> ## Content Index > Fetch the complete content index at: https://datadeep.tech/llms.txt > Use this file to discover other available public pages before exploring further. # HorizonSight 360° a Helmet-Integrated Rear-Awareness AR and Mission-Control Vision System - URL: https://datadeep.tech/horizonsight-rear-awareness-vision-system/ - Published: 2026-07-07T22:48:00.000Z - Updated: 2026-07-11T05:07:07.000Z - Description: HorizonSight 360 gives a helmeted operator eyes in the back of their head, and gives their team eyes everywhere at once. - Author: John D - Tags: DIY, Engineering, Survival ### Peripheral 360° Rear-Awareness Helmet System v1 Prototype Pilot Build **A helmet-integrated situational-awareness system that maps the rearward hemisphere into a stabilized, semi-transparent HUD in the lower visual field, while preserving unobstructed natural forward vision.** > 360° horizontal awareness; forward hemisphere at full natural fidelity, rear hemisphere as a stabilized horizon band. The remote operator gets the near-full visual sphere - **License:** Hardware — CERN-OHL-S v2\. Documentation — CC BY-SA 4.0. - **Version:** 1.0 (pilot prototype) — **Date:** 7 July 2026. --- ## TL;DR - **You can build a working, see-through rear-awareness HUD for roughly $210–290 (salvage/cheapest) or \~$510–700 for the recommended v1 (Raspberry Pi 5 8 GB + dual-fisheye USB camera + a $145 0.39" micro-OLED-plus-prism kit combined into the lower lens), staying far under the $5,000 ceiling.** The recommended architecture is a micro-OLED + beam-splitter combiner in the lower visual field — it is the only see-through option that is simultaneously cheap, hand-buildable, and bright enough to read. - **Accept \~150–200 ms glass-to-glass latency for v1** — adequate for "something is approaching from behind," not for maneuvering. Treat the low-latency path, on-board AI object detection (Coral/Jetson), and full-sphere streaming to remote mission control (HorizonSight 360, \~$1,300–2,000) as documented upgrades, not requirements. - **The single most important caveat is legal/safety, not technical: attaching anything to a certified helmet almost always voids its certification (EN 1078, ASTM F1447, EN 397, ANSI Z89.1) and the manufacturer's liability.** Build on a dedicated non-relied-upon helmet or as a fully removable, non-penetrating clip, keep on-head mass minimal, and never treat this prototype as certified PPE or a substitute for shoulder checks. ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/image-10.png) --- ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/EnhancedFieldOfView.png) ## Key Findings - **The compute is cheap but its price is unusually volatile.** A Raspberry Pi 5 8 GB is the right v1 brain (best-supported camera/OpenCV stack), but 2026 DRAM shortages have pushed prices up sharply — the 16 GB Pi 5 jumped from $120 to $205, "over 70% more expensive than original MSRP," per Tom's Hardware. Budget accordingly and consider a salvaged Pi 4. - **A single off-the-shelf part collapses most of the optical risk.** A 0.39" micro-OLED kit ships *with* a magnifier prism and HDMI driver board for $145 (yxmicrodisplay.com) This is what makes the see-through HUD DIY-feasible; waveguides are not hand-fabricable and were rejected. - **The AI upgrade path is affordable.** The NVIDIA Jetson Orin Nano Super dev kit "delivers up to 67 TOPS of AI performance… At just $249" (nvidia.com), with 102 GB/s bandwidth. Staying on the Pi, a Google Coral USB accelerator is "capable of performing 4 trillion operations per second (4 TOPS)… execute state-of-the-art mobile vision models such as MobileNet v2 at almost 400 frames per second" (coral.ai). - **No commercial product does exactly this.** The closest safety analogue, the radar-only Garmin Varia RVR315 (\~$149.99), gives an alert with no image; the camera-equipped Varia RCT715 has "a suggested retail price of $399.99" (Garmin newsroom) but still no live rear HUD; the newer RearVue 820 is $299.99\. Industrial AR helmets are forward-focused and cost 3–10× the full upgrade path. - **\~200 ms live streaming is a solved problem on this hardware.** WebRTC stacks (MediaMTX, TzuHuanTai/RaspberryPi-WebRTC) hit \~0.2 s latency on a Pi 5, enabling the HorizonSight mission-control feed over SATCOM/WiFi/LTE. --- ## Details ### 1\. At-a-Glance | Item | Value | | ------------------------------------ | --------------------------------------------------------------------------------------------------------- | | **Cheapest working path** | \~$210–290 USD (salvaged SBC, single rear camera, peripheral LCD/EVF) | | **Recommended v1 path** | \~$510–700 USD (Pi 5 8 GB, dual-fisheye USB cam, 0.39" micro-OLED + combiner) | | **Upgraded path (HorizonSight 360)** | \~$1,300–2,000 USD (Jetson Orin Nano Super or Pi 5 + Coral, LTE streaming) | | **Budget ceiling** | $2,500 (this build stays far below it) | | **Estimated build time** | 30–50 hours over 2–4 weekends | | **Difficulty** | Intermediate–Advanced (Linux, soldering, basic optics alignment, 3D print or hand-fab) | | **Key tools** | Soldering iron, multimeter, 3D printer (or print service), calipers, small hand tools, PC for calibration | > Prices are **estimates** and vary by region and date. Single-board-computer and RAM prices are exceptionally volatile in 2026\. Verify all prices, local laws, and electrical safety yourself before building. --- ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/Goggle3.png) ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/Goggle2.png) ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/Goggle1.png) Whether via on-board controls, or purely psychological filtering with semi-transparent HUD; the effect is pictured above. It allows you to have simultaneous peripheral awareness of your forward and rearview surroundings. The mission control upgrade also facilitates a remote operator to observe your surroundings in 360° view. ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/HorizonSight360.png) --- ### 2\. Abstract / Purpose **What it is.** A wearable rear-awareness system. A crown-mounted wide-angle (fisheye) camera captures the rear hemisphere. A small single-board computer (SBC) dewarps and stabilizes that feed and renders a compressed panoramic "rear strip" onto a micro-display, which is optically combined into the **lower** portion of the wearer's visor or goggle lens. The wearer keeps normal, unobstructed forward vision through analog-translucent optics at all times; the digital image occupies only the lower visual field, where humans naturally tolerate reference information (like a car dashboard or bicycle computer). **The problem it solves.** Cyclists, search-and-rescue (SAR) workers, industrial/mining workers, and skiers/ATV riders all suffer from a rearward blind zone. Shoulder-checks break forward attention and are impossible in some postures or helmets. Commercial answers are either narrow (rear radar such as Garmin Varia, no image) or extremely expensive and forward-focused (industrial AR helmets). This project gives continuous peripheral awareness of rearward motion for parts-cost in the low hundreds of dollars. **Who it's for.** A university-level builder comfortable with Raspberry Pi/Linux, wiring, and basic electronics, who accepts a bespoke prototype and is willing to verify safety locally. **Limitations:** - This is a **v1 pilot prototype**, not certified safety equipment. - **Attaching anything to a certified helmet almost always voids its certification** (EN 1078, ASTM F1447, EN 397, ANSI Z89.1) and its manufacturer liability. Addressed plainly in the Skills & Safety section. - It is an **awareness aid, not a substitute for shoulder checks or mirrors** in traffic. - Latency is \~175ms glass-to-glass in the recommended build; fine for detecting "something is approaching from behind," not for precision maneuvering. --- ![Conceptual / Illustrative Purposes Only](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/FutureGogglev2.png) Conceptual / Illustrative Purposes Only --- ### 3\. Design Rationale & Theory of Operation #### 3.1 Governing principles **Fisheye projection & dewarping.** A fisheye lens maps the scene by an angular projection (commonly equidistant, r = f·θ, where θ is the angle from the optical axis and r the image radius) rather than the rectilinear r = f·tan θ of a normal lens. This is what lets one lens see \~180–200°. The cost is heavy barrel distortion. OpenCV models this with a camera matrix **K** and fisheye distortion coefficients **D** (k1–k4); the `cv2.fisheye` module is distinct from the pinhole model and must be used for wide lenses (a common failure mode is running fisheye images through the pinhole calibrator and getting "avant-garde-style drawings," as the StereoPi tutorials warn). To turn raw fisheye into a usable rear strip we build a **remap table** once and then apply it every frame with `cv2.remap`. As the remap table is precomputed, per-frame cost is just a memory-bound resample; cheap enough to run in real time on a Pi 5. **Why a panoramic "strip," not a rectified rectangle.** The rear hemisphere is best shown as a wide, short panorama (a rear-view "letterbox"). We reproject the fisheye into a cylindrical/equirectangular strip: azimuth maps to horizontal position, a limited elevation band maps to the short vertical axis. This preserves "something is to my rear-left vs. rear-right" a spatial cue that matters for awareness. **Human lower-visual-field ergonomics.** The lower visual field is where humans habitually place reference instruments and where gaze naturally drops without losing forward road/hazard awareness. Placing the HUD strip low and small keeps the primary forward field completely clear. This is a deliberate safety choice: the display must never occlude the forward hazard field. **Vergence accommodation & focal distance.** A near micro-display seen directly would force the eye to focus at \~3 cm, uncomfortable and fatiguing. The combiner optic (a small lens/prism + semi-reflective surface) forms a **virtual image** at a comfortable distance. For v1 we set the virtual image at roughly 1–2 m so the eye's accommodation is close to its relaxed forward state, reducing the vergence–accommodation conflict that is a known cause of AR eye strain. We keep displayed content sparse and low to further reduce strain (INAIRSPACE.com). **Stabilization.** Head motion makes a raw rear strip nauseating. An [IMU](https://shop.pimoroni.com/en-us/products/adafruit-9-dof-orientation-imu-fusion-breakout-bno085-bno080-stemma-qt-qwiic?ref=datadeep.tech) (accelerometer + gyro + magnetometer, fused) on the helmet gives orientation. We counter-rotate the panoramic sampling window against measured head roll/pitch so the horizon in the rear strip stays level ("electronic horizon lock"). This is [digital](https://learn.adafruit.com/adafruit-9-dof-orientation-imu-fusion-breakout-bno085/report-types?ref=datadeep.tech), not mechanical - no gimbal needed. Yaw is intentionally NOT fully cancelled (to allow the rear view to turn with you a little), only smoothed (Pimoroni.com) #### 3.2 Display-architecture tradeoff and the commitment The analog forward visual requirement is absolute: **the visor/goggle must stay analog-translucent; forward vision is never digitally mediated.