Search

CN-121985029-A - Cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking

CN121985029ACN 121985029 ACN121985029 ACN 121985029ACN-121985029-A

Abstract

The invention provides a cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking, which comprises the following steps of using req_id timestamp as an asynchronous alignment mechanism of an index, caching a space-time backtracking annular buffer zone of a multidimensional evidence chain, backtracking reintegration reconstruction and injection mapping from Treq to Tnow, combining confidence level with asynchronous fusion updating of a door control, flexible injection non-jump mechanism of a rendering side, triggering communication strategy and power consumption/bandwidth self-adaption, and reversely aligning and projecting 'past moment authenticity' returned by a cloud to 'current moment trace' by utilizing a relative motion evidence chain (such as IMU original data, characteristic point tracking trace, pose transformation matrix/increment and the like) cached by a terminal, and completing fusion updating in a non-jump mode, so that AR/MR rendering stability can be maintained under high-delay, jitter network and disordered response conditions.

Inventors

  • CHEN HUICHONG

Assignees

  • 武汉华创全息数字科技有限公司

Dates

Publication Date
20260505
Application Date
20260120

Claims (10)

  1. 1. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking is characterized by comprising the following steps of: (1) Asynchronous alignment mechanism indexed by req_id timestamp; (2) Caching a space-time backtracking annular buffer zone of the multidimensional evidence chain; (3) Retrospective re-integration reconstruction and injection mapping from Treq to Tnow; (4) Asynchronous fusion updating combining confidence and gating; (5) A flexible injection non-jump mechanism at the rendering side; (6) Triggered communication strategies and power consumption/bandwidth adaptation.
  2. 2. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking of claim 1, wherein in step (1), the asynchronous alignment mechanism comprises: The sensor acquisition module is used for acquiring camera image streams and IMU original data; The local relative positioning module (VIO/equivalent module) outputs continuous relative movement results for real-time rendering and generates a relative movement evidence chain; and the timestamp and request identification module is used for generating a unique request IDreq _id for each cloud request and recording a request time timestamp Treq (which can be monotone clock time).
  3. 3. The cloud VPS and local VIO co-location method based on asynchronous timestamp traceback of claim 1, wherein in step (2), the spatiotemporal traceback ring buffer comprises a spatiotemporal traceback ring buffer module, and the spatiotemporal traceback ring buffer module caches a 'relative motion evidence chain' (not only pose) with time as an index to support subsequent traceback and re-integration reconstruction.
  4. 4. The method according to claim 1, wherein in step (3), the backward integration reconstruction and injection mapping process is performed by using a cloud request generation module, and the cloud request generation module sends visual information in one or more "data state" packets (original image/feature point cloud/descriptor vector/semantic feature, etc.).
  5. 5. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking according to claim 1, wherein in the step (4), an asynchronous fusion correction module is used in the asynchronous fusion update, and the asynchronous fusion correction module receives a global pose returned by the cloud and performs time sequence alignment, backtracking reconstruction, residual calculation and fusion update.
  6. 6. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking of claim 1, wherein in step (5), the flexible injection non-jump mechanism is used in a flexible injection and rendering output module, and the flexible injection and rendering output module propagates a fusion result to a rendering pose sequence in a non-jump mode to avoid instant jump.
  7. 7. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking of claim 1, wherein according to step (6), the triggered communication policy includes a triggered communication policy module that triggers or suppresses a cloud request according to drift, feature degradation, illumination variation, network quality, power consumption policy, etc.
  8. 8. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking according to claim 2, wherein the key data structure of the multi-dimensional evidence chain is a space-time backtracking annular buffer B (evidence chain buffer), in order to achieve deterministic alignment under the condition of delay and disorder, the annular buffer B is maintained at a terminal side and used for buffering the relatively moving evidence chain from history to current, the buffer capacity is N, and the coverage duration is not less than the maximum end-to-end round trip delay upper bound RTTmax plus a safety margin (for example, 1-3 seconds or longer for coverage, depending on the scene).
  9. 9. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking according to claim 3, wherein the backtracking and reintegration reconstruction comprises a pose backtracking and reintegration reconstruction algorithm (core: projected to the current from a past reality), pvps (Treq) returned by the cloud corresponds to a past observation time Treq, and accumulated relative motion from the Treq to a current time Tnow is reconstructed through a buffer B, so that the cloud reality can be mapped to the current time.
  10. 10. The cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking as recited in claim 6, wherein said trigger communication policy (Power consumption/Bandwidth optimization+coverage more trigger conditions) To reduce cloud invocation and power consumption, a triggered VPS request policy is provided, and the trigger may consist of one or more of the following conditions (scalable set): drift accumulation reaches a threshold η (any dimension such as position/heading/scale); degradation of visual quality, reduction of the number of feature points, increase of tracking failure rate, and reprojection The error increases; the environmental change is that the illumination change rate delta L > gamma and the dynamic shielding proportion are increased; Kinematic change, acceleration/angular velocity jump, velocity continuously higher than threshold; network and power consumption strategy, if the network is good, the request frequency is increased, and if the network is low-power/weak Reducing the frequency; task policy forcing requests at critical areas/critical nodes to guarantee content anchoring The trigger policy may be written as either a rule or a learning policy.

