Search

CN-121996917-A - Sequential abnormal observation eliminating method and device for two-dimensional code auxiliary positioning

CN121996917ACN 121996917 ACN121996917 ACN 121996917ACN-121996917-A

Abstract

The application discloses a sequential anomaly observation removing method and device for two-dimensional code auxiliary positioning, which relate to the technical field of two-dimensional code auxiliary positioning and solve the problem of insufficient robustness of the existing two-dimensional code data processing method, and the method comprises the following steps: the acquired two-dimensional code observation data are subjected to multistage sequential anomaly detection which is sequentially composed of offset filtering, adjacent measurement consistency checking, speed-based dynamic symbol checking, angle change rate checking, time sequence continuity and node screening, the anomaly data are removed, and an exponential moving average algorithm is adopted to carry out smooth fusion on the residual high-quality data so as to obtain state estimation, so that the accuracy, stability and reliability of the two-dimensional code-based positioning navigation system in a complex real environment are remarkably improved, and meanwhile, the sequential strategy has the advantages of being strong in real-time performance and small in calculation amount compared with a batch processing method.

Inventors

  • QU PEIPEI
  • LI WEIJUN
  • HU ZHIGUANG
  • Huang Lielie

Assignees

  • 浙江迈睿机器人有限公司

Dates

Publication Date
20260508
Application Date
20260121

Claims (10)

  1. 1. A sequential anomaly observation and rejection method for two-dimensional code auxiliary positioning is characterized by comprising the following steps: Acquiring two-dimensional code observation data; performing multistage sequential anomaly detection on the two-dimensional code observation data, wherein the multistage sequential anomaly detection sequentially comprises offset filtering, adjacent measurement consistency checking, dynamic symbol checking based on speed, angle change rate checking, time sequence continuity and node screening; and carrying out smooth fusion on the two-dimensional code observation data detected by the multilevel sequential anomaly by adopting an exponential moving average algorithm to obtain state estimation.
  2. 2. The two-dimensional code assisted positioning-oriented sequential anomaly observation rejection method according to claim 1, wherein the offset filtering comprises: setting an offset threshold; respectively judging whether the offset in the X direction and the Y direction exceeds a set offset threshold value; if the offset in the X direction or the Y direction of the two-dimensional code observation data exceeds the set offset threshold, rejecting, otherwise, reserving.
  3. 3. The two-dimensional code assisted positioning-oriented sequential anomaly observation rejection method according to claim 1, wherein the adjacent measurement consistency check comprises: checking consistency of two adjacent measurements by using time sequence correlation; if the labels are different but the offset is the same, or if the offset is strictly zero but the difference from the previous one is large, the label is regarded as abnormal and is removed.
  4. 4. The method for sequentially observing and rejecting anomalies in a two-dimensional code-oriented assisted positioning of claim 1, wherein the speed-based dynamic symbol inspection comprises the steps of measuring jump or mismatching and rejecting if adjacent measured offsets are subjected to symbol inversion within 1.5s when the movement speed exceeds 0.8m/s, wherein the product of the adjacent measured offsets is smaller than zero to indicate the symbol inversion, and the minimum threshold value is exceeded by the offset to perform the dynamic symbol inspection.
  5. 5. The method for sequentially observing and rejecting anomalies for two-dimensional code assisted positioning according to claim 1, wherein the angle change rate inspection comprises the steps of considering false detection and rejecting if the deviation of the heading angle observed by the two-dimensional code and the current pose exceeds 0.03 radian twice continuously when the movement speed of a robot exceeds 0.4m/s in a positioning mode.
  6. 6. The two-dimensional code assisted positioning-oriented sequential abnormal observation eliminating method according to claim 1 is characterized by comprising the steps of carrying out time sequence analysis on continuous observation of the same two-dimensional code, eliminating two-dimensional code observation data of repeated measurement and angle mutation, wherein the repeated measurement refers to measurement with identical positions and angles, and the angle mutation refers to heading angle which exceeds 3 times of theoretical variation based on gyroscope angular speed prediction.
  7. 7. The two-dimensional code assisted positioning-oriented sequential anomaly observation rejection method according to any one of claims 1 to 6, wherein the step of performing smooth fusion on two-dimensional code observation data detected by multistage sequential anomaly detection by using an exponential moving average algorithm to obtain a state estimate comprises: Obtaining a filtered observation pose at the current moment, wherein the observation pose comprises position observation and angle observation, and obtaining a historical estimated pose at the previous moment and the motion speed of a current carrier; Calculating a measurement quality factor representing the current observation quality based on the offset between the observation pose and the historical estimation pose; calculating a speed influence factor based on the movement speed of the current carrier, wherein the speed influence factor corresponding to the angle smoothing decays along with the speed increase and is larger than the speed influence factor corresponding to the position smoothing decays along with the speed increase; adaptively calculating a first learning rate for position smoothing and a second learning rate for angle smoothing according to the measured quality factor and the speed influence factor; Performing exponential moving average calculation on the position observation and the historical estimation position by using the first learning rate to obtain a smoothed position at the current moment; Converting the angle observation and the historical estimation angle into vector representations on a unit circle respectively, carrying out index moving average calculation on the vector by utilizing the second learning rate, and then restoring the vector after weighted average through atan2 to obtain a smoothed angle at the current moment; and outputting the final estimated pose of the current moment formed by the smoothed position and the smoothed angle.
  8. 8. The method for eliminating sequential anomaly in assisted positioning of two-dimensional codes according to claim 7, wherein when a first learning rate for position smoothing and a second learning rate for angle smoothing are obtained through self-adaptive calculation, a time factor is further introduced for adjustment, the time factor is calculated according to a time interval between a current observation and a previous observation, and when the time interval exceeds a preset threshold, the values of the first learning rate and the second learning rate are improved, and an upper limit exists for the improvement amplitude.
  9. 9. A two-dimensional code auxiliary positioning-oriented sequential abnormal observation removing device is characterized by comprising: The data acquisition module is used for acquiring the two-dimensional code observation data; the multistage filtering module is used for carrying out multistage sequential anomaly detection on the two-dimensional code observation data, wherein the multistage sequential anomaly detection sequentially comprises offset filtering, adjacent measurement consistency checking, speed-based dynamic symbol checking, angle change rate checking, time sequence continuity and node screening; And the self-adaptive smooth fusion module is used for carrying out smooth fusion on the two-dimensional code observation data detected by the multilevel sequential anomaly by adopting an exponential moving average algorithm so as to obtain state estimation.
  10. 10. A computer readable storage medium storing program code for execution by a device, the program code comprising steps for performing the method of any one of claims 1-8.

