This is the first physical-device calibration spine for the expanded spatial rig.
## Objective
Turn connected cameras, microphones, speakers, and near-field sensors into timestamped observations that can be calibrated into one world coordinate system.
-`msmf:0` LeapUVC accepts a `320x240x120` request and reports about 115 fps, with rough OpenCV measured capture around 52 fps in the first short probe. Treat that as "tracking mode is plausible," not calibrated throughput.
- Kiyo RGB devices produce snapshots, but initial isolated mode probes against Kiyo indices timed out. Do not trust FPS configuration for Kiyos until a backend-specific capture path is chosen.
- PS3 Eye default DirectShow capture works at `640x480` near 30 fps. High-FPS tracking modes are not proven yet; mode probes against the DirectShow filter can hang or collide with another open handle, so test them sequentially.
Do not install the driver over `MI_01`; that is the working generic USB audio interface.
CL Eye / Code Laboratories is historical reference material only. Do not make this repo depend on a paid, stale, or redistributed proprietary driver. Reverse engineering should stay inside interoperability research boundaries; the live machine should prefer the open driver.
- Both PS3 Eye `MI_00` video interfaces are now bound as `libusb-win32 devices` and show `OK` in PnP.
-`E:\Tools\opentrack\modules\ps3eye-mode-test.exe` sees the cameras well enough to enumerate good modes up to `320x240@205Hz` and `640x480@83Hz`.
- Frame reads still fail with repeated `ps3eye: payload error, data[1]=78` and `can't read any frame`.
That means the next fault is not "missing driver"; it is driver/backend/USB streaming. opentrack's guide recommends `libusb-win32` first and `libusbK` as a fallback. PSMove's PS3 Eye documentation says WinUSB through Zadig has been tested with PS3EYEDriver on Windows. If libusb-win32 continues to enumerate modes but fail frames, try the fallback in this order on `MI_00` only:
-`libusbK` with both Eyes direct to motherboard: all opentrack modes report `GOOD`, including `320x240@30`, but frame reads still fail with `payload error, data[1]=78/79` and `bad header`.
-`WinUSB` with both live `MI_00` instances: opentrack mode test still reports all modes `GOOD`, but one camera open reports `Access is denied` and frame streaming still repeats `payload error, data[1]=78/79`.
-`libusb-win32`, `libusbK`, and `WinUSB` have all failed frame reads through opentrack on this machine.
- OpenCV by index does not negotiate high-rate PS3 Eye modes correctly. It can report a requested FPS while delivering the wrong shape or frame rate.
- FFmpeg by DirectShow device name does negotiate the filter pins correctly. Verified modes:
- single `PS3 Eye Universal` at `640x480@30/40/50/60`; `640x480@60` delivered about 56 fps over three seconds
- single `PS3 Eye Universal` at `320x240@30/40/50/60`; `320x240@60` delivered 180 frames over three seconds
- single `PS3 Eye Universal2` at `640x480@60` and `320x240@60`; both delivered about 59-60 fps
- dual `640x480@60` delivered about 54-56 fps per Eye over five seconds
- dual `320x240@60` delivered 300 frames per Eye over five seconds
- Requests above 60 fps through the installed DirectShow filter do not increase delivered frame rate. `320x240@70` still delivered about 60 fps; `640x480@70` delivered about 51-52 fps.
That command emits raw BGR frames to stdout. A real tracker should launch each FFmpeg capture as a child process, read fixed-size `320 * 240 * 3` frame packets, and timestamp packet arrival at the process boundary.
Also test with one PS3 Eye unplugged and each camera connected directly to a motherboard USB 2.0/3.x port, not through a hub. A libusb issue report shows PS3 Eye/opentrack access can break when certain USB-C hubs are present, even when the camera itself is not plugged into that hub.
- The two PS3 Eyes, Razer Kiyo, Razer Kiyo Pro, and Leap Motion are effectively sharing the same external hub/root path.
- After binding both PS3 Eye `MI_00` interfaces, one PS3 Eye composite device and one PS3 Eye audio interface showed `Unknown`, while both `MI_00` interfaces remained visible as `libusb-win32`.
- Two `Unknown USB Device (Device Descriptor Request Failed)` entries were also present on the same hub path.
That makes the hub/root topology a live suspect. A USB3 hub does not make old USB2 high-bandwidth devices independent; USB2 traffic still shares the hub's USB2 side and transaction scheduling. iPiSoft's multiple-PS Eye guidance says a single USB2 hub is only acceptable for low-resolution capture, and opentrack's guide says to try another USB2/USB3 controller if the camera fails.
Practical next cut:
1. Test one PS3 Eye alone on the current hub.
2. If it still fails, test one PS3 Eye alone on a different motherboard root port/controller.
3. Once one streams, add the second PS3 Eye at low tracking mode first, e.g. `320x240@60`, then increase.
4. Put Kiyos/Leap on a different hub/root path if possible.
5. If no physical split is possible, plan for lower PS3 Eye rates and treat the hub as a scarce shared resource, not a magic blue rectangle.