From March 11–13, at Embedded World 2025, UNION IMAGE officially unveiled its first self-developed stereo depth camera, marking a major milestone in the company’s 3D vision technology roadmap.

First In-House Stereo Depth Camera Developed by UNION IMAGE

Backed by a strong global supply chain and a lean manufacturing system, UNION IMAGE integrates AIbase-driven product development deeply into the entire design and production process, ensuring fast iteration, stable quality, and cost efficiency.

With advanced intelligent optimization algorithms and strong cost-control capabilities, UNION IMAGE delivers a high-performance, cost-effective depth camera, providing accurate, efficient, and reliable solutions for intelligent vision applications.

Product Overview

The stereo depth camera is equipped with dual image sensors, featuring 1/4-inch 1.3MP CMOS sensors with a resolution of 1280 × 1080.

With a 2.7 µm × 2.7 µm pixel size and electronically synchronized global shutters, the camera ensures precise and consistent image capture.

Output Capabilities:

2 × 1280 × 1080 @ 15 fps

2 × 640 × 480 @ 15 / 30 fps (2×2 binning)

Output formats: Point Cloud, Depth Map, RGB Image

Optical & Imaging Performance:

Maximum SNR: 38.3 dB

Dynamic range: 64.6 dB

Sensitivity: 1900 mV/Lux-sec

Baseline: 63 mm

Focal length: 1.75 mm

Field of View: 115° (H) × 96° (V) × 157° (D)

Aperture: F2.0

TV distortion: < 20%

These specifications ensure stable, high-quality image output even in low-light environments and complex scenes.

High-Precision Sensor Fusion

In addition to powerful imaging sensors, the camera integrates a high-performance IMU and magnetometer:

Accelerometer:
±8G range, 0.244 mg resolution, noise density 3.2 mg

Gyroscope:
±1000 dps range, 0.03 dps resolution, noise density 0.16 dps
Sensitivity error within ±0.4%, output rate up to 400 Hz

Magnetometer:
Z-axis ±2500 µT, X/Y-axis ±1300 µT
Resolution 0.3 µT, output rate 50 Hz

Within a testing distance of 1 cm to 300 cm, depth accuracy reaches up to 1%, enabling comprehensive and precise environmental perception through multi-sensor fusion.

AI Large Model at the Core

The key highlight of this stereo depth camera lies in its AI-powered depth perception model:

Strong Adaptability to Low-Texture Scenes
Deep learning–based feature extraction enables robust depth estimation in challenging scenarios such as white walls, skies, and weak-texture surfaces, where traditional algorithms often fail.

Improved Accuracy and Robustness
Advanced AI algorithms, including representation-guided matching and optimized loss functions, significantly reduce depth errors and generate denser, more accurate point clouds.

Optimized Dynamic Scene Processing

Enhanced robustness against motion blur, multi-path interference, and complex lighting conditions.

Continuous Evolution Ahead

The UNION IMAGE self-developed stereo depth camera series will continue to evolve alongside advances in AI, moving toward smaller form factors, lower cost, higher performance, and greater intelligence.
The series will support customized development and scenario-specific optimization, adapting to diverse customer requirements.

Stay tuned.

AI + Vision at Embedded World 2025

Embedded World 2025 brought together nearly 1,000 exhibitors and over 32,000 professional visitors worldwide, serving as a core platform for embedded systems, Industry 4.0 technologies, and industrial innovation.

During the event, UNION IMAGE showcased its expertise in AIoT vision solutions, attracting strong interest and active engagement from industry partners.

Its camera modules, vision solutions, and complete imaging devices demonstrated the ability to rapidly match solutions to specific industries and application scenarios.

Star Products on Display

Unimage Pi Series

The Unimage Pi Series is a highly integrated intelligent vision solution designed for AIoT applications.
Powered by dual-core RISC-V processors, it offers multiple models and flexible memory configurations.

With rich interfaces supporting industrial control and multi-device expansion, and nine built-in high-precision vision models, it covers applications including security, transportation, health monitoring, and human–machine interaction.
Balancing high performance and easy deployment, Unimage Pi delivers stable and efficient edge computing power for AI-driven vision applications.

Dynamic Vision Sensor (DVS)

Based on bio-inspired retina principles, the DVS supports spatiotemporal event-based vision analysis, enabling high-speed motion trajectory reconstruction, behavior analysis, and anomaly detection.

It is widely applied in intelligent interaction, autonomous driving, medical monitoring, and Industrial IoT, demonstrating exceptional real-time performance, energy efficiency, and environmental adaptability.

AI Camera

The AI Camera is a high-performance intelligent vision device with built-in deep learning models capable of real-time detection of 17 human keypoints.

It supports motion recognition and posture analysis, and is widely used in sports science, rehabilitation training, medical assessment, human–computer interaction, security, and entertainment.
The solution delivers stable, high-precision motion capture and real-time analysis, even in complex environments.

Multispectral Camera

The Multispectral Camera utilizes high-precision spectral separation technology and is widely applied in:

Industrial inspection: PCB micro-defect detection, food sorting, color difference control

Agricultural monitoring: Crop 3D modeling, pest and disease early warning

Combined with multispectral fusion algorithms, it provides highly reliable vision solutions for smart manufacturing and precision agriculture.