A self-supervised monocular depth estimation method based on visual slam algorithm
By employing a self-supervised monocular depth estimation method, and utilizing depth estimation and pose estimation networks, combined with multi-scale feature fusion and displacement operations, the problems of difficult image spatial information acquisition and slow computation speed in monocular visual SLAM algorithms for mobile robots are solved, achieving high-precision depth estimation with low computational cost.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH BEIJING
- Filing Date
- 2026-05-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing monocular vision SLAM algorithms for mobile robots suffer from problems such as difficulty in acquiring spatial information from images, inaccurate localization in dynamic environments, slow computation speed, low prediction accuracy, and large computational load in depth estimation algorithms.
A self-supervised monocular depth estimation method is adopted, and an algorithm framework is constructed using a depth estimation network and a pose estimation network. The image is reconstructed by using the depth map and the pose transformation matrix. Multi-scale feature fusion and translation operations are combined. The network model is trained and optimized using the KITTI, Make3D and AirSim datasets.
It improves the prediction accuracy and computation speed of depth estimation, reduces the number of model parameters, and is suitable for integration into visual SLAM systems.
Smart Images
Figure CN122415705A_ABST