14-bit RAW Thermal Camera Module: FlexMini 640 vs InfiRay Mini2 640

14-bit data. 16-bit storage.

FlexMini 640 provides 14-bit original grayscale data in a 16-bit Y16 container. The original quantization has 16,384 theoretical code levels; the storage uses two bytes per pixel. A native 16-bit signal would have 65,536 theoretical code levels, but the same storage size.

Y16 enumeration and a uint16 buffer do not prove 16-bit original precision. Left-shifting 14-bit values by two or multiplying them by four does not add original image detail. Actual imaging performance depends on detector noise, sensitivity and correction; it is not a fourfold gain in image quality or temperature accuracy. See the Linux V4L2 luma-format specification. FlexMini 640 output is non-radiometric grayscale, not calibrated temperature data.

Use the decoding and transport details for the selected RAW mode. A format name alone does not define that mode’s packing or alignment. Do not apply another camera’s temperature-conversion formula to FlexMini 640 RAW image values.

Select a supported source, then verify the active source

Start with what the installed firmware actually exposes. Page 4 of the public guide introduces these source selectors; the current FlexMini 640 resources clarify their capability checks. The capability mask determines which can be used on the matched module.

Separate NUC (1) and SFFC (2) selectors are removed. Their entries in the guide’s explanatory table should not be treated as selectable options.

1. Read caps_mask and current source status.
2. Request an available source and inspect selected_source and pending.
3. If the change is pending, restart the Y16 stream.
4. Read active_source to verify the effective selection.

Requested is not the same as active. Treat the read-back as the completion check, not merely the successful SET response. Source selection is runtime-only and does not write Flash; do not assume a default source or retention after power cycling.

From a matched upgrade to Linux frame capture

1. Confirm the upgrade package

Before upgrading, agree the module and board revision, interface, host and compatible firmware with CAMCUDA. The public guide shows a Windows upgrade utility and calls for a power cycle if the device has not restarted after completion. A newer filename alone is not a compatibility check.

2. Enumerate the actual video device

The guide uses the following command to list available formats. /dev/video0 is an example path; identify the correct device on your host.

v4l2-ctl -d /dev/video0 --list-formats-ext

Look for Y16 and inspect its advertised sizes and intervals. The guide illustrates 640×512 at 50 fps; sustained receive performance still needs confirmation on the delivered module and host. The Linux API separates format enumeration from format negotiation. Check the driver’s returned format rather than assuming it accepted the request unchanged.

Use three distinct checks: advertised format, active image source and received frames. Seeing Y16 in the list answers the first question. Reading source status answers the second. Capturing on the intended host answers the third. Keeping those checks separate makes a support request more useful than reporting only that an image looks wrong.

3. Use the matching capture example

The Linux example builds with make. The guide then shows this capture invocation:

./raw_y16_grab /dev/video0 frame_y16.raw 640 512 5 frame_y16.pgm 10

This is the guide’s example invocation. Use the matching README for argument definitions and adapt the device path to your host; do not infer each number’s purpose from its position. Confirm the active source and preserve the original buffer before feeding the frames into an algorithm.

Preserve the RAW buffer and troubleshoot the right layer

The supplied capture example stores consecutive little-endian uint16 frames without per-frame headers. For its 640×512 two-byte storage layout, one frame contains 655,360 bytes. That file size is not a transport-throughput measurement or a universal driver-stride rule.

The optional PGM is an 8-bit, percentile-stretched preview: useful for seeing the scene, but different from the original analysis buffer. Keep the RAW file alongside the source status and negotiated format. Otherwise, a display transformation can be mistaken for a change in the acquired data.

For comparison work, retain an untouched acquisition file and record which output each processing stage consumes. A display preview and a RAW frame should not be compared as though they were the same representation. This simple separation helps both the integration team and CAMCUDA discuss the data path without guessing what a screenshot contains.

Agree acceptance checks around the delivered configuration: source-switch behavior, shutter behavior and received-frame performance on the intended host. Keep a short record of the module and board revision, firmware, selected and active source, mode and output files. These are practical evaluation checks, not a substitute for measuring the configuration you plan to deploy.

CAMCUDA FlexMini 640 at a glance

Use this summary to frame the evaluation, then check the current product page for configuration, optics and ordering options. Hardware specifications and the firmware-dependent acquisition path are listed separately.

Guides, SDK resources and compatibility

The four-page English and German guides bring the upgrade procedure, Linux format check, capture example and source-selector reference into one place.

• English firmware upgrade and RAW/Y16 guide · PDF
• Deutsche Anleitung: Firmware-Upgrade und RAW/Y16 · PDF
• FlexMini 640 SDK and integration resources · Google Drive

SDK compatibility: the reviewed control SDK v1.11.0 main header references ir_ctrl_temperature.h, which is absent from that package. Obtain a corrected matching package and confirm header, library and hardware compatibility before build integration. It is not a complete, ready-to-build SDK release. The independent Linux V4L2 capture example is separate from this packaging limitation.

These are current integration resources, not a record of the customer’s historical versions. Confirm the approved package because shared-folder contents can change. Where guide and archive folder labels differ, use the file basenames to identify the relevant tools.

Planning an evaluation? Share your host platform, lens, interface, required source and quantity through the FlexMini 640 product page. CAMCUDA can review the requested configuration before you commit to the integration.