** That rules out any opaque or fully occluding display and rules out video-passthrough VR. Three viable see-through architectures were considered: | Architecture | Pros | Cons | Verdict | | -------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | ------------------------------------- | | **A. Micro-OLED + combiner optic in lower lens** (recommended) | Cheap, bright, high-contrast micro-OLED kits exist off-the-shelf with HDMI driver + magnifier prism; combiner can be beam-splitter or coated acrylic; fully see-through; proven in DIY HUD prior art | Requires careful optical alignment (eye relief, combiner angle); modest FOV | **CHOSEN** | | **B. Waveguide** (diffractive/reflective) | Thin, elegant, large eyebox, used in commercial AR | Waveguide combiners are "technically sophisticated… usually hard to design," expensive, not DIY-fabricable | Rejected for v1 (cost + buildability) | | **C. Direct small transparent/peripheral LCD** (no combiner) | Simplest; cheapest; transparent OLED/LCD panels exist | Image sits at panel focal distance (too near → eye strain); low brightness/contrast; transparent panels wash out in daylight | Kept only as the *cheaper variant* | **Why A wins:** it is the only option that is simultaneously (1) truly see-through, (2) buildable with globally available parts and hand tools, (3) bright enough to read against real backgrounds (micro-OLEDs at 500–3000 cd/m²), and (4) cheap. Off-the-shelf 0.39" micro-OLED kits ship *with* the magnifier prism and HDMI driver board, collapsing most of the optical-engineering risk into a single $145 part. #### 3.3 Key parameter choices - **Captured FOV:** \~180–200° per lens. A single crown-mounted fisheye covers the rear hemisphere; a dual-fisheye module covers full 360° for the upgrade path. - **Displayed FOV:** a rear strip spanning \~180° azimuth (rear-left through rear-right), \~20–30° elevation band. - **Display resolution:** 800×600 (0.39" kit) is plenty; the strip is downsampled anyway. - **Latency budget:** \~150–200 ms glass-to-glass accepted for v1. - **Combiner angle:** \~45° between micro-display output and eye line is the nominal starting point; fine-tuned per build. ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/Helmetv1.png) --- ### 4\. Specifications & Performance Targets | Parameter | Target (v1 recommended) | Notes | | ---------------------- | ----------------------------------------------------- | --------------------------------------------------------------------- | | Captured FOV | 180–200° (single fisheye) / 360° (dual) | Lens-dependent | | Displayed FOV | \~180° az × \~25° el rear strip | Downsampled panoramic | | Display | 0.39" micro-OLED, 800×600, 500 cd/m² | Higher-brightness (1800–3000 cd/m²) panels exist for daylight | | Glass-to-glass latency | \~150–200 ms | Acceptable for awareness, not maneuvering; low-latency path = upgrade | | Frame rate | 15–30 fps | 15 fps adequate for motion cueing | | Power draw | \~7–12 W (Pi 5 + cam + display) | Pi 5 draws far more than Pi 4; budget accordingly | | Battery runtime | 2–4 h on a 10,000 mAh USB-C PD bank | Scales with pack size | | Added helmet mass | Target < 250g on-head; preferred max of \~500 g | **Neck-strain & impact-safety concern** | | Operating temp | \~ -10 to +50 °C (limited by micro-OLED & battery) | 0.39" kit rated -10 to +70 °C; Li-ion cold/heat limits are tighter | | Service life | Prototype; expect to iterate | Micro-OLED lifetime per its datasheet | | Duty cycle | Intermittent-to-continuous; thermal-limited under sun | Active cooling recommended on Pi 5 | **Mass safety note:** added mass on a helmet increases rotational and inertial loading on the neck in a crash and can change the helmet's certified behavior. Keep the heavy items (battery, SBC) **off** the helmet, when possible - (belt/vest mounted) with only the camera, display and combiner on the head. Optimize for cable management. --- ### 5\. Bill of Materials Prices are **estimates in USD**, single-unit, July 2026, and vary by region/date. #### Recommended v1 path | # | Item | Spec/size | Qty | Generic name (or specific model + why) | Est. unit | Line total | Source / notes | Salvage alt. | | -- | --------------------- | ------------------------------------------------------- | ----- | -------------------------------------------------------------------------------------------- | ------------- | --------------- | ----------------------------------------------------- | ------------------------------------------------------------------------------- | | 1 | Single-board computer | Quad-core ARM, ≥4 GB | 1 | Raspberry Pi 5 (8 GB) — best-supported camera/OpenCV stack | $95 | $95 | Authorized reseller. **Price volatile — DRAM crisis** | Salvaged Pi 4 (8 GB); Pi 5 4 GB (\~$75) | | 2 | microSD card | 32 GB A1/A2 | 1 | Generic 32 GB microSD | $8 | $8 | Big-box/online | Reuse any ≥16 GB card | | 3 | Camera | Dual-fisheye USB, UVC | 1 | USB2.0 dual-fisheye 360 module (\~180° ×2), UVC 1080p | $139 | $139 | Generic online marketplace (FEBON-class) | Salvaged Ricoh Theta / Insta360 / GoPro in USB webcam mode; single RPi wide cam | | 3b | (Alt camera) | RPi Camera Module 3 Wide | 1 | Raspberry Pi Camera Module 3 Wide (120° FoV, autofocus); $35 MSRP, \~$38.50 retail | $38.50 | — | Authorized reseller (PiShop) | Any CSI cam + fisheye lens; std Camera Module 3 \~$29.25 | | 4 | Micro-display kit | 0.39" OLED 800×600 + HDMI driver + prism | 1 | 0.39" micro-OLED