画像特徴点に基づくSLAMの精度向上を目的としたニューラルネットワークによる不要特徴点判別

抄録

In this research, we propose the method for improving the accuracy of feature-based SLAM. Unnecessary feature points that SLAM uses for mapping are removed by discrimination by a neural network system. Our neural network system was trained by the coordinates and pixel values of the feature points as feature vectors and outputs a feature point in a surgical image is a biological part or a surgical instrument part. Feature points of surgical instruments may cause noise in SLAM, and the elimination of these feature points is expected to improve the accuracy of SLAM. The experimental results showed our neural network system was able to discriminate the feature points with high accuracy in the validation images, which is the same scene as the training images.