Evaluating Jetson-Inference on NVIDIA Jetson Xavier NX with Sony FCB-ER8530

Using Custom Carrier Board with Inbuilt HDMI-to-MIPI CSI-2 Conversion

A complete evaluation of real-time AI inference on NVIDIA Jetson Xavier NX using the Sony FCB-ER8530 4K block camera and a custom carrier board featuring built-in HDMI-to-MIPI CSI-2 conversion. Includes performance benchmarks, latency results, and Jetson-Inference metrics.

---

Preview

This article explores integrating the Sony FCB-ER8530 4K block camera with the NVIDIA Jetson Xavier NX, using a custom-designed carrier board that includes inbuilt HDMI-to-MIPI CSI-2 conversion.

Although OPPILA offers a standalone HDMI-to-MIPI bridge board as a product,

👉 we are NOT using that bridge here.

Instead, we are using a custom Xavier NX carrier board where the HDMI-to-MIPI circuitry is already embedded on the board. This enables a direct 4K HDMI video feed from the FCB-ER8530 into the Jetson Xavier NX for high-performance AI inference using the Jetson-Inference framework.

---

Introduction

Edge AI continues to advance across defense surveillance, smart cities, industrial inspection, and autonomous systems. High-quality imaging paired with strong AI compute performance is essential.

The NVIDIA Jetson Xavier NX delivers up to 21 TOPS, making it ideal for real-time object detection and vision analytics. However, cameras like the Sony FCB-ER8530, which output 4K HDMI, cannot be directly connected to Xavier NX because:

Sony Camera Output → HDMI Jetson CSI Input → MIPI CSI-2

To solve this, we designed a custom Xavier NX carrier board with:

✔ Inbuilt HDMI → MIPI CSI-2 conversion

✔ No external HDMI-to-MIPI module required

✔ Direct high-speed CSI-2 input into Jetson

✔ Stable 4K@30fps performance

This article shares the hardware/software architecture and evaluates performance using Jetson-Inference.

---

Objective

To validate AI inference on Xavier NX using the Sony FCB-ER8530 via a custom HDMI-to-MIPI integrated carrier board, focusing on:

1. Reliable HDMI → MIPI camera input
2. Real-time AI object detection throughput
3. Per-frame and end-to-end latency
4. System-level stability

---

Hardware Architecture

Sony FCB-ER8530 — 4K LVDS Block Camera

Key Features:

1. 4K UHD 3840×2160 @ 30 FPS
2. 20× Optical Zoom + 12× Digital Zoom
3. LVDS Y/Pb/Pr 4:2:2 output
4. Ideal for long-range detection applications

Custom Xavier NX Carrier Board with Built-In LVDS-to-MIPI CSI-2 Conversion

This custom carrier board integrates all HDMI-to-MIPI circuitry internally.

---

Key Capabilities :

1. Accepts 4K HDMI directly from FCB-ER8530
2. Converts HDMI → 4-lane MIPI CSI-2 in real time
3. Low-latency internal design
4. High-bandwidth support for 4K@30fps input
5. Industrial, compact, and thermally optimized

This internal conversion eliminates the complexity of HDMI-to-MIPI bridges and ensures lower latency with better reliability.

Jetson Xavier NX Developer Kit

1. Volta GPU with 384 CUDA cores + 48 Tensor Cores
2. AI performance: 21 TOPS
3. JetPack 5.1.2 (L4T 35.4.1)
4. Dual MIPI CSI-2 camera inputs
5. Ideal for TensorRT accelerated inference

---

System Diagram

Block Flow

Connection Diagram

---

Software Stack

Tools & Frameworks Used

1. JetPack 5.1.2
2. Custom V4L2 camera driver for MIPI-CSI capture
3. Jetson-Inference (SSD-MobileNet-V2 TensorRT optimized)
4. OpenCV, Jetson-Utils, GStreamer pipelines
5. Python API for rapid prototyping

Inference Pipeline

1. HDMI Input from Sony camera
2. Internal conversion to MIPI CSI-2 by carrier board
3. Frame preprocessing
4. TensorRT inference (SSD-MobileNet-V2)
5. Post-processing & overlay output

Performance Evaluation

We performed inference using downscaled 1920×1080 frames from the 4K input.

Metric	Result
Inference Throughput	28–30 FPS (1080p)
TensorRT Latency	12–18 ms
End-to-End Latency	~45 ms
Thermal Behavior	Stable under passive cooling

---

Video Demo

🎥 https://youtu.be/Tu6TYtENLVs

Live inference using our custom carrier board + Sony FCB-ER8530.

Key Observations

1. HDMI → MIPI internal conversion produced very low latency
2. Excellent stability during extended testing
3. SSD-MobileNet-V2 offered strong performance for edge deployments
4. System can scale to higher resolution and more complex AI models

---

Conclusion

By integrating the Sony FCB-ER8530 4K camera with the NVIDIA Jetson Xavier NX using a custom-designed carrier board featuring inbuilt HDMI-to-MIPI conversion, we achieved:

1. Reliable, low-latency 4K camera input
2. Real-time AI inference at ~30 FPS
3. Strong thermal and operational stability
4. Zero need or external conversion hardware

This architecture is ideal for:

1. Defense & homeland security
2. Smart city monitoring
3. Industrial AI
4. Traffic and transport analytics
5. Long-range imaging & autonomous systems

---

FAQs