kit w/ HDMI board & resin prism — ships with optics, collapses optical risk | $145 | $145 | Specialist microdisplay vendor (yxmicrodisplay) | Salvaged EVF from dead camera/camcorder | | 5 | Combiner optic | 50/50 beam-splitter glass or coated acrylic, \~40–50 mm | 1 | Beam-splitter ("teleprompter") glass 50/50 | $15 | $15 | Optics/teleprompter supplier; eBay | Coated acrylic; salvaged game beam-splitter | | 6 | IMU | 9-DOF fused | 1 | Adafruit BNO085 ($24.95, on-chip fusion → trivial horizon lock) | $24.95 | $24.95 | Adafruit / Mouser | MPU-6050/ICM-20948 (\~$4–15, more code) | | 7 | HDMI adapter | micro-HDMI → HDMI, short | 1 | Pi 5 micro-HDMI cable/adapter | $6 | $6 | Big-box | Salvage | | 8 | Power bank | USB-C PD, 10,000 mAh, ≥27 W | 1 | USB-C PD power bank | $25 | $25 | Big-box | Reuse phone power bank | | 8b | PD trigger board | PD→5V/5A for Pi 5 | 1 | USB-PD to 5V/5A converter board (Pichondria/GeeekPi-class) | $12 | $12 | Online | — | | 9 | Helmet mounts | GoPro-style adhesive (curved+flat) | 1 kit | Generic GoPro-style adhesive mount kit (e.g. 8-pack: 4 flat + 4 curved + 3M pads) | $12 | $12 | Online marketplace | Salvaged action-cam mounts; zip-ties | | 10 | Enclosure | 3D-printed PETG for SBC + battery | 1 set | Printed enclosure & combiner mount | \~$5 filament | $5 | Self-print or print service | Project box; foam | | 11 | Wiring/consumables | Silicone wire, JST, heatshrink, standoffs | 1 set | Generic | $15 | $15 | Big-box/online | Salvage | | 12 | Active cooler | Pi 5 heatsink+fan | 1 | Pi 5 active cooler | $7 | $7 | Reseller | Salvaged heatsink | | | | | | | **Subtotal** | **≈ $508–$525** | | | #### Cheapest path (single rear cam, peripheral LCD) | Item | Choice | Est. | | ------------------- | ---------------------------------------------------- | ---------------------------------- | | SBC | Salvaged Pi 4 / Pi Zero 2 W | $0–35 | | Camera | Single wide/fisheye USB or CSI cam | $15–38 | | Display | Small transparent/peripheral LCD **or** salvaged EVF | $10–30 | | Combiner | Coated acrylic | $5 | | IMU | MPU-6050 | $4 | | Power | Reused phone power bank | $0–15 | | Mounts/wiring/print | Salvage + filament | $10 | | **Total** | | **≈ $210–290** (less with salvage) | --- ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/image-13.png) #### Upgrade path (HorizonSight 360) - added items | Item | Choice | Est. | | --------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------- | | SBC | NVIDIA Jetson Orin Nano Super dev kit — "up to 67 TOPS… At just $249" [Notebookcheck](https://www.notebookcheck.net/The-Nvidia-Jetson-Orin-Nano-Super-a-powerful-generative-AI-SBC-is-now-available-worldwide-for-249.934029.0.html?ref=datadeep.tech) (NVIDIA), 102 GB/s, 7–25 W | $249 | | AI accelerator (if staying on Pi) | Google Coral USB (Edge TPU, 4 TOPS, 2 W, MobileNet v2 \~400 fps) or Hailo-8 M.2 (26 TOPS) | $60–110 | | Cellular | LTE CAT4 HAT/USB modem (Quectel-class) + data SIM | $50–90 + SIM | | Bonded uplink (optional) | 2nd modem / bonding service | varies | | Higher-brightness micro-OLED | 0.49" 1920×1080 1800 cd/m² | \~$150–300 | | Protected 18650 pack | 2× protected 3500 mAh cells (\~$10–25 each) + 2S BMS + holder | \~$30–60 | | **Upgrade delta** | | **≈ $600–900** on top of v1 | **Cost summary:** Cheapest ≈ **$210–290**; Recommended v1 ≈ **$510–700** with cables/tax; Full HorizonSight ≈ **$1,300–2,000**. --- ### 6\. Tools & Equipment | Tool | Category | Est. cost | Manual / borrow fallback | | ------------------------- | ---------------- | ------------ | --------------------------------------------------- | | Soldering iron + solder | Likely owned | $25–40 | Borrow; makerspace | | Multimeter | Likely owned | $15–30 | Borrow | | Digital calipers | Likely owned/buy | $15 | Ruler (less precise) | | 3D printer | Borrow/rent | $0 (service) | Online print service; hand-fab from ABS sheet/epoxy | | PC/laptop | Likely owned | — | Required for OpenCV calibration | | Small screwdriver/hex set | Likely owned | $10 | — | | Hot-glue gun / epoxy | Buy | $8 | — | | Heat gun (heatshrink) | Borrow | — | Lighter (carefully) or hairdryer | | Helmet clamp/vise | Borrow | — | Hand-hold during work | --- ### 7\. Skills & Safety **Required competencies:** Linux command line; Python + OpenCV basics; soldering; reading a wiring diagram; basic 3D CAD or willingness to use provided files; patience for optical alignment. **Hazards specific to this build** 1. **Modifying certified PPE voids certification!** Bicycle helmets (EN 1078 / ASTM F1447), industrial helmets (EN 397 / ANSI Z89.1), ski helmets (EN 1077), and eye protection (ANSI Z87.1 / EN 166) are certified **as tested**. Drilling, gluing, adding mass, or attaching hardware **almost always voids the certification and the manufacturer's liability**. Under EN 397:2025, per JSP Safety's technical guidance, "helmet accessories that may affect conformity of the helmet (e.g. during impact testing), must be mounted during testing and effectively certified as compatible" - i.e., an untested accessory breaks compliance. **Do not modify a helmet you rely on for occupational or legal protection.** For a test build, use a **dedicated, non-relied-upon helmet**, or design the rig as a fully removable clip that does not penetrate the shell. State clearly to any user that the modified unit is not certified PPE. 2. **Lithium battery handling.** Use protected cells or a quality PD power bank with BMS. Never charge unattended; never puncture/crush; keep the pack off the head; fuse the 5 V line. A brownout can crash the Pi mid-use. 3. **Added helmet mass & retention loading.