Computer Vision Products & Solutions Guide
This guide presents EMG2’s products and solutions for computer vision: 3D cameras, camera modules and lenses, SBCs and box PCs, FPGA boards and video acquisition boards, monitors and panel PCs, software and hardware vision solutions, the NATvision multi-camera platform, and custom vision and AI development by our design office ABTICS. For each range, you will find the key features and strengths. Browse it below or download the PDF.
Booklet contents
Detecting a defective part, counting objects or people, recognizing and tracking an object, finding your way in space… Computer vision gives your systems the ability to see and analyze autonomously.
At EMG2, we supply the components needed to build these systems and advise you on choosing them: 3D cameras, camera modules and lenses, FPGA and video acquisition boards, SBCs, box PCs, GPGPUs, monitors and panel PCs. We also provide software solutions.
With our sister company ABTICS, a design office specializing in vision and RF, we go further: ABTICS works with you to design, develop and integrate custom, high-performance perception systems, from study through to integration, with the option of incorporating AI and deep learning.
The entire vision chain, one single contact
- 1. Capture: 3D cameras, camera modules, lenses
- 2. Acquire: video acquisition boards
- 3. Process: CPU, GPU, FPGA · SBC, box PC, GPGPU
- 4. Display & control: monitors, panel PCs
- 5. Leverage: 3D vision SDK, software & hardware solutions, NATvision
- Steps 1 to 4: end-to-end support by ABTICS
Our fields of application
- Industry & quality control: high-speed inspection, defect detection, sorting
- Logistics & warehousing: parcel and pallet dimensioning, counting objects
- Robotics & AMR: pick-and-place, palletizing, navigation, obstacle detection
- Security, retail & healthcare: face detection, people tracking and posture analysis
- Transport & railway: embedded computers certified E-Mark and EN 50155
- Defense & space: rugged computing for demanding defense and space applications
The right technology for your needs
In computer vision, there is no one-size-fits-all technology. The right choice depends on the answers to six questions:
- What must the system see? Presence, dimension, defect, face, posture… in 2D or in 3D?
- How fast? Production rate, maximum acceptable latency
- How many cameras? A single stream, or several synchronized cameras
- In what environment? Temperature, dust, humidity, shocks, vibrations
- What integration constraints? Size, power consumption, power supply, connectors
- What volume? Prototype, small series or industrial production
The table below summarizes the main technology options, their typical uses and the solutions we offer for each. They are not equivalent alternatives: CPU, GPU and FPGA are computing resources, MicroTCA is a modular hardware architecture that hosts them, and AI is an algorithmic approach that can run on several of these resources.
| Technology | Category | Typical use cases | Principle | Typical strengths | EMG2 & ABTICS solutions |
|---|---|---|---|---|---|
| CPU | Computing resources: what executes the processing | Presence checks, barcode reading, color-threshold sorting | Sequential processing, fixed algorithm | Simple to implement; often the most economical choice for light workloads and a limited number of streams | Camera modules, 3D cameras + 3D vision SDK, SBCs, box PCs, panel PCs, monitors |
| GPU | Computing resources: what executes the processing | Simultaneous analysis of several video streams on the same line | Same programming logic as the CPU, but massively parallel | High throughput on parallel workloads (the number of streams depends on resolution, frame rate and model); broad ecosystem and accessible frameworks | GPGPUs, SBCs, box PCs, panel PCs |
| FPGA | Computing resources: what executes the processing | Very high-speed inspection, real-time embedded processing, object tracking | Reconfigurable circuit executing the processing directly in hardware | Very low, deterministic latency; high energy efficiency on fixed, well-defined pipelines; hardware parallelism | FPGA boards + video acquisition boards |
| MTCA | Hardware architecture: how the boards are assembled | High-end synchronized multi-camera systems (1 to 12 boards) | Modular industrial architecture with interchangeable AMC boards and a MicroTCA Carrier Hub (MCH) | High availability with built-in fault management and redundancy; high density (up to 12 AMC boards per chassis) | NATvision solution |
| AI | Algorithmic approach: runs on CPU, GPU, FPGA or NPU | Face detection, detection of undefined anomalies, high visual variability | Model trained on representative data | Handles cases too variable for fixed rules; results depend on the model and on the training data, and can improve when the model is retrained | 3D vision SDK; custom edge-AI development, up to model training, with ABTICS |
These technologies are often combined: for example, FPGA preprocessing upstream of AI inference, or a GPU running the same model on several streams in parallel. Our role at EMG2 is to analyze your needs and help you choose the most suitable products and solutions. With ABTICS, we can also carry out a comparative analysis of these architectures or of different software packages, to determine what best meets your needs.
