Search

CN-122021679-A - Two-dimensional code image signal processing method and system for AGV navigation

CN122021679ACN 122021679 ACN122021679 ACN 122021679ACN-122021679-A

Abstract

The invention discloses a two-dimensional code image signal processing method and system for AGV navigation; the method comprises the steps of collecting an original image containing a two-dimensional code, carrying out noise suppression, contrast reconstruction and local structure compensation through an image restoration module, outputting an enhanced image, determining two-dimensional code corner points or boundaries through a two-dimensional code positioning module, obtaining a standardized two-dimensional code area image through geometric correction, inputting the standardized image into a decoding module, carrying out semantic feature extraction and sequence prediction through a deep neural network, directly outputting coding information and confidence level, and finally transmitting the information to an AGV control system for navigation decision. The system correspondingly comprises image acquisition, restoration, positioning correction, decoding and interface modules, and each module is matched with the embedded platform through light weight optimization, so that processing delay is 30ms. The method can effectively cope with complex industrial scenes such as greasy dirt, dust, defects and the like, and remarkably improves the robustness of two-dimension code identification and the reliability of AGV navigation.

Inventors

  • LI HONGWANG
  • ZHOU RUIFENG

Assignees

  • 深圳市福联实业有限公司

Dates

Publication Date
20260512
Application Date
20260202

Claims (10)

  1. 1. The two-dimensional code image signal processing method for AGV navigation is characterized by comprising the following steps: S1, acquiring an original image containing a two-dimensional code through an image acquisition device arranged on an AGV body; s2, inputting the original image into an image restoration module, performing noise suppression, contrast reconstruction and local structure compensation processing, and outputting an enhanced image; s3, inputting the enhanced image into a two-dimensional code positioning module, determining the angular point position or boundary region of the two-dimensional code in the image, and performing geometric correction to obtain a standardized two-dimensional code region image; S4, inputting the standardized two-dimensional code area image into a decoding module, extracting semantic features and predicting sequences through a deep neural network, and outputting coding information and confidence corresponding to the two-dimensional code; And S5, outputting the encoded information to an AGV control system for AGV positioning or path planning.
  2. 2. The two-dimensional code image signal processing method for AGV navigation according to claim 1 wherein said image restoration module employs a convolutional neural network structure comprising an encoder and a decoder and is optimized in training using a composite loss function comprising at least pixel level loss, perceptual loss and structural consistency loss.
  3. 3. The two-dimensional code image signal processing method for AGV navigation according to claim 2, wherein the training sample of the image restoration module is generated by a synthesis method, comprising: Simulating an oil stain layer by using Perlin noise, simulating dust by using Gaussian noise and speckle noise superposition, and simulating image tearing or defect by using a random polygonal mask; The composite sample is combined with the true impairment-sharp image to form a hybrid training set.
  4. 4. The two-dimensional code image signal processing method for AGV navigation according to claim 1 wherein the two-dimensional code positioning module predicts the coordinates of the corner points through a regression network or detects the positions of the corner points through a heat map network, and performs geometric correction through a Space Transformation Network (STN) based on the position information to obtain a standardized two-dimensional code area image with uniform size, direction and proportion.
  5. 5. The two-dimensional code image signal processing method for AGV navigation according to claim 1 wherein the decoding module performs grid division on the standardized two-dimensional code area image, extracts local features of each grid, inputs the local features into a transducer encoder for global context modeling, realizes context compensation through a self-attention mechanism, and finally outputs a coding sequence and confidence thereof through a full-connection layer and softmax.
  6. 6. The two-dimensional code image signal processing method for AGV navigation according to claim 1 is characterized in that steps S1 to S4 are periodically executed in the running process of the AGV, and when the output confidence is lower than a set threshold, repeated acquisition, cloud high-precision recovery or manual review flow is triggered.
  7. 7. A two-dimensional code image signal processing system for AGV navigation, which is characterized by comprising: The image acquisition module is used for acquiring an original image containing the two-dimensional code; The image restoration module is used for carrying out noise suppression, contrast reconstruction and local structure compensation on the original image and outputting an enhanced image; The two-dimensional code positioning and correcting module is used for detecting two-dimensional code corner points or boundaries, executing geometric correction and outputting a standardized two-dimensional code area image; The decoding module is used for extracting semantic features through the deep neural network, carrying out sequence prediction and outputting coding information and confidence coefficient; And the interface module is used for transmitting the coded information to the AGV control system and triggering a corresponding processing strategy according to the confidence coefficient.
  8. 8. The two-dimensional code image signal processing system for AGV navigation according to claim 7 wherein the image restoration module adopts a Residual U-Net network structure, the two-dimensional code positioning and correction module integrates a key point regression network and a Space Transformation Network (STN), the decoding module adopts a Transformer encoder structure, the system is deployed on an embedded processor, and a neural network model corresponding to each module is subjected to at least one of model quantization, structured pruning, knowledge distillation or hierarchical pushing to realize a processing delay of not more than 30ms.
  9. 9. An automatic guided vehicle, AGV, comprising: A vehicle body; the image acquisition device is arranged on the vehicle body and is used for acquiring images containing the two-dimensional code; A processor; A memory storing a computer program; Wherein the computer program, when executed by the processor, implements the two-dimensional code image signal processing method according to any one of claims 1 to 6.
  10. 10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the two-dimensional code image signal processing method according to any one of claims 1 to 6.

