Technical System Architecture © Our-Eyes, Inc.

4D Spatial Capture & Playback Stage

This project establishes a real-time volumetric capture, processing, and multi-view presentation stage. The pipeline ingests live 3D space, encodes it into ultra-high-bandwidth 4D media codecs, and utilizes raw GPU power to calculate millions of light vectors for headset-free 3D viewing on a 65" light-field display.
[Volumetric Stage] ──> [Master Sync Hub] ──> [Capture Workstation] ──> [Storage System] ──> [65" Looking Glass] (ZED X Cameras) (GMSL2 Precision) (TouchDesigner Matrix) (NVMe RAID / Dual RTX) (8K Light-Field)
1. Sensor & Capture Layer (Volumetric Input)
Sensors

Multiple Stereolabs ZED X IP67 stereo depth cameras. Global shutter sensors prevent motion blur, which is critical for real-time spatial reconstruction.

Interconnect

GMSL2 (Gigabit Multimedia Serial Link) cabling links the cameras to the workstation via specialized capture cards. This maintains low latency and high bandwidth over distances up to 15 meters.

Temporal Alignment

The Master Sync Hub utilizes hardware trigger pins on the ZED X cameras. This ensures every sensor captures depth frames at the exact same microsecond, eliminating temporal tearing when stitching the volumetric mesh.

2. Compute & Encoding Layer (The Workstation Core)
The Matrix Core

TouchDesigner acts as the central ingestion engine. It pulls the raw depth maps and color feeds via the ZED SDK, aligning them into a unified 3D point cloud or signed distance field (SDF).

Render Multiplexing

TouchDesigner processes the stitched scene and structures it into a Quilt Matrix (the specific multi-view image grid format required by Looking Glass Factory).

Compression Pipeline

For real-time recording or optimized playback, the volumetric data is encoded using high-performance codecs:

HAP Q: Excellent for ultra-low CPU overhead during fast decoding, using the GPU to handle decompression textures.
NotchLC: Provides 10-bit color accuracy and clean alpha channels, minimizing color banding across the billion-color depth spectrum of the 65" display.

3. Storage & Processing Topology
GPU Infrastructure

Twin NVIDIA RTX 6000 Ada Generation GPUs. Packed with a combined 96GB of ECC GDDR6 VRAM, providing the massive frame buffer necessary to manipulate raw 8K textures and multi-view frame streams.

NVLink / PCIe Peer-to-Peer

Active distribution handles the computational load. One GPU drives heavy real-time camera stitching and geometry deformation shaders in TouchDesigner. The second GPU renders the final multi-view light field calculations.

Storage Array

A dedicated NVMe PCIe Gen 5 SSD RAID 0 array. Standard drives will choke on the data rate of uncompressed or lightly compressed NotchLC/HAP Q videos running at 8K spatial resolutions. A multi-drive RAID ensures sustained read/write speeds exceeding 14+ GB/s to keep playback at a smooth 60fps.

4. Presentation & Display Layer
The Canvas

The 65" Looking Glass Synapse Light Field Display. Powering this hardware requires immense computing power, as it renders up to 100 distinct spatial viewpoints simultaneously at an 8K base input resolution.

Optics & Illusion

The display projects up to 100 distinct perspectives simultaneously into a 53-degree viewing cone. Instead of flat pixels, it maps millions of directional light rays. Multiple viewers can walk around the physical display and see "around" the captured subject naturally without headsets.