ing at the problem of low accuracy of blood vessel segmentation caused by the small and blurred outline of fundus retinal blood vessels, a retinal blood vessel segmentation method using wavelet transform to fuse the contour feature and detailed feature of the blood vessel under a multi-scale framework is proposed. The contrast between the blood vessel and the background is enhanced by preprocessing, the contour feature and detail feature of the blood vessel are extracted in a multi-scale framework, and image post-processing is performed. The wavelet transform is used to fuse the two feature images, the maximum value of the corresponding pixel of each scale is calculated to obtain the blood vessel detection image, and finally the Otsu method is used for segmentation. Through the test experiment of the DRIVE data set, the average accuracy, sensitivity, and specificity are 0.9582, 0.7086, and 0.9806, respectively. The method in this paper can accurately segment the contour of the blood vessel while retaining more branches of small blood vessels, and the accuracy is high.
.ing at the problem of poor localization accuracy or failure of traditional visual odometry in low-texture scenes, we propose a method of RGB-D visual odometry that combines point and line features. This method unites the unique characteristics of points and lines, and combines the photometric residual of point features and the local gradient fitness error of line features to construct a joint error function that is robust to low-texture scenes. The Gauss-Newton iteration method is used to perform nonlinear iterative optimization of the joint error function to obtain the accurate pose of each frame. The proposed method is evaluated on the public real-world RGB-D dataset and synthetic benchmark dataset. Experimental results show that, compared with other state-of-the-art algorithms, the proposed algorithm has better accuracy and robustness.
.