rpng/open_vins

Features

• State Estimation Libraries - Uses Extended Kalman Filters to track a robot's velocity and pose while correcting for sensor drift.
• Visual-Inertial Odometry Frameworks - Fuses inertial data and visual feature tracks using an Extended Kalman Filter to determine device position and orientation.
• Computer Vision Features - Detects salient image points using grid-based patterns for uniform visual feature extraction.
• Visual Feature Tracking - Follows point features across multiple camera configurations using KLT or descriptor-based methods.
• Loop Closure Detection - Detects revisited locations in a background process to correct global drift via loop closure.
• Visual Odometry Systems - Implements a full visual-inertial odometry system using an EKF to estimate device state.
• Stereo Triangulation - Estimates 3D positions of visual landmarks by analyzing their projections across multiple camera views.
• Sensor Calibration - Determines intrinsic and extrinsic parameters of cameras and IMUs for accurate spatial and temporal synchronization.
• EKF-Based State Estimators - Implements an Extended Kalman Filter to fuse inertial measurements and visual features for real-time tracking.
• Visual SLAM Implementations - Implements a full visual SLAM system to build environmental maps and correct global positioning errors.
• Sparse Feature Mapping - Tracks sparse visual features across multiple representations to maintain a map and improve long-term state estimation.
• Inertial Preintegration - Processes high-frequency inertial data into single constraints between image frames to reduce computational load.
• Inertial Preintegrations - Processes high-frequency inertial data into single constraints to reduce the computational load of the state filter.
• State Covariance Modeling - Tracks error states and covariance indices to update manifold representations during filter corrections.
• Kalman Filter Localization - Implements an Extended Kalman Filter for real-time robotic state estimation by fusing inertial and visual data.
• Uniform Feature Distribution - Uses a spatial grid to ensure uniform visual coverage and prevent feature clustering during detection.
• Manifold Pose Interpolations - Performs B-Spline interpolation over the SE(3) manifold to produce smooth pose estimates between samples.
• Online Sensor Calibrations - Estimates camera and inertial parameters during operation to correct temporal offsets in real time.
• Geometric Distortion Correction - Corrects optical imperfections and geometric distortions using pinhole, fisheye, and radial-tangential models.
• Dynamic State Initializations - Recovers initial velocity and gravity by solving a linear system while the platform is moving.
• Stationary State Initializations - Determines initial orientation by aligning the state with the gravity frame while the platform remains still.
• Pinhole Camera Models - Represents camera geometry using radial-tangential and fisheye equidistant projections to simulate lens behavior.
• Manifold Interpolations - Uses B-Spline curves over the special Euclidean group to produce smooth trajectories between state estimates.
• SE(3) Manifold Interpolations - Uses B-Spline curves over the special Euclidean group to produce smooth trajectory paths between discrete poses.
• Synthetic Sensor Simulation - Generates synthetic sensor measurements and trajectories to validate navigation algorithms in controlled virtual environments.
• Ground Truth Trajectory Generation - Combines motion capture data with inertial information to create high-accuracy reference paths for evaluation.
• State Estimation Performance Metrics - Includes a suite of metrics and scripts to quantify the precision and accuracy of state estimation results.
• Trajectory Accuracy Evaluation - Compares estimated movement paths against ground truth data using absolute and relative pose error metrics.
• Trajectory Alignment Tools - Transforms and aligns two separate trajectories to enable a direct comparison of estimated paths.
• Trajectory Error Quantification - Calculates absolute trajectory error and relative pose error across varying segment lengths.

Star history