vslam

simple Visual SLAM (VSLAM) problem

objective: replace an inertial navigation with vslam algorithm.

1 simulate a circular motion of a robot in a room. 2 update the trajectory from a perfect sensor. 3 assume a perfect control system.

Estimate the camera trajectory and reconstruct a 3D map of a scene using a monocular camera.

Given a sequence of images captured by a monocular (single lens) camera, the goal is to estimate the camera's trajectory (i.e., its position and orientation) over time and reconstruct a 3D map of the observed scene.

Solution Steps:

Feature Extraction: Extract key features from each image, such as keypoints or corners. Common feature detection algorithms include ORB, SIFT, or SURF.

Feature Matching: Match corresponding features between consecutive image pairs. This can be done using feature matching algorithms like the nearest neighbor or brute-force matching.

Pose Estimation: Estimate the camera's pose (position and orientation) relative to the previously observed frame. This can be done using techniques like the 5-point or 8-point algorithm, which solve the perspective-n-point (PnP) problem.

Keyframe Selection: Identify keyframes from the sequence of images. Keyframes are frames that provide significant changes in the camera's pose or scene structure.

Map Initialization: Initialize a sparse 3D map by triangulating matched features from the keyframes.

Bundle Adjustment: Refine the camera poses and 3D map by jointly optimizing them using bundle adjustment. Bundle adjustment minimizes the reprojection error between the observed image points and the projected 3D map points.

Loop Closure: Detect and close loops in the trajectory to reduce drift and improve accuracy. Loop closure techniques identify previously visited locations in the scene by matching features between distant frames.

Map Optimization: Further optimize the map and camera poses by performing global optimization, such as pose graph optimization or loop closure optimization.

asked here for quickstart: https://github.com/changh95/visual-slam-roadmap

https://www.mathworks.com/help/vision/ug/monocular-visual-simultaneous-localization-and-mapping.html Monocular Visual Simultaneous Localization and Mapping Glossary Overview of ORB-SLAM Download and Explore the Input Image Sequence Map Initialization Store Initial Key Frames and Map Points Initialize Place Recognition Database Refine and Visualize the Initial Reconstruction Tracking Local Mapping Loop Closure Compare with the Ground Truth Supporting Functions