Abstract: A method based on image processing was proposed to realize an online detection of rubber hose surface defects. An image shear method was taken to select hose area. The median filter and Canny edge detection were used to extract graph. To avoid the boundary influence in separating the defects from the hose, a morphological expansion method was used for the defect area separation. According to the character of defects, the defects were classified. Three CMOS cameras within the same plane were layout around the hose axis, interval of 120°. Coaxial light source was used to improve the brightness of hose surface. The mechanical platform of the detection system could adjust the cameras in three directions to ensure the hose in the center position. This system was tested in the hose workshop of Codan-Lingyun Automotive Rubber Hose Co Ltd. The experiment results show that the correct rate of detection is over 96 percent and it can detect the hose surface defects automatically and alarm timely.