Key features
- Three 3D vision technologies: active stereo, time of flight (ToF) and structured light
- Unified development kit; ROS/ROS2, NVIDIA Isaac, GenICam, Halcon, OpenNI, OpenCV and AI platforms (Jetson, Ambarella, SiMa…), depending on the model
- Industrial active stereo models: GigE with PoE, secure M12 connectors, IP67, built-in heat sink, 6H scratch-resistant tempered glass
- On these models, the session is restored in < 15 s after a disconnection
Active stereo
Uses two lenses from two different viewpoints to generate an accurate 3D image, ideal for object recognition or robot guidance.
Best suited to: medium range, robot and AMR guidance. Accuracy decreases with distance.
Specifications: LIPSedge™ AE400/AE430 and AE450/AE470
- Depth range: 0.45–4 m (AE400/430); 0.6–6 m (AE450/470)
- Depth: 1280 × 720 @30 fps; 640 × 480 @60 fps (AE400/430)
- FoV (H×V×D): 65×40×72° (AE400/430); 87×58×95° (AE450/470)
- Depth accuracy: ≤ 2% up to 2 m (AE400/430) or 4 m (AE450/470), over 80% of FoV
- Edge-AI SoC (AE430/470): Arm Cortex-A73/A53, 5 TOPS
- 6-axis IMU (AE470) · Connection: GigE, PoE
ToF (time of flight)
Projects modulated light and measures the reflected signal to determine distance: fast, real-time depth for people tracking, logistics or navigation applications.
Best suited to: fast depth over a wide field; sensitive to strong sunlight and shiny surfaces.
Specifications: LIPSedge™ DL and M3 LED
- Depth range: 0.16–4 m (DL); 0.2–4 m (M3 LED)
- 850 nm: 2 edge-emitting lasers (EEL) (DL) or 2 LEDs (M3 LED)
- Depth @30 fps: 320 × 240 native, 640 × 480 output (DL); 80 × 60 (M3 LED)
- FoV (H×V×D): 74.2×58.1×88° (DL); 74.1×57.5×92° (M3 LED)
- Depth accuracy: ≤ 1% (DL); ≤ 3% (M3 LED)
- Connection: USB
Structured light
Projects a pattern (stripes, grids or dots) and measures its deformation to generate a highly accurate point cloud, for example for 3D reconstruction.
Best suited to: close-range, high-detail measurement of static objects; sensitive to strong ambient light.
Specifications: LIPSedge™ L210u (module) / L215u (enclosure)
- Depth range: 0.4–1.2 m
- VCSEL 940 nm laser
- Depth: 1280 × 800 @6 fps or 640 × 480 @25 fps
- FoV: 50.7 × 74.2° (H × V)
- Depth accuracy: 0.3% at 100 cm
- Connection: USB 2.0 (Micro USB)
Key features
- Excellent image quality, essential for critical applications
- Compactness: easy integration into space-constrained devices without compromising performance
- IP cores: ready-made processing blocks for your FPGA or chip designs, reducing development time
- Solutions designed for imaging systems in many fields: security, surveillance, industry and medical
Smart camera modules
- Renesas RZ/V2L processor (dual-core Arm Cortex-A55 at 1.2 GHz + Cortex-M33)
- Distortion correction, viewpoint conversion and panoramic conversion
- C/C++ applications can be developed and integrated directly on the camera for series production
- Sample programs for Windows (C#) included for USB and LAN communications
- Optional interface boards: LAN, GPIO, I²C, SPI and SD card storage
- Development kit available
Industrial camera modules
- Interfaces, depending on the model: USB, LAN, MIPI, GigE (PoE), NTSC
- Resolution: from 0.3 MP (640 × 480) to 2 MP (1920 × 1080)
- Sensors: 1/4-inch to 1/2.7-inch color CMOS with rolling shutter (global shutter on some models)
- Frame rate: 5 to 30 fps
- Input voltage: 5 V; power: 0.9 to 2.4 W
- Operating temperature: −10 °C to +60 °C
Lenses
- Supported sensor sizes: 1/2 or 1/4 inch
- Mount: M12 or M9
- Focal lengths: 0.82 to 9.3 mm; apertures: F/2 to F/2.8
- Fields of view: horizontal 43° to 194°, vertical 32.6° to 156°, diagonal 52° to 206°
- Pixel size: 1.8 to 5.6 µm; flange distance: 0.9 to 5.02 mm
