Monocular vision inertial pose estimation method based on deep learning point and line features

The invention discloses a monocular vision inertial pose estimation method based on deep learning point and line features. The monocular vision inertial pose estimation method comprises the following steps: step 1, parameter calibration and IMU data pre-integration; step 2, obtaining feature points, feature lines and feature line matching pairs by using an LPSD network and a deep learning DG-Line network; step 3, obtaining feature point matching point pairs and IMU pre-integration values in adjacent frames, and predicting to obtain an initial pose Tinit of the current frame of two-dimensional image; and step 4, minimizing re-projection errors of the feature points and the feature lines, performing BA optimization in combination with IMU pre-integration errors, and determining pose estimation Tlast after smoothing of each frame of image. The monocular vision inertia estimation method based on the deep learning point and line features has the advantages of being high in precision, high in robustness, capable of