In order to detect cotton position on a plant, an infrared distance measurement device and the computer image processing technology, combined with the growth characteristics of cotton, were used to measure the distance of single cotton. The infrared measurement model was established in the distance of 0.7m to 1.6m. The distance image of plant surface point cloud was obtained with the measurement accuracy of 0.008 5 m. The Gaussian filter size was designed with the standard deviation of 0.65. 1D Gaussian filter was used to smooth the plant image and 2D Gaussian filter was used to obtain the gradient image. With the strong threshold of 0.2m and the weak threshold of 0.08m, the closed loop edge was constructed based on Canny edge detection operator. The image of cotton area was extracted by morphological operations, and the cotton sticks background was removed. Peak filter was used to weaken the adhesion and overlap of the cotton image. The cotton distance image was segmented based on watershed algorithm to extract the single cotton image and obtain the distance. The results show that the image recognition rate of single cotton on the plant is 91.3% and the correlation coefficient between the infrared and artificial measurement results is 0.992 2, which provides parameters for cotton picker robot to plan motor path.