Obstacle contour extraction method based on improved Grabcut algorithm

Abstract

Abstract Obstacle distance measurement has an important role in the design of mobile robots, which allows the robot to timely complete the movement steering to avoid damage caused by crushing. The accurate segmentation and edge detection of the target obstacle in the image is the premise of completing the measurement. In this paper, based on the classification and location of specific obstacles, a method of obstacle segmentation based on improved Grabcut algorithm is proposed, and the edge contour is extracted by Canny operator. The method classifies the image’s background pixels using a Laplacian Mixture Model and modifies the covariance matrix to change the probability density function of that model, thereby reducing the misclassification probability of the pixels. In addition, the method crops the input image in advance to reduce the running time of the algorithm. This paper also introduces a way to initialize the image’s Trimap, considering the visual angle of the sweeper. The above four improvements enable the method to accurately segment obstacle targets and extract contour curves. The accuracy of the method and the good real-time performance on the embedded platform are verified by comparison experiments.