Subpixel laser centerline extraction via channel separation and adaptive Gaussian sliding window

Abstract The structured light scanning binocular stereo vision technique enhances surface texture information by projecting structured light, enabling high-precision 3D data acquisition. Accurate and efficient extraction of the laser stripe centerline at the subpixel level is crucial for high-quality 3D measurements. This study proposes an adaptive Gaussian sliding window method based on channel separation for precise centerline extraction. The original image is first segmented using the Hue, Saturation, Value color space to eliminate background interference. The RGB channels are separated, and the brightest channel is selected for centerline extraction. An adaptive Gaussian sliding window filtering algorithm is then applied, enhancing the stripe signal and approximating a Gaussian distribution. Finally, the gray-level centroid method is used for subpixel centerline localization. Experimental results show an average processing time of 0.0958 s per frame, demonstrating high efficiency. The proposed method reduces centerline extraction errors by 1.0384 px and 0.2531 px, and 3D measurement errors by 0.5160 mm and 0.1108 mm, compared to conventional methods. By integrating the ‘maximum brightness’ strategy with adaptive Gaussian filtering, this method addresses the limitations of traditional grayscale centroid approaches, improving precision while maintaining computational efficiency.