Aiming at the recognition problem of robot adaptive grinding curved surface welding area， an algorithm based on linear structured light for automatic recognition of starting and ending points of grinding robot and a depth image enhancement operator are proposed in this work. The enhancement operator takes the sum of the absolute value of the intensity difference between the center pixel and the pixels in the 8 neighboring regions as the value of the center pixel to enhance the visualization features of the depth image and reveal the texture features of the polished area. First， the point cloud data is filtered and the hole filling is processed； second， the standard deviation of each scan line point cloud in the height direction is calculated； finally， the obtained features are identified to find the position with large feature change in a certain range， so as to extract the area that needs grinding. Experimental results show that the enhancement operator has excellent effect on depth image enhancement， and the average recognition accuracy of the algorithm for the start and end positions is less than 1 mm， which can achieve pixel level accuracy， and has strong robustness and is insensitive to noise. The effectiveness and feasibility of the algorithm are also verified in the field test.