Abstract: The path edge recognition strategy of the crane metal structure climbing robot is divided into three steps. Firstly, image pre-processing which means using the improved over-color operator for grayscale. Secondly, the gradient threshold is determined by the method based on the optimal classification line of the support vector machine, in addition, main direction angle constraint is added to improve line segment detector (LSD) algorithm, and obtain the straight line detection image for clustering. Thirdly, the clustering data set is constructed by the feature extraction of straight line segments. Based on the dynamism of the data set feature of, the improved AP clustering algorithm is established by combining the prior information based discriminant model with the affinity propagation (AP) clustering algorithm to cluster the line segments and screen out the line segments constituting the edge of the path, and obtain the final path edge line by fitting. The experimental results show that compared with the traditional AP clustering and other clustering algorithms, the improved AP clustering algorithm has the highest screening accuracy for path edge lines. The recognition success rate of path edge recognition strategy based on improved LSD and AP clustering is 96% which meets the accuracy and real-time requirements.

Key words: edge recognition, over-color operator, LSD, SVM, feature extraction, AP clustering