Pose detection of parallel robot based on improved Hough-K-means and SURF algorithms

Abstract

It is difficult to detect end pose of auto electrocoating conveying parallel robot based on binocular vision because of less features, noise and blurred edges in image. So a pose detection method is presented. Firstly, according to distance and similarity between image points, the bilateral filtering is adopted to reduce noise. Secondly, for extracting straight lines on blurred edges and feature points in lines accurately, a Hough-K-means algorithm using K-means clustering analysis in parameter space after Hough transform is proposed. Thirdly, the discrete Gaussian–Hermite moment is used as feature descriptor to solve the low accuracy problem of principal directions of feature points described by speeded-up robust features (SURF) descriptor. The extracted feature points are selected according to the similarity metric between constructed feature vectors. Finally, the vision model is developed to calculate the three-dimensional pose parameters of parallel robot based on obtained point pairs. The experiment results show that, compared with the SURF algorithm, the average matching time decreases by 21.58%, the average deviations of detected poses in x, z and β decrease by 1.149 mm, 0.646 mm and 1.164° respectively by using the proposed method. The speed and accuracy of pose detection of parallel robot can be improved.

Keywords

• pattern clustering
• feature extraction
• pose estimation
• edge detection
• hough transforms
• robot vision
• image restoration
• image denoising
• image filtering
• parallel robot
• surf algorithm
• auto electrocoating
• binocular vision
• blurred edges
• pose detection method
• hough transform
• discrete gaussian–hermite moment
• extracted feature points
• hough-k-means algorithm
• k-means clustering analysis
• speeded-up robust features descriptor