Dual Camera to UVC Bridge for Hyperspectral Imaging
An innovative startup pioneering hyperspectral imaging partnered with tinyVision.ai to develop a dual-camera interface that consolidates scene and spectral sensor streams into a single USB3 UVC output. The objective was to streamline high-speed, synchronized data acquisition for their hyperspectral systems, using a modular, scalable architecture.
The collaboration led to a proof-of-concept (PoC) and follow-on production-ready platform based on tinyVision.ai's tinyCLUNX33 FPGA SoM, enabling precise sensor synchronization, power-efficient design, and low-latency streaming—all essential for spectral data fidelity.
Client Background
The company operates in the high-performance imaging domain, developing hyperspectral sensors that deliver detailed optical data across a wide spectrum. Their solutions are used in scientific research, precision agriculture, and industrial inspection. Given the data-intensive nature of hyperspectral imaging, the client sought a custom interface solution to consolidate multi-sensor output into a unified, host-friendly stream.
The Challenge
The technical challenge centered on aggregating video streams from two 5MP sensors (RGB and monochrome) into a single UVC interface while maintaining precise synchronization. The key constraints included:
-
Sensor synchronization: Required accurate alignment of sensor frames without introducing large RAM buffers that lead to latency.
-
Power and clock integration: Needed unified clock distribution and power sequencing for dual-camera operation.
-
Custom interface: Use of MIPI CSI-2 protocols required specialized adapter design. Off-the-shelf solutions lacked support for sensor-level synchronization and real-time, low-latency UVC aggregation.
Our Solution
tinyVision.ai implemented a custom MIPI-to-UVC bridge using its tinyCLUNX33 SoM featuring the Lattice CrossLinkU-NX33 FPGA. The design included:
-
Dual MIPI interfaces: 2-lane support per sensor with synchronized data capture
-
I2C MUX architecture: Enabled independent sensor gain/exposure control
-
Custom RTL pipeline: Pixel pass-through design with minimal latency and no ISP dependency
-
Synchronization logic and image stacking support: provides perfectly synchronized images to the host image processing stack.
-
Standard UVC output: Data streamed to host as a stacked frame for simplified host processing without requiring custom drivers.
For PoC, dual MIPI input tinyVision.ai RPi adapter boards were leveraged; for production, a new dual-camera board was designed incorporating power, clock, and sync circuitry.
Development Process
The project was executed in three structured phases:
Phase 1: PoC and Adapter Integration
- Verified dual-sensor input with off-the-shelf adapters
- Aggregated 1080p streams into UVC output
- Demonstrated synchronization without frame buffering
Phase 2: Production-Grade Board and Drivers
- Designed and assembled dual-camera boards with 2-lane MIPI per sensor
- Integrated I2C expander, LED driver (Zephyr), and clock routing
Throughout, tinyVision.ai maintained tight collaboration via design reviews, communication, and iterative testing.
Results & Impact
Synchronized streaming of dual-camera data to a host over USB3 in real-time
Enables future hyperspectral decoding directly on FPGA in real-time
Technologies Used
- FPGA: Lattice CrossLinkU-NX33 on tinyCLUNX33 SoM
- Sensors: RGB and mono
- Interfaces: 2-lane MIPI CSI-2, USB 3.2 gen 1 UVC
- Firmware: Zephyr RTOS, custom RTL (MIPI2UVC), I2C MUX, sensor drivers
- Development Tools: Lattice Radiant
The tinyCLUNX platform was selected for its high-speed I/O support, minimal power footprint, and proven MIPI2UVC IP.
Conclusion / Key Takeaways
The dual-camera bridge solution developed by tinyVision.ai enables the client to integrate synchronized, high-resolution spectral and scene imaging into a compact, reliable USB3 interface. With a future-ready architecture, driver support, and RTL pipelines tailored for spectral workloads, this project highlights tinyVision.ai's ability to rapidly deliver production-capable, high-performance imaging interfaces.
Looking to accelerate your embedded vision project? Contact our engineering team to discuss your requirements.
Executive Summary
An innovative startup pioneering hyperspectral imaging partnered with tinyVision.ai to build a dual-camera UVC bridge that supports real-time hyperspectral imaging. Leveraging the tinyCLUNX33 SoM and Lattice CrossLinkU-NX33 FPGA, the solution enables synchronized data from two sensors to be streamed via USB3 using custom RTL and Zephyr drivers. The project spanned PoC, custom board design, and FPGA integration—paving the way for FPGA-based decoding in future product phases.
Connect us