Description
The lab BOM is for the current Jetson Orin Nano prototype. It prioritizes model iteration, Linux camera debugging, and video pipeline development over cost, power, and weatherproofing.
Functionality covered
- Runs the current
entrance-observer application locally.
- Captures USB UVC video and records selected clips.
- Tests detection, tracking, direction counting, and telemetry upload.
- Tests on-demand video sessions through
gate-video-stream.
- Produces benchmark data for production alternatives.
Bill of materials
| Tier |
Component |
Example part |
Qty |
Rough cost |
Existing note |
Lab recommendation |
| Required |
Edge AI computer |
NVIDIA Jetson Orin Nano Super Developer Kit, 8 GB |
1 |
$249 |
Orin Nano |
Keep as the reference platform for model and API development. |
| Required |
Camera |
MOKOSE 4K USB UVC camera |
1 |
$154.50 |
Camera |
Good lab choice because UVC works with OpenCV, GStreamer, and common Linux tools. |
| Required |
Lens |
5-50 mm CS/C mount CCTV lens or camera bundled 6-12 mm lens |
1 |
€43.35 |
Camera lens |
Use varifocal only while the entrance field of view is unknown. Lock focus/zoom during tests. |
| Required |
Storage |
250 GB M.2 NVMe SSD |
1 |
€23.88 |
M2 SSD |
Stores OS, Docker images, logs, clip buffers, and benchmark videos. |
| Required |
Network |
Ethernet or Waveshare AC8265 WiFi module and antennas |
1 |
€0-22.92 |
WiFi antennas |
Prefer Ethernet for repeatable tests; use WiFi to reproduce field conditions. |
| Required |
Power |
Official-quality USB-C power supply for Jetson |
1 |
not yet documented |
Add exact part |
Missing from the old BOM. Must be sized for Jetson, camera, SSD, and WiFi under load. |
| Required |
Camera mount |
Adjustable 1/4 inch camera wall mount |
1 |
$9.59 |
Mounts |
Useful for quick alignment. Mark final distance, height, and angle after each test. |
| Recommended |
Temporary frame |
2020 aluminium extrusion pack |
1 pack |
€40.62 |
2020 Aluminum Extrusion |
Good for a lab fixture, not a field enclosure. |
| Recommended |
Optical cover sample |
Plexiglass/acrylic sheets |
1 pack |
€8.32 |
Plexiglass |
Use to test glare, blur, scratches, condensation, and cleaning. Do not assume it is production-grade. |
| Optional |
Local setup display |
7 inch HDMI touchscreen |
1 |
$47.99 |
Display |
Bench-only. Remove from field and production BOMs. |
| Optional |
Acrylic dev case |
Acrylic case originally noted for Jetson Nano 2 GB |
1 |
€11.36 |
Case |
Treat as incompatible until Orin physical fit and cooling are confirmed. |
Component decisions for Phase 1
| Decision |
Recommendation |
Why |
| Compute |
Keep Jetson Orin Nano as the reference. |
It gives the fastest route from Python model code to edge inference and reduces early software risk. |
| Camera interface |
Keep USB UVC in lab. |
USB cameras are easiest to inspect with v4l2-ctl, ffmpeg, GStreamer, OpenCV, and browser previews. |
| Lens |
Use varifocal only for discovery. |
The lab should discover working distance and entrance coverage before a fixed lens is selected. |
| Storage |
Keep NVMe. |
Video datasets, Docker images, and local buffers quickly exceed small built-in storage. |
| Networking |
Test both Ethernet and WiFi. |
Ethernet isolates camera/model bugs; WiFi reveals retry and bandwidth behavior. |
| Enclosure |
Use fixture, not final enclosure. |
A desk fixture should make camera movement and cable access fast. Field sealing is Phase 2. |
| Power |
Add a real power meter to the BOM. |
FPS/W and Wh/day cannot be estimated reliably from datasheets. |
Missing lab items to add before the next build
| Missing item |
Why it is needed |
| USB-C inline power meter or AC wall power meter |
Required for FPS/W and energy benchmarks. |
| Known test video set or printed moving-target fixture |
Makes model changes comparable. |
| USB3 cable of the same length planned for field tests |
Long or poor cables can cause camera drops. |
| Lens lock or thread-locking method |
Prevents accidental focus/zoom changes during repeated tests. |
| Small fan or thermal probe |
Helps detect thermal throttling before outdoor testing. |
| Label set for camera distance, angle, lens setting, and profile |
Makes video datasets reproducible. |
Alternatives not recommended for Phase 1 default
| Alternative |
Why not default in lab |
When to revisit |
| Raspberry Pi 5 + Hailo AI HAT+ |
Requires model conversion and Hailo runtime work before the algorithm is stable. |
Phase 2 side-by-side benchmark or Phase 3 cost-down path. |
| RK3588 NPU board |
Tooling and model conversion are less predictable than Jetson. |
If video encoding and media pipeline efficiency become more important than ML ecosystem speed. |
| Coral TPU |
Current YOLO-style detector may need major simplification. |
If a very small telemetry-only model is trained. |
| Smart/IP camera with onboard AI |
Can create vendor lock-in and less control over model iteration. |
Long-term production research after count logic and dataset stabilize. |
Estimated lab cost
Because the current notes mix USD and EUR, keep totals as planning ranges rather than a single quote:
- Required documented USD items: $413.09 (
$249.00 + $154.50 + $9.59).
- Required documented EUR items: €90.15 (
€43.35 + €23.88 + €22.92) before fixture/cover.
- Recommended fixture/cover EUR items: €48.94 (
€40.62 + €8.32).
- Optional bench/debug items: $47.99 + €11.36.
At a rough 1:1 USD/EUR planning rate, the current lab prototype is about €550-650 before shipping, taxes, power supply, cables, power meter, and assembly time.
Exit criteria
- All required parts are documented with links or substitute rules.
- The camera can run for several hours without USB resets.
- The Jetson can process the reference clips at the target FPS profile.
- Telemetry and on-demand video can be demonstrated from the same hardware.
- Power and thermal measurements are recorded so Phase 2 does not inherit unknown energy risk.
