zzsssky/time_autosync

Automatically syncs a camera to a rigidly attached IMUs time frame

auto_timesync

A ros node with two functions

• Making camera arrival timestamps a hair less garbage.
• Estimates the absolute time offset between the camera and a ridgidly attached IMU

Under the skin the system is just a central difference kalman filter.

• It assumes a constant camera framerate and uses this to get rid of some of the jitter from arrival times
• Absolute offsets are calculated by aligning the IMU's measured angular velocity to that perceived by a simple Lucas-Kanade tracker running on the camera images. Obviously this will only work on systems that undergo significant movement.

This filter is pretty new and only tested on a couple of datasets I had lying around, so expect some issues when you first run it.

Parameters

---

General Parameters

Parameter	Description	Default
stamp_on_arrival	If true it will throw out the image timestamps and replace them with arrival times.	false
max_imu_data_age_s	How long in seconds to keep IMU data, the camera timing offset needs to be in this range.	2.0
delay_by_n_frames	Delays putting camera images into the filter by this many frames. Used to ensure there are plenty of IMU messages ahead of the frame to get IMU messages from. Note this does not cause any delay in the output of timesynced images.	3
focal_length	The focal length of the camera. A rough guess is fine and the system seems to run ok on distorted images	460
use_pointcloud	Uses pointcloud as input rather then an image. In this mode ICP matiching is used to estimate the rotation	false
calc_offset	If true the absolute time offset between the camera and an IMU is calculated. If false only some dejittering is performed (this saves a lot of CPU and most of the above parameters don't apply).	true

Filter Parameters

Parameter	Description	Default
verbose	Prints a large amount of information about the filters internal state	false
mah_threshold	Mahalanobis distance theshold for accepting measurement updates	10.0
inital_delta_t	Inital estimate in seconds for rate at which images are received. Set to 1/framerate. Note if you set it too small by a factor of 2 or more the automatic frame drop detection will cause weird effects	0.05
inital_offset	Inital estimate in seconds for the offset between the camera and IMU timestamps	0.0
inital_timestamp_sd	Inital standard deviation of systems timestamp in seconds (which is initalized to the first images timestamp)	0.1
inital_delta_t_sd	Inital standard deviation of the time between frames	0.1
inital_offset_sd	Inital standard deviation of the offset between the camera and IMU	0.1
timestamp_sd	Standard deviation of the received image timestamps	0.02
angular_velocity_sd	Standard deviation of the angular velocity estimated from the camera images	0.03
delta_t_sd	Standard deviation of the process noise for the time between frames	0.0001
offset_sd	Standard deviation of the process noise for the time offset between the IMU and camera	0.0001