Adaptive Laser Stripe Extraction Method Based on Semantic Segmentation for Industrial Measurement

The 3-D measurement based on line structured light sensors plays an important role in industrial measurement, but the harsh environment (lighting, physical properties, noise) during industrial measurement seriously affects the accuracy of laser stripe extraction, resulting in inaccurate measurement. This paper introduces an adaptive ROI laser stripe extraction algorithm that combines deep learning and traditional methods to address the challenge of stripe extraction in industrial measurement. Aiming at the problem of small proportion of laser stripes in images, a network structure based on ResNet encoder and MultiAttention FusionNet based on FCN Net decoder is proposed, which combines multi-scale feature fusion and channel attention mechanism to improve the robustness and extraction accuracy of stripes. Finally, the final extraction result is obtained by filtering the stripe length and direction vector. The experimental verification was carried out using wheelset measurement under lathe re-profiling conditions as an example, and the results showed that the extracted stripes met the measurement requirements, indicating the advantages of the proposed method.