- Optical distortion down to −1% on low-distortion models
- Versions with infrared transmission available
IP cores
Hardware processing blocks integrated into your own FPGA or chip design, rather than separate cameras:
- Two families of JPEG cores: KJN-F (for FPGA: compact, high-speed, low power) and KJN-S (lossless compression)
- Fast, low-latency JPEG encoding, designed to minimize the number of logic gates and power consumption
- MIPI cores: direct interface with image sensors and processors
- ISP (image signal processor) cores: color interpolation, white balance and noise reduction
Key features
- Wide choice of high-performance processors and compute modules
- Multiple form factors, COTS or custom
- Complete rugged and space-grade systems possible, with Ethernet switch, memory, interface boards…
- Carrier board: hosts a compute module (e.g. NVIDIA Jetson) and provides its interfaces
- SBC: a complete computer on a single board
- Box PC: a ready-to-use computer in its enclosure
- GPGPU: a system built around a GPU for parallel processing
Carrier boards
- For Qualcomm Dragonwing or NVIDIA Jetson modules
- By model: GbE/10G RJ45, Wi-Fi 6/6E, Bluetooth, M.2 Key B for 4G/5G
- Camera/video: GMSL2, FPD-Link III, MIPI CSI-2; HDMI/DP outputs up to 4K
- From −40 °C up to +85 °C, depending on the model
Military & space SBCs
- Intel processors, 3U VPX format
- Up to 64 GB DDR4 with ECC
- With or without XMC slot; PCIe Gen3 and/or Gen4
- −40 °C to +85 °C; resistant to shocks and vibrations
- Radiation-tolerant versions (> 100 krad (Si))
Box PCs
Edge AI
- NVIDIA Jetson, Qualcomm Dragonwing, or Intel Core i3–i9 (Gen 10/14) + NVIDIA RTX Embedded Ada 2000/3500/5000
- PoE Ethernet, GPIO, USB, UART, SPI, I²C, CAN bus; Wi-Fi or 4G/5G SIM expansion
- M.2 port, HDMI output
- 9–60 V DC (backup battery option)
- −40 °C to +70 °C
Industrial & in-vehicle computers
- Intel Atom or Core i3/i5/i7 CPU (Gen 11), or AMD Ryzen R2312
- SODIMM DDR4 up to 64 GB
- Ethernet, GPIO, USB; M.2 for Wi-Fi or 4G/5G SIM, and for NVMe or SATA SSD
- 9–60 V DC; −40 °C to +70 °C
- E-Mark and EN 50155 certified models
- DEEPX DX-M1 NPU option
Rugged & space-grade computers
- 2 to 8 slots; 3U CompactPCI, 3U/6U VPX/VME
- Intel processors
- −46 °C to +60 °C
- Compliant with MIL-STD-461, -462, -704, -1275
- Resistant to extreme environments (altitude, humidity, shocks, vibrations)
AI GPGPU
- Form factors: SFF, ultra-SFF, 3U VPX, based on NVIDIA® Jetson modules
- Top configuration (Jetson AGX Orin Industrial): 2,048 CUDA cores, 64 Tensor cores, up to 248 TOPS (INT8)
- −40 °C to +55 °C; MIL-STD-461, -704, -1275 and IP66 by model; military and space versions
FPGA boards
- AMC* form factor
- Zynq UltraScale+ version: Xilinx (AMD) ZU7 or ZU11 MPSoC with quad-core Arm Cortex-A53; 8 GB DDR4 for the processing system (PS) + 8 GB DDR4 for the programmable logic (PL); MicroSD card with Linux BSP
- Arria 10 version: Altera (Intel) Arria 10 FPGA
- High Pin Count (HPC) FMC site: one FMC, or two on the full-size version
Video acquisition boards
FMC-PoE (for FPGA AMC boards)
- 4 front RJ45 Gigabit Ethernet ports for cameras, IEEE 802.3af PoE
- GigE Vision support; SFP version (10 GigE) available
- FMC VITA 57.1 High Pin Count (HPC) connector, with 4× SGMII and I²C to the carrier board
- Highly efficient power conversion
PCIe
- PCIe Gen2 x4, low-profile form factor available
- 4 HDMI channels, input up to 4096 × 2160 @30 fps
- 8/10-bit color, YUV444/422/420 and RGB formats
- Built-in hardware video processing (scaling, de-interlacing, color space conversion)
- Windows 7/8.1/10 and Linux compatible; FCC/CE certification
M.2
- PCIe Gen2 x2 host interface, M.2 (22 × 80 mm)
- HDMI 1.4a / DVI-D input, 4096 × 2160 @30 fps
