slamlearn
-
Multi-view 3D reconstruction based on SFM (Structure from Motion)。The project utilizes SIFT for feature point detection, initializes the camera poses by solving the essential matrix, calculates the poses for each subsequent image using EPNP, and ultimately triangulates to reconstruct a sparse point cloud of landmark features. (I also attempted dense reconstruction using epipolar line search and depth filtering but did not include it in the final program due to computational time and memory constraints.)
-
Monocular SLAM (Simultaneous Localization and Mapping) front-end odometer. Using ORB-FAST feature point detection, in addition to calculating the pose and mapping, incorporating Bundle Adjustment for pose optimization of each frame image and multi-frame co-visible landmark point optimization. Adding feature point removal to ensure unique co-visible points. Additionally, performing plane detection in the point cloud map.
--- singlereconstruction 双视图重建
--- multiviewreconstruction sfm多视图三维重建
slam
k.txt 相机内参
rgb.txt 视频帧图片地址
--- 平面检测
