Abstract: To avoid sample collection,manual labelling and local extremum in the target detection method based on support vector machine or neural network,detection on geometrical feature of the common industrial circular valve handles was transformed into detection on two sub-features,circle and line segments in this paper.Namely,Hough transform is employed to detect circles and lines on an image captured by robot firstly.Only three lines are remained which best correspond to geometrical feature of circular valve handles in the designed algorithms.Then,combing the detected circle,lines and the angles features,the detected circle will be judged whether it is a handle or not.And the positions of inserting robot’s three fingers to operate the handle are obtained.The experiment results indicate that there’s no rigorous requirement for shooting angle and brightness,the circular valve handles can be detected effectively,and the position of inserting robot’s fingers can be calculated accurately by the method.The correct detection and location ratio is 90.7%.

Key words: Circular valve handle,Detection and location,Hough transform,Geometrical feature,Radial line