** Extra mass increases neck load and rotational injury risk in a crash and can defeat the helmet's energy management. Keep on-head mass minimal (target <250 g) and re-do a retention/strap check. Note that EN 397 chinstraps are *designed to release* at 150–250 N to prevent injury— do not obstruct this with your cabling. 4. **Vision obstruction & distraction.** The HUD must stay in the lower field and be dimmable. Distraction is a primary concern. Never occlude the forward visual field. 5. **Night brightness / eye safety.** Auto/manual-dim the display at night; an over-bright near-eye source degrades dark adaptation and causes strain. 6. **Soldering fumes / epoxy / hot glue.** Ventilate; eye protection; avoid burns. 7. **Road legality.** Some jurisdictions restrict head-mounted displays or helmet cameras/attachments for road users, and some regulate/prohibit the camera function on privacy grounds (Garmin ships the Varia camera with exactly this warning). **Check local law.** **Required PPE while building:** safety glasses (soldering, cutting acrylic), vapor protection / ventilation for fumes. --- ### 8\. Build Instructions Reference parts by BOM #. Go/no-go checks are marked **\[GATE\]**. **Phase 1 - Bench bring-up (compute + camera)** 1. Flash Raspberry Pi OS (64-bit) to microSD (#2); boot Pi 5 (#1) with active cooler (#12). 2. Plug in dual-fisheye USB cam (#3). Confirm enumeration (`v4l2-ctl --list-devices`) and grab a frame. **\[GATE\]** You see two raw fisheye circles. 3. Install OpenCV, numpy, and your capture stack. **Phase 2 - Software pipeline** 4\. **Calibrate** the fisheye (checkerboard, `cv2.fisheye.calibrate`) → save K, D. 5\. Build the **remap table** from fisheye to a rear-panoramic strip; apply with `cv2.remap` each frame. 6\. **Stabilize:** read BNO085 (#6) over I²C; counter-rotate the sampling window for roll/pitch; smooth yaw. 7\. **Motion highlighting:** run OpenCV MOG2 background subtraction (cheap) to outline approaching movement; optionally a lightweight object detection algorithm ([YOLO](https://www.datacamp.com/blog/yolo-object-detection-explained?ref=datadeep.tech), on Coral/Jetson in the upgrade). Draw a subtle box/arrow. 8\. Output the composited strip to HDMI at the display's native timing. **\[GATE\]** Strip is level when you tilt your head; approaching motion is highlighted. **Phase 3 - Display + combiner bench prototype** 9\. Drive the 0.39" micro-OLED kit (#4) from micro-HDMI (#7); confirm image. 10\. On a bench (the "Lego-block" optical-bench method common in DIY HUD prior art), position micro-display, prism, and combiner glass (#5). Set combiner \~45°; adjust distances so the **virtual image sits \~1–2 m away** and is sharp. Set eye relief so the full strip is visible without hunting. **\[GATE\]** You can read the strip *and* see straight through the combiner. **Phase 4 - Helmet integration & cable management** 11\. Mount fisheye at the **crown**, aimed to cover the rear hemisphere, on a GoPro-style adhesive mount (#9) - or a removable clip that does not penetrate the shell. 12\. Mount combiner + micro-display in the **lower** lens region on a printed bracket (#10). Route the thin display + camera cables along a strap channel; strain-relief everything. 13\. Keep SBC + battery **off-head** (belt/vest) with a single tidy umbilical, or in a rear counterweight pod if on-head. **\[GATE\]** No cable can snag; nothing occludes forward view; retention still closes and releases properly. **Phase 5 - Battery & power** 14\. Power Pi 5 via PD bank (#8) + PD-trigger board (#8b) delivering 5 V/5 A. Fuse the line. Verify no undervoltage warnings under load (`dmesg`). **\[GATE\]** 30-min run with no brownout/throttle. **Phase 6 - Field trials** 15\. Static test → walking test → intended-activity test at low intensity, in a safe area, with a spotter. Verify that the rear strip truly helps and never distracts from forward hazards. *Suggested figure captions:* (P1) close-up of the two raw fisheye circles in the capture window; (P2) side-by-side of raw fisheye vs. dewarped rear strip with a level horizon line drawn; (P3) bench optical layout showing micro-OLED, prism, combiner at 45°, and the eye position, with the virtual image plane marked at \~1–2 m; (P4) helmet with crown camera and lower-lens combiner, cable routing highlighted; (P5) power umbilical and off-head pod. --- ### 9\. Drawings & Schematics Recommended free tools: **FreeCAD** (enclosure, brackets), **KiCad** (any custom PCB), **LibreCAD** (2D fab drawings), **OpenSCAD** (parametric mounts), **Inkscape** (panel/labels). **Exploded view (describe):** helmet shell; crown camera mount + fisheye; strap-routed cable; lower-lens combiner bracket holding beam-splitter at 45° and micro-display facing it through the prism; off-head pod with Pi 5, cooler, PD board, battery. **Fabrication drawings & tolerances:** - *Combiner mount:* combiner angle 45° ± 2°; display-to-combiner distance set at bench (±0.5 mm affects focus); eyebox aligned to wearer's resting gaze, biased low. - *Crown camera mount:* rigid, aimed to place rear horizon in mid-strip; ±3° aim tolerance. - *Electronics enclosure:* clearance for Pi 5 active cooler airflow; cable strain-relief; battery retained separately. **Wiring schematic (power tree):** ``` USB-C PD bank ──▶ PD-trigger (5V/5A) ──▶ Pi 5 (fused) Pi 5 USB ─────────────────────────────▶ Dual-fisheye USB camera Pi 5 micro-HDMI ──────────────────────▶ micro-OLED HDMI driver ──▶ micro-OLED panel Pi 5 I²C (SDA/SCL, 3V3, GND) ─────────▶ BNO085 IMU (Upgrade) Pi USB3 ─────────────────────▶ Coral USB TPU (Upgrade) Pi USB/HAT ──────────────────▶ LTE modem ``` **ASCII side-profile optical path (lower-lens combiner):** ``` eye \ (line of sight, forward, THROUGH combiner) \ ______ combiner glass (50/50) @ ~45° \______/______________ forward world (see-through) \ /| virtual \ / | reflected HUD ray image ~1–2m \/ | / | prism + | micro-OLED --+ (image source, faces up into combiner) ``` --- 0:00 /0:06 1× --- ### 10\. Testing, Calibration & Validation **Pre-first-use safety checks (all must pass):** - **Camera calibration:** ≥15–20 checkerboard captures; fisheye reprojection error < \~1 px. - **Dewarp accuracy:** straight rear reference lines map smoothly across the strip; no discontinuities. - **Horizon stabilization drift:** tilt head ±30°, hold; residual horizon tilt in strip < \~3°; drift after 5 min within a few degrees (BNO085 fusion should hold it). - **Display alignment & eye-strain screening:** strip readable at resting gaze without refocusing effort; run 10 min, self-check for strain/nausea; stop if present. - **Retention pull test after added mass:** with rig installed, verify straps/retention still hold and release as designed; confirm on-head mass within budget. - **Battery load & thermal test:** full-brightness, full-pipeline 30-min run; log Pi temperature; confirm no throttle/undervoltage. - **Glass-to-glass latency measurement:** point the camera at a running millisecond stopwatch on a screen, photograph the stopwatch and the HUD strip together, subtract. Verify within the \~150–200 ms goal. (For reference, un-optimized Pi camera pipelines measure \~120–200 ms glass-to-glass by the stopwatch method on older Pi hardware; a Pi 5 with a precomputed remap should land in this range.) - **Motion-detection true/false-positive bench test:** wave objects behind at varying speeds; log detections vs. false alarms; tune MOG2 thresholds until useful. Do not rely on the system in real activity until every check passes. --- ### 11\. Operation **Do:** use it as a peripheral awareness aid; keep it dim at night; keep forward vision primary; power up and confirm the strip is stabilized before moving. **Don't:** treat it as a substitute for shoulder checks/mirrors in traffic; stare at it; use it in heavy rain/fog beyond the camera's or your own limits; do NOT rely on it as certified PPE. **Operating envelope:** day/night with brightness adjustment; avoid water ingress (prototype is not weather-sealed); thermal-limited under direct sun. ### 11.1 FAQ **What does it look like from inside?** > A small, semi-transparent letterbox screen floating in space roughly 1–2 m ahead of you and about 25° below your normal gaze, which is about where the top of a car's hood sits, or where a bicycle computer would be if it hovered off the handlebars. Due to the combiner being half-mirrored, the real world remains visible *through* the panorama; the rear-view strip looks like a faint ghost overlay on whatever is actually down-forward of you (ground, hood, handlebars). It moves with your head like anything mounted to your goggles would, but it *focuses* like a distant object. **Would it appear in front of the user?** > Yes; in front and below. Down-forward along your glance direction. Not pasted on the lens, not hovering at your nose. That's the whole trick: the physical display is at 25 mm, the *apparent* display is at 1.5 m. **How this avoids eyestrain and discomfort:** > When you flick your eyes from the road (focused at 10+ m) down to the panorama (focused at 1.5 m), that's a small, fast, natural refocus, the way you've comfortably glanced at a car dashboard. Glancing at a screen focused at 25 mm would instead demand more accommodation than a human eye possesses; you'd see blur plus eyestrain from the failed attempt. The combiner geometry converts an impossible focal demand into a routine one. This is the same principle as a car windshield HUD or a teleprompter: the newsreader isn't focusing on the glass in front of the lens, they're focusing on the virtual image of the text placed near the camera's distance. Same physics, rotated 90°. ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/SideviewHUD.png) *One practical implication*: since the virtual image distance is set by the display-to-lens spacing inside the prism kit, you get this behavior essentially for free, as the kit's optics are pre-spaced. Your alignment jobs are the 45° combiner angle and aiming the eyebox at the wearer's natural downward glance. --- ### 12\. Maintenance | Interval | Task | Consumable/wear part | Est. cost | | --------- | ------------------------------------------------------------------ | -------------------- | --------- | | Each use | Wipe combiner & lens; check mounts/cables; battery charge | — | $0 | | Weekly | Re-seat connectors; verify stabilization; inspect adhesive mounts | GoPro adhesive pad | \~$2 | | Monthly | Re-run quick latency & motion check; clean fan/heatsink | — | $0 | | \~6–12 mo | Re-calibrate fisheye; inspect battery health; replace worn cabling | Wire/JST | \~$10 | | As needed | Replace micro-OLED (aging/burn-in) or battery | Micro-OLED / cell | $10–145 | ### 13\. Troubleshooting | Symptom | Likely cause | Fix | | ---------------------------------------- | ------------------------ | ---------------------------------------------------------------------------------------------- | | Video dropout | USB bandwidth / cable | Use USB3 port; shorten/replace cable; lower resolution/fps | | Camera won't enumerate | UVC/driver | Check v4l2-ctl; try another port; confirm module powered | | Choppy/high