Research on Position and Posture Estimation of Rotated Workpiece Based on Image Recognition

Abstract

In order to improve production efficiency, industrial robots are gradually applied to the manufacturing industry. It is necessary for industrial robots to obtain the position information and posture information of workpieces placed randomly to complete the tasks such as grasping, sorting and assembling. However, the algorithms based on traditional machine vision are sensitive to a series of factors such as illumination intensity and workpiece surface smoothness, which makes it poorly adaptable. To enhance the robustness and accuracy of robot operation, a position and posture estimation algorithm of the rotated workpiece based on image recognition is proposed. The proposed algorithm adopts the TextBoxes++ algorithm to predict the four vertices of the workpiece. The position information and posture information could be transformed from the smallest rotated rectangle containing the four vertices. The proposed algorithm has been evaluated on the custom dataset including six kinds of workpieces. The experimental results show that the proposed method can control the average position error and the peak position error within 0.11 mm and 0.5 mm, while the average posture error and the maximum posture error are no more than 0.40° and 2.92°, respectively. The proposed method could detect workpieces of uncertain positions and different postures, which is beneficial for the robots to adapt to the external environment automatically.