Description

Sequential abnormal observation eliminating method and device for two-dimensional code auxiliary positioning Technical Field The application relates to the technical field of two-dimensional code auxiliary positioning, in particular to a sequential anomaly observation eliminating method and device for two-dimensional code auxiliary positioning. Background In the positioning navigation module of intelligent systems such as mobile robots (e.g. AGV, AMR), automatic driving vehicles, etc., two-dimensional codes (e.g. AprilTag, arUco, etc. visual tags) are widely used as a stable, low-cost Landmark (Landmark) to provide absolute position references for correcting odometer accumulated errors, assisting SLAM (synchronous positioning and mapping) processes, especially in structured environments (e.g. warehouse, factory shop). The system detects and calculates the pose (including the position offset x, y and yaw angle yaw) of the two-dimensional code relative to the robot in real time through the mounted camera. The current mainstream method for processing the two-dimensional code observation data generally directly uses the original calculation result of the sensor, or only performs simple data validity check (such as a confidence threshold). The method specifically comprises the following steps: 1. The pose data calculated by the visual algorithm is directly sent into a filter (such as Kalman filtering and graph optimization), and noise is processed by depending on the robustness of the filter. 2. Simple threshold filtering-a fixed threshold is set for the uncertainty (covariance) or confidence of the resolved pose, and data below the threshold is discarded. 3. Time stamp alignment and interpolation, namely time synchronization is carried out on two-dimensional code observation and other sensor data (such as IMU and wheel speed meter). When facing a complex practical application scene, the existing method lacks deep verification of physical rationality of observed data, so that a positioning system is easy to be interfered by abnormal observation: 1. The method can not effectively deal with dynamic interference, namely when the robot passes through the two-dimensional code rapidly, the image blurring, partial shielding or illumination mutation can cause a transient but huge jump error (Outlier) in visual calculation, and the real movement and the abnormal jump are difficult to distinguish by a simple threshold. 2. And neglecting the time sequence correlation among the data, namely, only independently judging single observation, and failing to identify a continuous observation sequence (such as high-frequency oscillation of a two-dimensional code position symbol) which appears in a short time and is not physically possible to appear. 3. The sensor is sensitive to the installation error and calibration error, namely, the small deviation of the installation angle can cause continuous observation error, and the simple fixed threshold value cannot be self-adaptive. 4. The filter divergence or map distortion may be caused by that once the abnormal observed values are incorporated into the positioning algorithm, the state estimation may be polluted, the positioning drift and track jitter may be caused, even the ghost or structural distortion may occur in the map of the SLAM, and the reliability and accuracy of the system may be seriously affected. Disclosure of Invention The application aims to solve the problem of insufficient robustness of the existing two-dimensional code data processing method and provides a sequential anomaly observation eliminating method and device for two-dimensional code auxiliary positioning. In a first aspect, a sequential anomaly observation rejection method for two-dimensional code assisted positioning is provided, including: Acquiring two-dimensional code observation data; performing multistage sequential anomaly detection on the two-dimensional code observation data, wherein the multistage sequential anomaly detection sequentially comprises offset filtering, adjacent measurement consistency checking, dynamic symbol checking based on speed, angle change rate checking, time sequence continuity and node screening; and carrying out smooth fusion on the two-dimensional code observation data detected by the multilevel sequential anomaly by adopting an exponential moving average algorithm to obtain state estimation. In some possible implementations, the offset filtering includes: setting an offset threshold; respectively judging whether the offset in the X direction and the Y direction exceeds a set offset threshold value; if the offset in the X direction or the Y direction of the two-dimensional code observation data exceeds the set offset threshold, rejecting, otherwise, reserving. In some possible implementations, the neighbor measurement consistency check includes: checking consistency of two adjacent measurements by using time sequence correlation; if the labels are different but the offset