- Hardware video processing: scaling, de-interlacing, color space conversion
- 8/10-bit color, YUV444/422/420 and RGB
- Windows / Linux (V4L2); FCC/CE
Mini PCIe
- Mini PCI Express host interface (PCIe Gen2 x1), compact 50.80 × 30 mm form factor
- HDMI 1.4a input, 1920 × 1080 @60 fps capture
- 8/10-bit color, 4:4:4 RGB and YUV
- Built-in video processing (scaling, de-interlacing, frame rate conversion)
- Windows 7/8.1/10, DirectShow, multi-board
USB
- USB 2.0; composite, S-Video and audio L/R inputs; NTSC and PAL compatible
- Up to 720 × 480 @60i (NTSC) and 720 × 576 @50i (PAL)
- Multi-language SDK: VB, VC, Delphi, C#, VB.NET (generic DLL framework)
- 5 V, 2.06 A power supply
- Windows 7/8.1/10 compatible
* Advanced Mezzanine Card: a compact modular board inserted into a MicroTCA (MTCA) chassis to add a specific function (computing, inputs/outputs, storage, etc.)
Key features
- High-brightness, sunlight-readable displays and flexible mounting options to meet a wide variety of installation requirements
- Versatile I/O interfaces and international certifications (E-Mark, FCC…), depending on the model
- Suited to industrial automation, intelligent transportation, self-service kiosks and critical control centers where precision and durability are essential
- High-performance processing and highly sensitive touchscreens
- Rugged models with metal enclosures and reinforced tempered glass, resistant to severe shocks and vibrations (MIL-STD-810H)
- Seamless integration and long-term scalability
- Fanless platforms for silent operation
Monitors
- 7 or 10.1-inch capacitive touchscreen
- Resolution 1280 × 800; brightness 560 or 1000 nits
- HDMI and DVI ports
- Stereo speakers
- 9 V to 48 V DC power supply
- Operating temperature: −20 °C to +60 °C
Panel PCs
- Intel Atom, NXP i.MX 8X or NVIDIA Jetson Orin processor
- 10.1 or 15.6-inch capacitive touchscreen
- Resolution 1280 × 800, brightness 1000 nits
- Ethernet, GPIO, USB, CAN bus; M.2 expansion for Wi-Fi or 4G/5G SIM; M.2 for NVMe or SATA SSD
- 9 V to 60 V DC power supply (backup battery option)
- Operating temperature: −25 °C to +60 °C; IP66
3D vision software, SDK & solutions
LIPSARM™: robotic-arm solution
- 3D vision-guided robotic arm system, based on NVIDIA Isaac Manipulator
- For industrial automation: pick-and-place, palletizing and bin-picking
LIPSAMR™: AMR solution
- System designed specifically for autonomous mobile robots (AMR)
- For path optimization, factory mapping, and obstacle and pedestrian detection
LIPSMetric™: hardware + software solution
- Stationary 3D measurement of logistics and warehouse parcels up to 1 m, with ±1 cm accuracy on box-shaped parcels
- Pallet dimensioning version: measurement in about 1 s, able to detect and measure stacked pallets
LIPSense™: SDK
- Real-time 3D body-pose tracking based on skeleton keypoint detection
- Up to 32 skeletal joints per person, and up to 43 people tracked simultaneously
- For gesture-based and touchless interfaces, motion and posture analysis
Camera-based face detection (hardware & software solution)
- Pre-installed algorithms for face detection, gender and age estimation, and dwell time measurement (cumulative time a person remains in the field of view)
- Detects up to 10 faces simultaneously in a single image; detection range of 2 to 5 m (adjustable)
- 1/4-inch color CMOS sensor (rolling shutter)
- Resolution: 0.3 megapixel (640 × 480)
- Wired LAN: 100BASE-TX / 10BASE-T with PoE support
- USB Video Class (UVC) output
If you need:
- Several high-end cameras, with a total data rate above 10 Gb/s
- A modular vision architecture
- Time synchronization via PTP*
- Deterministic, low-latency processing
- An SDK and application examples
… NATvision is made for you!
