A traffic metal sign surface defect detection system and method

Abstract

This invention discloses a system and method for detecting surface defects on traffic metal signs. One method includes the following steps: First, acquiring multiple two-dimensional images under different lighting conditions, and simultaneously obtaining three-dimensional point cloud data of the sign surface; second, performing pixel-level registration between the two-dimensional images and the three-dimensional depth map to generate 4-channel fused data containing RGB and height information; third, constructing a digital twin virtual template to generate a virtual standard image and a virtual standard height map; fourth, using a dual-branch deep learning network to extract two-dimensional semantic features and three-dimensional geometric features, and introducing a deformable Transformer encoder to compare the measured features with the virtual template features layer by layer to generate a residual map; and finally, after classification and identification, outputting the defect type, location, and size. This invention effectively solves the problems of metal reflection interference, lack of three-dimensional geometric defect detection, difficulty in verifying customized sign content, and low recognition rate of minor defects, achieving high-precision, full-dimensional automated detection.

Images