Abstract: Image processing algorithms of welding robot structure light detection were studied.Compared with the traditional image processing algorithm,some improved methods were proposed and tested,In filter of image preprocessing,a method combing region segmentation with multi-scale detection was adopted,and could overcome the disadvantage of contour easy breaking with single dimension LOG filter and minish the region to be processed,and easy to improve the processing speed.In contour pick-up of post-processing,Genetic Algorithm as a global optimization algorithm,was adopted to realize template matching.By grey coding method,fitness function,selection operators,crossover operators,mutation operators and so on,finally realize template matching and coordination pick-up the experimental results show that these improved methods can improve the precision and adaptability of welding robot structure light detection.