leokoppel / msckf

Multi-State Constraint Kalman Filter with ROS interface.

MSCKF

Experimental implementation of the Multi-State Constraint Kalman Filter with ROS interface.

Based on the paper by Mourikis and Roumeliotis (PDF).

Also includes a feature tracking node using OpenCV's SIFT implementation.

This is a work in progress. In particular, The API is a bit non-standard and is subject to change.

ROS nodes

feature_tracking_node

Subscribed topics

• camera/image_rect (sensor_msgs/Image) - Undistorted image stream.
• camera/camera_info (sensor_msgs/CameraInfo) - The corresponding camera metadata.

Published topics

• output_image (sensor_msgs/Image) - Image with arrows drawn on it, showing matches.
• features (msckf/ImageFeatures) - custom message describing feature positions

msckf_node

Subscribed topics

• imu_vel (geometry_msgs/TwistStamped) - IMU measurement of linear and angular velocity.
• odom (nav_msgs/Odometry) - Position of robot, used to initialize the filter
• camera/image_rect (sensor_msgs/Image)- the same image stream used by feature_tracking_node. Currently, only the header information is used.
• features (msckf/ImageFeatures) - the feature information published by feature_tracking_node

Published topics

• odom_combined (geometry_msgs/PoseWithCovarianceStamped) - Estimated robot pose