latency | CPU-bound dewarp | Precompute remap; lower fps/res; offload to Coral/Jetson | | Thermal throttling | No cooling under sun | Fit active cooler; shade the pod; reduce clocks | | IMU drift / tilted horizon | Magnetometer/calibration | Recalibrate BNO085; add complementary filter; keep IMU away from magnets/motors | | Combiner fogging | Humidity/temp | Anti-fog coating; small vent; warm-up | | Battery brownout / undervoltage warnings | Weak PD negotiation | Use 5V/5A profile + PD-trigger board; thicker/shorter cable; bigger pack | | Software crash | Pipeline exception | Run under a **systemd watchdog**/supervisor that auto-restarts the vision service; log to file | | Washed-out HUD in daylight | Display too dim | Use higher-nit micro-OLED; increase combiner reflectivity; shade | --- ![Wide 2:1 equirectangular full-sphere panorama from the HorizonSight 360 camera's video feed - illustrative Purposes Only](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/HorizonSight_Mongolia.png) Wide 2:1 equirectangular full-sphere panorama from the HorizonSight 360 camera's video feed - illustrative Purposes Only ### 14\. Variations, Scaling & Customization - **Cheaper version:** single rear standard/wide camera, no stitching, small peripheral transparent LCD or salvaged EVF. \~$210–290\. Loses full 360°, dimmer, but proves the concept. - **HorizonSight 360 upgrade:** [Jetson Orin Nano Super](https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/nano-super-developer-kit/?ref=datadeep.tech) (67 TOPS) or Pi 5 + Coral/Hailo for on-board 360 stitch + object detection; **stream a full spherical feed to remote mission control** over WiFi/LTE via **WebRTC** (≈200 ms achievable on Pi-class hardware with MediaMTX/pi-webrtc; RTMP for record-only). Budget extra bandwidth (1–5 Mbps typical LTE uplink; more for full sphere), power, and thermal headroom; consider bonded cellular for reliability; add a signaling/relay server for NAT traversal. - **Per-helmet integration:** - *Bicycle (EN 1078/F1447):* minimize mass, aero pod on rear; strong road-legality caveat. - *SAR:* ruggedize, weatherproof, add mission-control streaming; helmet often non-certified-critical. - *Industrial/mining (EN 397/Z89.1):* **accessory-compatibility is a certification issue under EN 397:2025** \- use only vendor-sanctioned accessory rails; consider intrinsically safe (ATEX) constraints in gas/dust zones. - *Ski/ATV goggles (EN 1077):* integrate combiner into goggle lower lens; cold-rated battery kept in a warm pocket. - **Regional adaptation:** source SBC/camera locally; verify voltage/charger and radio (WiFi/LTE band) legality. --- ### 15\. Cost Analysis | Solution | Approx. cost | What you get | | ------------------------------------- | --------------------------- | --------------------------------------------------------- | | **This build — cheapest** | $210–290 | Rear motion awareness, DIY, repairable | | **This build — v1 recommended** | $510–700 | Stabilized 360-capable rear HUD, see-through | | **This build — HorizonSight 360** | $1,300–2,000 | Above + remote mission-control streaming + on-board AI | | Garmin Varia rear radar (RVR315) | $149.99 | Radar alert only (140 m range, 51 g), no image | | Garmin Varia radar + camera (RCT715) | $399.99 MSRP | Rear radar + 1080p recording, no live HUD | | Garmin Varia RearVue 820 | $299.99 | Radar + brake light, threat/size tracking, no image | | RealWear Navigator 500 | \~$2,300 (quote-only; est.) | Industrial monocular assisted-reality (forward, not rear) | | Trimble XR10 + HoloLens 2 | \~£4,450 (\~$5,600) | Hard-hat mixed reality (forward) | | Industrial AR smart helmet (enhanced) | multiple $1,000s | Forward AR + comms | **Interpretation:** No commercial product does exactly what this does (a stabilized, see-through *rear* HUD). The closest safety analogue, Garmin Varia, gives only a radar alert with no image (RVR315) or records without a live HUD (RCT715), yet costs $149.99–399.99 - so the cheapest build is cost-competitive while adding an actual rear *view*. Commercial AR helmets cost 3–10× the full upgrade path and are forward-focused. **Payback** is really about capability you can't otherwise buy at this price, plus full repairability and no subscription/lock-in. **Total build time:** \~30–50 h. --- ### 16\. References, Prior Art & Attribution - **OpenCV** — fisheye camera model & calibration (`cv2.fisheye`), MOG2 background subtraction, `remap`. - **Raspberry Pi** — camera stack (libcamera), Pi 5 platform, OS. - **DIY HUD / smart-glasses prior art** — Hackaday "homemade Google Glass" builds (beam-splitter + micro-display + prism; "Lego-block" optical alignment); the $60 Bluetooth head-mounted display; June\_Glasses (ESP8266 HUD) on GitHub; INAIRSPACE DIY HUD guides (vergence–accommodation, combiner, exit-pupil). - **Fisheye/stereo tooling** — StereoPi fisheye calibration tutorials; surround-view fisheye projects on GitHub. - **Streaming** — TzuHuanTai/RaspberryPi-WebRTC; MediaMTX WebRTC on Pi 5. - **Standards to check locally** — EN 1078, ASTM F1447 (bicycle); EN 397 (incl. 2025 revision, Type 1/Type 2), ANSI/ISEA Z89.1 (industrial); EN 1077 (ski); ANSI/ISEA Z87.1, EN 166 (eye protection). - **Commercial comparators** — Garmin Varia line (RVR315, RCT715, RearVue 820); RealWear Navigator 500; Trimble XR10. --- ### 17\. License & Contribution - **Hardware:** CERN-OHL-S v2 (strongly reciprocal). Distribute your source (CAD, wiring, BOM) with any hardware you make/sell based on this. - **Documentation:** CC BY-SA 4.0 (attribute, share alike). - **Contribute back:** fork the repo, improve the optics/mounts/software, publish your calibration files and STLs, and share latency/mass measurements per helmet type so others can build on real data. --- ## Recommendations 1. **Build the recommended v1 first, on a dedicated throwaway helmet; do not touch relied-upon PPE.