Description

Two-dimensional code image signal processing method and system for AGV navigation Technical Field The invention belongs to the technical field of image processing, and particularly relates to a two-dimensional code image signal processing method and system for AGV navigation. Background In the field of intelligent manufacturing and warehouse logistics, automated guided vehicles (AGVs/AMRs) widely employ ground-laid two-dimensional bar codes (e.g., QRCode, dataMatrix) for high-precision positioning and instruction acquisition. The conventional code reading flow generally comprises the steps of image acquisition, image preprocessing, binarization, locator detection, perspective correction, grid sampling, error correction decoding and the like. However, the prior art has significant shortcomings in several conditions in the industrial field: 1. The image restoration capability based on rules is limited, the traditional filtering, superdivision or deblurring method is effective on light blurring/low resolution, but when a two-dimensional code is covered by greasy dirt, dust is blocked or physical tearing to cause large-area loss of a module, missing textures are difficult to complement on a semantic level to cause error correction failure of a back end, the visual effect can be improved partly based on the operation of generating a model, but special constraint is not made on machine readability, and the generated result is possibly unfavorable for reading by a standard decoder. 2. The geometric positioning is strongly dependent on the locator, the traditional method relies on FINDER PATTERN or TIMING PATTERN for positioning, and when the locator is worn or lost, the positioning based on the geometric rule is invalid, and the whole decoding flow is interrupted. 3. The modularized processing causes information loss, namely if the enhancement, the positioning and the decoding are independent modules, the information lost by the front end is difficult to compensate by the back end, the existing depth enhancement method stays in visual restoration, and the end-to-end mapping which shares the restoration characteristics with the semantic decoding is lacking. 4. The engineering deployment real-time challenge is that many complex models depend on a high-performance GPU, navigation real-time and power consumption constraints are difficult to be directly met on an embedded AGV platform, and a lightweight and hierarchical reasoning strategy is required to be designed to meet the field deployment requirements. Therefore, a processing method and a system capable of stably and efficiently identifying two-dimension codes under complex working conditions are needed to improve the reliability and adaptability of AGV navigation. Disclosure of Invention The invention aims to provide a two-dimensional code image signal processing method and system for AGV navigation, which solve the existing problems. In order to solve the technical problems, the invention is realized by the following technical scheme: The invention relates to a two-dimensional code image signal processing method and a system for AGV navigation, comprising the following steps: S1, acquiring an original image containing a two-dimensional code through an image acquisition device arranged on an AGV body, wherein the image acquisition device is preferably a global shutter industrial camera, and the resolution is not lower than 640480 so as to ensure the definition and instantaneity of image acquisition. The method comprises the steps of S2, inputting an original image into an image restoration module, carrying out noise suppression, contrast reconstruction and local structure compensation processing on the original image, and outputting an enhanced image, wherein the image restoration module adopts a convolutional neural network structure comprising an encoder and a decoder, preferably a Residual U-Net, a multi-scale Residual block is added between the encoders to enhance high-frequency texture restoration capability, the network is restrained in a training process by using a composite loss function, the composite loss function comprises L1/L2 pixel loss, perception loss (based on characteristics of an intermediate layer of a VGG network of pre-training, such as VGG Relu3_3 and Relu _3), structure or edge constraint loss (gradient difference, such as Laplacian) and antagonism loss (PatchGAN), each loss weight proportion is preferably _pixel _perc _struct _adv=10:1:1:0.5, the training sample of the image restoration module can be synthesized in a mode according to actual scene, the training sample of the image restoration module comprises a transparent oil layer (frequency range of Per-3:1:1:1:0.5) and a random noise correction model, and a noise correction model is combined according to the actual noise loss, and the noise correction model is combined to the noise loss, and the noise loss is improved by using a random model, and the noise correction model is suitable