NATvision is an integrated hardware and software platform based on the modular MicroTCA (MTCA) standard, for multi-camera systems that require FPGA preprocessing: a coherent base onto which your application building blocks are added.
- Cameras: GigE Vision and other standards
- Acquisition: FMC boards (e.g. FMC-PoE)
- FPGA preprocessing: AMC FPGA boards
- CPU / GPU processing: host computing
- Application: your software building blocks
- Time synchronization: IEEE 1588 PTP*
What’s inside the NATvision solution?
- A scalable architecture from 1 to 12 MTCA boards, based on the PICMG MicroTCA.0 Rev. 3 standard
- FPGA boards for (pre)processing, supplied with: a camera stack (binary code), a GenICam configuration tool (source code), a sample application and its CPU driver (source code)
- GigE Vision stack available as an option (source code)
- FMC acquisition boards for other camera standards
- IEEE 1588 PTP* time synchronization
Use case: warp transformation (real-time geometric correction)
- Many camera systems produce distorted images (perspective, wide-angle/fisheye, rolling shutter…), which reduces the accuracy of measurement, detection and analysis.
- Warp transformation recalculates the position of each pixel. Applied on the fly in the NATvision FPGA, a transformation matrix corrects perspective; lens or rolling-shutter distortion requires a suitable correction model.
- Warping demands computing power and memory bandwidth, often in real time: the FPGA handles it with massively parallel pipeline processing, at very low latency, without loading the CPU/GPU.
- A dynamic matrix, generated externally (CPU/GUI) and applied instantly by the FPGA, keeps the correction up to date in real time, even when the camera moves or the viewing angle changes.
- Example result: with the NATvision GigE Vision stack on the FPGA, estimated pipeline latency of 6.3 ms for a 1280 × 720 stream at 55 fps in this configuration.
* Precision Time Protocol: protocol for high-precision time synchronization between network devices
Our design office ABTICS designs, develops, integrates and industrializes custom, high-performance perception systems, from sensor selection through to deployment, drawing as needed on advanced algorithms, edge computing or complex architectures (CPU, GPU, FPGA).
Key features
- Expertise in vision, image processing, AI and related tools, with a multi-sensor “next-generation vision” approach
- Mastery of various processing cores (CPU, GPU, FPGA…), edge computing, advanced algorithms and various sensors (visible, IR, LiDAR…)
- End-to-end support: feasibility study, proof of concept (PoC), development of hardware and software building blocks, system integration
- Strong capacity for innovation and adaptation to specific technical or application constraints
What’s inside our solutions?
- AI/ML model generation, training and inference
- Advanced computer vision algorithms
- High-bandwidth simultaneous image capture
- 3D tracking, recognition and spatial representation
- Timing and synchronization management
- Open sensor fusion, camera-agnostic
- Development on specific CPU, GPU and FPGA platforms
- Training and upskilling of the client’s teams
Development examples
Multi-sensor 3D detection & tracking
- Multi-camera solution for 3D detection and tracking of objects or people
- Acquisition of multiple video streams from various sensors
- Simultaneous tracking of multiple objects (identical or not)
- Real-time analysis and traceability
- Reduced response time, flexible deployment
Semi-automatic annotation tool
- Quasi-automatic generation of accurately annotated datasets
- Capture and processing of images from multiple streams and sensors
- Intuitive interface for editing images and annotations
- Image, annotation and dataset management, model training assistance
Human posture analysis
- 3D reconstruction and interactive visualization of human posture (joints, anatomical landmarks, visible or not) within an accurate digital model
- Synchronized acquisition of video streams via a calibrated multi-camera system
- Real-time estimation of up to 133 anatomical keypoints
Any questions about one of these ranges? Contact our teams: we will help you choose the right product for your project.
Also discover all our vision solutions.