CyperStereo four-camera module

Demo and values

Four-camera inertial module from Shenzhen Tujian Intelligent Technology (图间智能). Four global-shutter RGB cameras sit on separate 25 × 25 mm boards, cabled to one controller with a Bosch BMI088 IMU; two cameras have 160° lenses and two have 200° lenses, and the baseline is set by how the boards are mounted. One recording, solved as four cameras + IMU.

The 3D scene MVIS recovered from this recording, first 20 s, each camera in turn.

Reprojection median0.19 px

cam0–cam1 baseline66.6 mm

cam0 td+0.80 ms

cam0 lever arm59.7 mm

Reprojection is how far, in pixels, the detected corners lie from where the solve puts them. td is the camera–IMU time offset, and the lever arm is the camera–IMU position offset.

No independent reference, so these values show what MVIS solved, not how close they are to the truth.

Calibration process

The path of the rig around the board, as MVIS solved it. To explore it:

• Press Play or drag the bar to follow the path.
• Use the view buttons to change the view.
• Pick a camera card to put that camera on the frustum, the pyramid that shows its view.

Each card draws the board from the solved pose, with the corners MVIS detected. We publish no camera images.

Solved rig

The solved rig in 3D, to scale. Tap a sensor to see its values. In full screen you can rotate and zoom. Grid squares are 20 mm.

3DFrontTopSide+−

To scale, from above and the front

From above

From the front

Camera	Lever x, y, z (mm)	Distance (mm)	log(R) (rad)	td (ms)	Field of view
cam0	+52.9, +13.4, +24.3	59.7	[−0.016, −1.583, −2.710]	+0.80	131° × 110°
cam1	−13.7, +12.7, +24.6	30.9	[−0.002, −1.579, −2.711]	+0.80	131° × 110°
cam2	+85.5, +2.1, +15.6	87.0	[+0.515, +1.958, +1.908]	+0.79	170° × 137°
cam3	−47.1, +0.1, +16.0	49.7	[+0.523, −1.937, −1.918]	+0.80	~171° × 137°

Solved by MVIS. Lever arms (camera–IMU position offsets) are in the IMU frame. Rotation is log(R), camera in the IMU frame.

Specifications

Published by the maker
Maker	Shenzhen Tujian Intelligent Technology (深圳市图间智能科技有限公司)
Image sensor	SmartSens SC136HGS, RGB, global shutter, back-illuminated, 4 × 4 µm pixels
Wide lenses (cam2, cam3)	1.67 mm, F2.0, M12; 200° diagonal, 174° × 139°
Narrow lenses (cam0, cam1)	1.79 mm, F2.0, M12; 160° diagonal, 133° × 112°
Synchronization	Camera to camera 10 µs, camera to IMU 100 µs on the sheet; identical camera stamps in this recording
Link and power	USB 3.0, UVC; 1.5–2 W at 5 V
Measured in these recordings
Cameras	4 × 1280×1024; 30 fps on the sheet, 15 Hz in this recording
IMU	Bosch BMI088, 200 Hz
Calibrated field of view	cam0 131° × 110° · cam1 131° × 110° · cam2 170° × 137° · cam3 ~171° × 137°

Sales contact Provided by the maker

Email

support@tujian.com

Dataset

Recordings	1 recording, 74.6 s, hand-held. About 1,120 frames per camera, 14,920 IMU samples
Format	ROS1 bag
Board	AprilGrid 6×6, tag 55 mm, spacing 0.3
Licence	None
Hosted	Not hosted

AprilGrid 6 × 6, 412 × 412 mm. Standard pattern, to scale

Detailed results

Lens Calibrated field of view

Calibrated field of view, horizontal × vertical. cam0: 131° × 110°, corners reach 77° from the axis · cam1: 131° × 110°, corners reach 77° from the axis · cam2: 170° × 137°, corners reach 86° from the axis · cam3: ~171° × 137°, corners reach 87° from the axis

cam0 cam1 cam2 cam3

We read these from the calibrated lens fit at the image edges, through the principal point (where the optical axis meets the image). ~ marks an axis past where the corners reached, so the fit is extrapolated there. Past the fit means an edge lies beyond where the fit holds, so we give no value.

IMU Gyro–accel time offset t_ga −4.36 ms

Each bar shows a scale as its deviation from 1. IMU model: calyx. t_ga, the time offset between the gyroscope and accelerometer channels: −4.36 ± 0.12 ms.

Gyro scale x−0.14 %

Gyro scale y−0.25 %

Gyro scale z−0.06 %

Accel scale x+1.21 %

Accel scale y+0.91 %

Accel scale z+1.22 %

−8 %0+8 %

Timing cam0 td +0.80 ms

The time offsets, to scale. td is the camera–IMU time offset, with t_imu = t_cam + td.

Full results Every solved parameter

Rotation is log(R) in rad, camera in the IMU frame.

Parameter	MVIS
cam0
fx px	449.26
fy px	449.82
cx px	657.09
cy px	526.38
k1	0.23203
k2	−0.01098
k3	−0.02659
k4	0.00821
Lever x mm	+52.88
Lever y mm	+13.36
Lever z mm	+24.26
log(R) rad	[−0.0159, −1.5832, −2.7104]
td ms	+0.800
Reprojection px	0.621
cam1
fx px	450.41
fy px	450.48
cx px	636.48
cy px	518.24
k1	0.22978
k2	−0.00836
k3	−0.02944
k4	0.00910
Lever x mm	−13.68
Lever y mm	+12.74
Lever z mm	+24.63
log(R) rad	[−0.0015, −1.5789, −2.7113]
td ms	+0.804
Reprojection px	0.643
cam2
fx px	417.88
fy px	417.70
cx px	641.28
cy px	507.31
k1	0.01399
k2	0.01291
k3	−0.00923
k4	0.00154
Lever x mm	+85.53
Lever y mm	+2.13
Lever z mm	+15.56
log(R) rad	[+0.5153, +1.9580, +1.9083]
td ms	+0.785
Reprojection px	0.398
cam3
fx px	418.82
fy px	419.14
cx px	624.83
cy px	510.81
k1	0.01040
k2	0.01466
k3	−0.00990
k4	0.00161
Lever x mm	−47.06
Lever y mm	+0.11
Lever z mm	+15.97
log(R) rad	[+0.5227, −1.9369, −1.9176]
td ms	+0.803
Reprojection px	0.416
IMU
t_ga ms	−4.36 ± 0.12

Limits

• No independent reference, so these values show what MVIS solved, not how close they are to the truth.
• The board never reached the outer side of the two 200° images (about two-thirds of each image is covered), so the lens model there is extrapolated.
• The IMU noise set for the solve is wider than the residuals need (IMU χ²/dof 0.03); a solve with tighter noise is still to do.
• The tag size is taken from the target file sent with the data and isn't confirmed on the printed board. It scales every baseline and lever arm.

How it was run

• All four cameras and the IMU together. Camera intrinsics (Kannala–Brandt kb4) from four single-camera solves of the same recording, then held fixed; IMU scale and misalignment, gyro-to-accel offset t_ga and camera time offsets estimated. IMU noise from the BMI088 datasheet, about 2.5× wider. Solved 2026-11-11.