caused by "bright drift" is identified and corrected. When the backend detects a transient jump in the system, the overall cost of each branch is calculated based on Formula 1 to compare the performance of each branch in terms of observation residual, transient jump amplitude, feature quality, and branch duration; Formula 1 is: ； in is the branch comprehensive cost; For branch The set of keyframe indices included; Keyframe The observed residual vector of ; Keyframe The residual difference between its adjacent keyframes; Keyframe The time weight coefficient of is the residual surge factor; Keyframe The corresponding characteristic health value; Refers to the health decay function; Indicates the time span that the branch lasts; is the branch duration penalty term.

6. A VR large space spatial mapping accuracy assessment method according to claim 5, characterized in that: Based on formula 2, the characteristic health and observation failure function are constructed; formula 2 is the characteristic Provides health quantification to reflect the adaptability of the feature under different viewing angles and appearance conditions. The system uses the value of formula 2 to screen feature factors and adjusts the weight of the feature according to the health during backend optimization; Formula 2: ； in Features health; is the cumulative amount of visual angle difference; Used to describe the sub-deviation; , are the weight coefficients of viewing angle and appearance deviation respectively; , is the nonlinear index; Based on formula 3, the observation failure function is constructed; during system operation, if the characteristic If it continues to fail to match the observation, then its observation failure Increase, prompting the backend to reduce the weight of the feature or remove its factor; Formula 3 combines time with health and matching confidence to amplify or reduce the "failure trend" and maintain the dynamic cleanliness of the optimization graph; Formula 3 is: ； in is the observed failure degree; Features The length of time since the last successful observation that the failure has occurred; is the time factor function; Features The health degree is based on formula 2; is the confidence level of the last match; is the comprehensive amplification function.

7. A VR large space spatial mapping accuracy assessment method according to claim 6, characterized in that: Based on formula 4, a local reoptimization and snapshot backoff residual model is constructed; when performing "local reoptimization" through formula 4, the residual sum of the metric subgraph or keyframe subset is calculated. After multiple hypothesis branches are switched or rolled back to a previous snapshot, the system reprojects the affected keyframes and features to repair the "bright drift" deviation that has occurred; Formula 4 is expressed as: ； in is the total value of the local reprojection residual; Represents the set of feature or keyframe indices covered by this local reoptimization; For features or keyframes The corresponding status; For the moment Sensor bias; represents the real measurement in the observation space; is the projection function; represents the robust norm with Huber kernel, is the nuclear threshold; is the residual power; The time-temperature bias evolution equation is constructed based on Equation 5; Formula 5 is used to combat "dark drift" by and , as well as Combined with Eq. 5, the sensor bias is continuously estimated and updated; in addition, the local reoptimization in Eq. 5 and Eq. 4 operates simultaneously, so that the system can resist both instantaneous drift and long-term drift. Formula 5 is: ； in For the moment The sensor bias vector of is the time derivative of the bias vector; For the moment Temperature; Indicates at time The residual vector obtained by comprehensive statistics; is the instantaneous residual change; is the temperature coupling matrix; is the residual amplification function; is the bias attenuation function.

8. A VR large space spatial mapping accuracy assessment method according to claim 7, characterized in that: Based on Equation 6, an absolute anchor point global constraint is formed; Equation 6 is expressed as the absolute anchor point residual. Equation 6 is used to introduce the known external reference coordinates into the system to form a global constraint, thereby correcting the global coordinate system offset caused by "dark drift". By continuously minimizing , allowing the system to maintain alignment with a real-world reference over large spaces or long periods of time; Formula 6 is: ； in is the anchor factor residual vector; is the global state; b is the current sensor bias; For external benchmark observations; Calculate the function for the anchor point residual.

9. A VR large space spatial mapping accuracy assessment method according to claim 8, characterized in that: The final state fusion update equation is constructed based on equation 7; the calculation results of equations 1 to 6 are merged into a unified state update process through equation 7 to generate the global state at the next moment ; Formula 7 is expressed as: ； in , Respectively represent Frame and The global state of the frame; is the state synthesis operator; is the fusion function; is a local incremental solution; For sensor bias at keyframe The value of the corresponding moment.

Images