** Prove the pipeline on the bench (Phases 1–3) before you ever attach anything to a helmet. Threshold to proceed to head-mounting: dewarped strip stays level under ±30° head tilt and MOG2 reliably flags approaching motion. 2. **Keep mass off the head.** Belt/pack-mount the Pi 5 and battery; only camera + combiner + display go on the helmet. If on-head mass exceeds \~250 g, redesign before field use; hard-stop at 500 g. 3. **Gate every field trial on the §10 checklist**, especially the retention pull test and a measured glass-to-glass latency within 150–200 ms. If latency exceeds \~250 ms, precompute the remap and drop resolution/fps before adding hardware. 4. **Only spend on the HorizonSight upgrade once v1 is genuinely useful.** The trigger to move to a Jetson Orin Nano Super ($249) or add a Coral USB ($60–110) is a concrete need for real-time object *classification* or live spherical streaming, not before. For streaming, target ≥2 Mbps sustained LTE uplink and add bonded cellular only if you see dropouts. 5. **Budget defensively on the SBC.** Pi 5 pricing is swinging with the DRAM market; if the 8 GB is inflated when you buy, a salvaged Pi 4 8 GB or a Pi 5 4 GB is an acceptable v1 substitute. Re-check prices the week you order. 6. **Document and publish** your calibration files, STLs, and measured latency/mass per helmet type under the project licenses so the next builder starts from data, not guesses. --- ## Caveats - **This is a v1 prototype, not certified equipment**, and it is an awareness aid, never a replacement for shoulder checks, mirrors, or certified PPE. - **All prices are estimates** and were captured in July 2026; SBC/RAM prices in particular are unusually volatile (the Pi 5 has seen repeated price hikes). Verify before purchase. - **The measured-latency reference figures (\~120–200 ms) come from older Raspberry Pi camera pipelines**, not a published Pi 5 OpenCV fisheye-dewarp benchmark; measure your own build. - **The RealWear Navigator 500 comparison price is an estimate** — vendors sell quote-only; treat \~$2,300 as indicative, not authoritative. - **Certification and road-legality vary by jurisdiction and change over time** (e.g., EN 397 was revised in 2025). What is legal/compliant where you build may differ. - **Micro-OLED daylight legibility, combiner ghosting, and IMU magnetometer drift** are the most likely real-world disappointments; plan for a higher-nit panel, an anti-reflective combiner, and careful IMU siting away from motors/magnets. --- *This is community documentation provided as-is; prices are estimates; the builder is solely responsible for local code compliance, PPE/certification decisions, and safe practice.* --- ![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/07/image-12-1.png) ![The Means Initiative Logo](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/2026/04/MeansPNG-7.png) Content Provided by The Means Initiative --- [Heads Up Display Glasses DIY: Your Ultimate Guide to Building Smart EyewearA comprehensive guide to building your own Heads Up Display glasses from scratch, covering optics, electronics, software, and assembly for a truly personalized wearable tech experience.![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/icon/favicon-5f751113-f102-4831-a1a9-2f5adcd2b49c.png)INAIRSPACEwangfred![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/thumbnail/1_61340dfe-0c9c-480c-9a83-c270cd56321c-309d1e12-23e9-465a-b0f5-d565f81fc40e.webp)](https://inairspace.com/blogs/learn-with-inair/heads-up-display-glasses-diy-your-ultimate-guide-to-building-smart-eyewear?ref=datadeep.tech) [Adafruit 9-DOF Orientation IMU Fusion Breakout - BNO085 (BNO080)Here it is, the motion sensor you were looking for: the one that just gives you the directly usable information without requiring you to first consult with a PhD to learn the arcane arts of Sensor Fusion.![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/icon/favicon-83d77244-9ed3-4ac6-a045-418356540e52.png)BNO085 (BNO080)Adafruit![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/thumbnail/4754-00-74687d95-ed05-49a4-b025-57b56be86c6a.jpg)](https://shop.pimoroni.com/products/adafruit-9-dof-orientation-imu-fusion-breakout-bno085-bno080-stemma-qt-qwiic?variant=32272762437715&ref=datadeep.tech) [Jetson Orin Nano Super Developer KitThe most affordable generative AI supercomputer.![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/icon/favicon-744c084b-ba0b-401d-8845-b33cf03ef709.ico)NVIDIA![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/thumbnail/jetson-orin-nano-super-developer-kit-og-d2615b05-71a1-4d5e-9c45-03363d6cbce4.jpg)](https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/nano-super-developer-kit/?ref=datadeep.tech) [GoPro MAX 360 Action Camera (Waterproof + Stabilization)Shop the GoPro MAX 360 degree action camera. Single-lens HERO and dual-lens 360 camera. Capture immersive 360 footage in 6K! MAX out the radness.![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/icon/favicon-12b5a03a-a68b-4ae1-a8fd-4c0ca58cb06d.ico)![](https://storage.ghost.io/c/1d/fa/1dfa0703-59cd-42c7-a4f8-b16e218c2d7c/content/images/thumbnail/1-pdp-max-gallery-768-375-ada6d4c0-17fb-419c-a70e-507bd00eea36.png)](https://gopro.com/en/us/shop/cameras/learn/max/CHDHZ-203-master.html?ref=datadeep.tech)