Structured light residual channel attention network for super-resolution enhancement of 3D measurement images

The precision of structured light 3D measurements is often limited by resolution degradation during image acquisition and processing, leading to blurred edge features. Interpolation-based upsampling methods are insufficient, and improving camera resolution in large field of view systems is costly. We propose the structured light residual channel attention network (SLRCAN), a super-resolution network tailored to the characteristics of structured light imagery. The network integrates residual architectures with a designed channel attention mechanism. We apply SLRCAN to measurement for continuous casting billet in an industrial environment. Dedicated datasets tailored to this scenario are used to train and evaluate the model, offering insights applicable to other scenarios. To address the limitations of conventional image quality metrics in evaluating geometric fidelity, we propose the assessment of smoothness (AS), a task-specific metric designed to quantify edge continuity. Experimental results at scales 2× and 4× show that SLRCAN outperforms existing methods, achieving state-of-the-art performance. Additionally, laser measurement experiments on standardized small objects validate its applicability, emphasizing its potential for precise measurement in structured light.