Understanding computer and machine vision
delivery ■ Preferred depth cameras for humanoids 1. Stereo vision cameras ■ Stable depth perception, adaptability to varied
autonomous navigation, industrial automation, medical imaging, and augmented/virtual reality (AR/VR) due to their compact form factor, real-time processing capabilities, and extensive software support. Intel RealSense Depth Cameras leverage a combination of stereo vision, active infrared (IR) projection, and structured light to compute high-fidelity depth maps in real time. The stereo IR cameras capture left and right image pairs, and an onboard Intel Depth Sensing ASIC computes depth using disparity matching algorithms. The active IR pattern projector improves accuracy in low-texture environments by adding depth cues where natural visual features are sparse. Intel’s D400 series are a popular choice, for example: ■ D415 – features a rolling shutter with a narrow field of view, suited for applications requiring precise object scanning ■ D421 - module brings advanced depth-sensing technology to a wider audience at an affordable price point ■ D435/D435i – utilizes a global shutter, making it ideal for fast- moving objects in robotics and automation ■ D455 – offers an extended baseline (95mm), improving depth accuracy for mid-range applications. These cameras see use anywhere from autonomous mobile robots
(AMRs) and drones, 3D scanning and volumetric measurements, medical imaging, or even biometric security measures. Interfaces in computer and machine vision The efficiency and scalability of machine vision systems depend on the interfaces used to transfer images data between cameras, processing units, and control systems. Different applications require specific connectivity solutions, balancing bandwidth, latency, power efficiency, and environmental robustness. The most widely used interfaces in machine/computer vision are GMSL/FAKRA, USB/MIPI CSI, Ethernet, and Power over Ethernet (PoE). GMSL/FAKRA Gigabit Multimedia Serial Link (GMSL) is a high-speed serial interface designed for automotive and industrial vision applications. It supports long-distance, high- bandwidth video transmission with low latency, making it ideal for autonomous vehicles, ADAS (Advanced Driver Assistance Systems), and robotic vision. GMSL operates over FAKRA connectors, which provide rugged, shielded connections suitable for harsh environments. GMSL can achieve data rates of up to 6Gbps per link while maintaining low latency, ensuring
Structured light excels in short-range precision, stereo vision provides passive depth estimation, ToF balances real-time performance with moderate range, and LiDAR leads in long-range accuracy.
reconstruction which is crucial for applications such as autonomous navigation, mapping, and industrial automation. Technical advantages: ■ Superior range (tens to hundreds of meters) and depth accuracy ■ Robust performance in diverse lighting conditions, including direct sunlight ■ Essential for applications like autonomous vehicle perception, aerial mapping, and infrastructure inspection Challenges: ■ High cost and power consumption compared to other depth sensing methods ■ Requires substantial data processing for real-time operation, often necessitating hardware acceleration (e.g., FPGAs, GPUs) ■ Performance can be affected by rain, fog, or highly specular surfaces Each depth sensing topology offers distinct advantages and limitations depending on the application. Structured light excels in short-range precision, stereo vision provides passive depth estimation, ToF balances real-time performance with moderate range, and LiDAR leads in long-range accuracy. The choice of technology depends on environmental constraints, computational resources, and the precision required for machine vision applications.
lighting conditions, multi- camera coordination and 360° view, low interference and high compatibility
1. Stereo vision cameras ■ Objects in motion,
Orbbec cameras implementing mainstream 3D technologies ■ Monocular Structured Light: Astra Mini Pro, Astra 2 ■ Stereo vision structured light camera: Gemini 330 series, Gemini 2 series (Gemini 2, Gemini 2 L, Gemini 2 XL) ■ TOF: Femto series (Femto Bolt, Femto Mega, Femto Mega I) Orbbec Gemini 330 series stereo vision 3D cameras integrate two infrared imaging modules, a laser diode module (LDM) for infrared speckle pattern projection, an RGB imaging module, a depth engine processor (MX6800), an image signal processor (ISP), and an inertial measurement unit (IMU). The LDM emits infrared speckle patterns onto the target scene, while the dual infrared imaging modules capture synchronized images from distinct viewpoints. The depth engine processes these images using advanced depth reconstruction algorithms to generate a high-precision depth map of the scene. Typical applications: ■ Preferred cameras for robotic arm applications
2. ToF cameras
require good accuracy at short distances, Varied/ un-controlled lighting conditions, Multiple cameras with shared FoV ■ Example: bin-picking
■ High precision depth sensing, real-time
performance, adaptability to complex lighting, multi- target detection and tracking
2. ToF cameras
■ Others 1. Structured light cameras: ■ Facial recognition (e.g. facial payment kiosks)
■ Stationary objects, require high-fidelity edge definition, Require good accuracy at medium-to- long distances, Controlled lighting conditions ■ Example: palletization/de- palletization
■ 3D scanning (e.g. body part scanning, object scanning)
Intel RealSense depth and tracking cameras Intel RealSense cameras have become a go-to product within the machine and computer vision space, offering engineers high- precision depth perception and positional tracking. These solutions are widely adopted in robotics,
3. Structured light cameras
■ Stationary objects, require highest accuracy at short- to-medium distances, controlled lighting conditions ■ Example: defect detection
■ Preferred depth cameras for AMR applications 1. Stereo vision cameras
■ Require stable depth maps while in motion, operate in varied lighting conditions, often require multiple cameras for 360° view, multi-camera interference ■ Example: pallet/tote- moving, forklifts, cleaning,
Figure 9: Intel RealSense D435f depth camera Credit: Intel Corporation
we get technical
46
47
Powered by FlippingBook