Multiple Device Sync Issues

[wes-b]

Multiple Device Synchronization has a couple issues that prevent it from working reliably.

Check out Multi-camera sync documentation

Issues that the user must be aware of and work around

• Color to Depth Timing may have an offset if setting manual exposure repeatedly. #471 Color to Depth Timing may have an offset if setting manual exposure repeatedly.
• Document how to avoid depth camera interference when using multiple camera's #436 Document how to avoid depth camera interference when using multiple camera's.
• Understanding of the timing of Kinect #868 Understanding of the timing of Kinect.
• libusb and libuvc errors when using multiple devices on Linux #485 libusb and libuvc errors when using multiple devices on Linux

**Issues that are actively being worked on / fixed

• k4a_image_get_exposure_usec returns different value than exposure set with k4a_device_set_color_control #362 Setting exposure does not result in the expected exposure value. Fixed in SDK version 1.2.0.
• Powerline setting is changed by firmware after the camera's are started #448 Power Line setting is being changed by firmware after being set by SDK. Fixed with FW 1.6.104075014.
• Subordinate timing is off by ~100us #529 Subordinate timing is off by ~100us. Fixed with FW 1.6.104075014.
• k4a_image_get_exposure_usec() doesn't match value set by k4a_device_set_color_control() #784 k4a_image_get_exposure_usec() doesn't match value set by k4a_device_set_color_control(). Fixed with FW 1.6.104075014.
• Cannot restart multiple synced cameras if RGB stream is OFF in subordinate #818 Cannot restart multiple synced cameras if RGB stream is OFF in subordinate Fixed with FW 1.6.110079014.
• Multiple Kinect devices connected to min requirements PC do not work properly #971 Multiple Kinect devices connected to min requirements PC do not work properly Fixed with FW 1.6.110079014.

[YuXinFan]

I meet some problem with sync. When connects external sync only one master which is not a ring:

1. In some running case, the timestamp of master capture is 100,000 usec forward. In this case, the other devices are aligned .
2. In some running case, the timestamp of some devices is about 30,000 usec lag, but they sometimes can also have the same timestamp in a loop round.
   I test many times. And I want to get stable sync. Why this problem occurs?

[wes-b]

I have not heard of the timestamp being that far off, but I am able to run 3 devices at my desk at 30FPS (no image processing), where I have seen it off by 4-5 ms.

The problem is from #362, you need to be setting a manual exposure value. But what you set / get via the API is different then what is actually used and returned via k4a_image_get_exposure_usec. We also have a firmware fix coming.

For now try building a version of the SDK with the fix for #362 (until 1.2.x is released). The timing won't be perfect yet, but is should get you within a few milliseconds or less.

[initialneil]

@wes-b I'm having the same sync issue. I'm connecting 3 devices to one PC, wire-synced one after one. One of the device is master and the others are subordinates.
When I use system timestamp from color image, the results look synced, yet the device timestamp are far from close to each other.
See the screenshot below. The device timestamp was divided by 1,000 and system timestamp divided by 1,000,000:

In other tests, subordinates are more likely to have close timestamp while the master is usually 100ms off.

[initialneil]

Hi, I just want to give an update on my previous problem. I've only recently realized that ALL subordinate devices must be started before the master. After I fix this, my sync problem seems to be gone, the device timestamp works now.
My bad. But I suggest in the white paper of sync, make sure people notice this. Thanks.