Description

Cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking Technical Field The invention relates to the technical field of large-scale space calculation and augmented reality/mixed reality positioning navigation, in particular to a cloud VPS and local VIO co-positioning method based on asynchronous timestamp backtracking. Background In large-scale space computation (Spatial Computing) and AR/MR application, in order to obtain sharable and long-term reusable global coordinates in a large-scale and trans-regional environment, an end-cloud co-location architecture of end-side continuous tracking and cloud global repositioning is commonly adopted in the industry, and typical processes of the end-cloud co-location architecture comprise end-side continuous tracking, cloud global positioning and end-side alignment/correction; In the above end cloud collaborative architecture, the cloud VPS results are not generated in real time. Because links such as network transmission, concurrent queuing, scheduling, search matching and optimization resolving are overlapped, the cloud generally has non-negligible time delay from receiving a request to returning a result, and the time delay has obvious time variability and uncontrollability. The delay often ranges from hundreds of milliseconds to seconds, and may further deteriorate under weak network, congestion or long link conditions, and under concurrent request or network jitter conditions, there is a possibility that the cloud Response has Out-of-order Response (Out-of-order Response) in addition to delay fluctuation, that is, a later-issued request may return first, a first-issued request returns instead, thereby causing the traditional processing logic based on ' first-in-first-Out ' (FIFO) ', and further amplifying uncertainty and instability of pose injection. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking to solve the problems in the prior art, and the invention has novel structure, and utilizes a relative motion evidence chain (such as IMU original data, feature point tracking tracks, pose transformation matrix/increment and the like) which is cached by a terminal and contains a multidimensional motion state to reversely align and project 'past moment truth' returned by the cloud to 'current moment track', and then finish fusion updating in a non-jump mode, thereby still keeping AR/MR rendering stable under the conditions of high delay, jitter network and disorder response. In order to achieve the above purpose, the invention is realized by the following technical scheme that the cloud VPS and local VIO co-location method based on asynchronous timestamp backtracking comprises the following steps: (1) Asynchronous alignment mechanism indexed by req_id timestamp; (2) Caching a space-time backtracking annular buffer zone of the multidimensional evidence chain; (3) Retrospective re-integration reconstruction and injection mapping from Treq to Tnow; (4) Asynchronous fusion updating combining confidence and gating; (5) A flexible injection non-jump mechanism at the rendering side; (6) Triggered communication strategies and power consumption/bandwidth adaptation. Further, according to step (1), the asynchronous alignment mechanism includes: The sensor acquisition module is used for acquiring camera image streams and IMU original data; The local relative positioning module (VIO/equivalent module) outputs continuous relative movement results for real-time rendering and generates a relative movement evidence chain; and the timestamp and request identification module is used for generating a unique request IDreq _id for each cloud request and recording a request time timestamp Treq (which can be monotone clock time). Further, according to step (2), the space-time traceback ring buffer includes a space-time traceback ring buffer module, and the space-time traceback ring buffer module caches a "relative motion evidence chain" (not only pose) by using time as an index, so as to support subsequent traceback and reintegration reconstruction, and the relative motion evidence chain adopts a generalized data structure decoupled from a sensor, so that even if the IMU original data is missing, the system can complete traceback by using an inter-frame transformation matrix sequence generated by a Visual Odometer (VO), and universality of the scheme under different hardware architectures (such as pure visual AR glasses) is ensured. Further, according to the step (3), a cloud request generation module is used in the traceback re-integration reconstruction and injection mapping process, the cloud request generation module sends visual information in one or more data state packets (original image/feature point cloud/descriptor vector/semantic feature, etc.), in addition, an overflow fusing mechanism is built in the recon