Abstract: In order to accurately and in real-time obtain weld seam information for intelligent welding, a laser stripe feature extraction framework based on improved Steger algorithm and Hough detection was proposed. First, the collected laser stripes were preprocessed for the weld image, utilizing a combination of median filtering and bilateral filtering for denoising, employing Otsu's thresholding to extract the laser stripes, and removing small noise points through morphological operations. Then,to address the low extraction efficiency of traditional Steger algorithm, the improved Steger algorithm was used to set thresholds focusing on the region of interest（ROI） to enhance algorithm efficiency, initially extracting the centerline of the light stripe where breakpoints existing. Finally, all straight lines in the light stripe centerline were detected through Hough transformation, and weld seam feature corner points were filtered from the intersection points of different lines by setting specific areas where feature points were located. Experimental results indicate that the peak signal-to-noise ratio after denoising is greater than 30 dB. The average extraction time of the improved Steger algorithm is 0.070093 s, which is more than twice as fast as the traditional Steger algorithm, meeting the real-time requirements. The coordinate error of the extracted weld seam feature corner points is at the millimeter level, satisfying the welding requirements.