Machine vision from SICK – cameras for measurement, defect detection and sorting

Machine vision is an automation technology that allows machines to "see" and analyze visual information, performing quality control, measurement, and identification tasks. SICK is one of the leading manufacturers of sensors and machine vision systems with over 60 years of experience. SICK offers a very wide range of 2D and 3D vision cameras for industrial automation.

Table of Contents

• What is machine vision?
• SICK machine vision camera range
• SICK machine vision camera selection table (2D / 3D)
• Basic functions of machine vision
• Simple and complex application examples
• Defect, quality and sorting detection
• Similarities with license plate recognition and people detection
• Traditional and deep learning approaches
• Advantages and ROI
• Future trends 2025

What is machine vision?

Machine vision is a technology that uses cameras, sensors, and computer vision algorithms to automatically inspect, analyze, and make decisions based on visual data. It is similar to human vision but operates with much greater precision, speed, and consistency. The system captures images, processes them with special algorithms, and makes decisions — for example, whether a product is of high quality or defective.

Machine vision systems consist of four main components:

1. Image acquisition (cameras and sensors)
2. Data transmission (Ethernet, USB)
3. Information extraction (edge detection, measurements, pattern recognition)
4. Decision-making (based on AI/ML algorithm analysis)

SICK machine vision camera range

SICK offers several camera series for different applications:

2D vision sensors and cameras

• Inspector83x series — AI-powered camera with deep learning capabilities that allows unskilled operators to quickly configure quality control, defect detection, and sorting by simply "teaching" the camera with examples. Capable of up to 15 inspections per second.
• InspectorP61x — ultra-compact 2D vision camera, ideal for confined spaces or mounting on a robot arm end-effector.
• InspectorP64x — programmable 2D camera with 1.7 megapixel resolution for operation at greater distances and high-speed tasks.

3D vision cameras

• Ranger3 series — high-performance 3D streaming cameras with up to 69 kHz profile acquisition speed, RGB color scanning up to 5120 pixels, and ROCC technology for outstanding 3D quality. Suitable for industrial inspection and measurement.
• Ruler3000 series — 3D cameras with extremely high scanning speed (up to 46 kHz) and the ability to work with dark materials, such as tire rubber.
• Visionary-T, Visionary-S, Visionary-B — “snapshot” 3D sensors for various applications: indoor (Time-of-Flight technology), stereo vision with structured illumination, or rugged sensors for outdoor conditions.
• TriSpectorP1000 and ScanningRuler — programmable 3D cameras with laser triangulation method for robot guidance and 3D measurements.

SICK machine vision camera selection table (2D / 3D)

The table shows some of the most popular SICK cameras.

Basic functions of machine vision

Machine vision systems perform four basic tasks:

1. Pattern Recognition — recognition of simple shapes, colors, barcodes, faces, fingerprints.
2. Visual Positioning — precise determination of an object's coordinates and angle.
3. Measuring — determination of geometric dimensions with high accuracy.
4. Appearance Inspection — defect detection: assembly errors, missing components, print defects, shape damage.

Simple and complex application examples

Simple examples:

• Presence check — whether all components are installed on the circuit board or included in the packaging.
• Dimension measurement — control of product height, width, or volume.
• Barcode and QR code reading — product traceability on production lines.
• Color sorting — sorting fruits and vegetables by ripeness/color.
• Simple text reading (OCR) — recognition of expiry dates, batch numbers on products.

Complex examples:

• AI-supported defect detection — detection of unpredictable defects with deep learning algorithms that can adapt to new types of defects.
• 3D shape analysis — reading tire sidewall markings and checking integrity in 3D profile using AI/ML algorithms.
• Foreign object detection (FOD/FOS) — 3D scanning of electric vehicle battery surfaces to detect foreign objects (tools, screws) that could cause a short circuit.
• Complex assembly verification — checking the correctness of multiple components in automotive manufacturing with calibrated 3D cameras.
• Robot control with 3D vision — “random bin picking”: locating parts and calculating optimal gripping positions in chaotic containers.

* AI - Artificial Intelligence
* ML - Machine Learning

Defect, quality and sorting detection

Machine vision systems are crucial in quality control:

Defect detection:Systems identify cracks, scratches, dents, color inconsistencies, missing parts, or incorrect assembly with more than 99% accuracy. AI systems such asInspector83xuse convolutional neural networks (CNNs) that analyze images pixel by pixel and classify defects by severity. Traditional methods compare images to a reference, but AI models can learn and adapt to even unforeseen defects.

Quality inspection:Cameras check label accuracy, seal integrity, barcode presence, and print quality. In the pharmaceutical industry, systems inspect tablet blister packs, confirming batch codes and completeness, processing hundreds of packages per minute.

Sorting by various criteria:

• By color— RGB cameras recognize ripened fruits or defects in food products.
• By shape— systems distinguish properly shaped products from broken or deformed ones.
• By size— automatic size classification on the conveyor belt.
• Combined sorting— systems simultaneously analyze color, shape, and size, sorting at speeds of up to 3 m/s.

Similarities with license plate recognition and people detection

The principle of machine vision is identical to that used in automatic number plate recognition (ANPR/LPR) and pedestrian detection on streets.

Number plate recognition (LPR/ANPR)

Uses optical character recognition (OCR) algorithms that identify and read letters and numbers on vehicle license plates in images. Modern deep learning-based systems can read them at various angles, in poor lighting conditions, and even damaged plates with more than 98% accuracy. They utilize similar CNN algorithms as industrial machine vision systems.

Human/pedestrian detection

Uses computer vision algorithms to identify and track pedestrians in images or video streams. The systems analyze human shape, movement, and behavior using deep learning models. This is critical for autonomous vehicles (collision avoidance), security monitoring, and smart city infrastructure.

Common principles

All these systems — whether SICK industrial cameras, LPR systems, or pedestrian detectors — use a similar technological foundation:

1. High-quality image acquisition using cameras and sensors.
2. Image pre-processing (normalization, noise reduction).
3. Feature extraction using traditional methods (SIFT, HOG) or deep learning models.
4. Classification or detection using AI algorithms.
5. Real-time decision making.

Traditional and deep learning approaches

Traditional machine vision

The engineer manually defines which features to look for (edges, colors, textures), and creates rules. This approach is simple, fast, and interpretable, but requires a lot of manual work and is less flexible for complex tasks.

Deep learning approach

The AI model automatically "learns" what to look for by analyzing example images ("end-to-end learning"). This requires larger amounts of data and computational resources, but achieves better results for complex tasks and can adapt to new problems. SICK Inspector83x cameras use exactly this approach—the operator simply shows the camera good and bad examples, and the camera learns to distinguish them on its own.

Advantages and ROI

Machine vision systems provide significant return on investment:

• Productivity increase:20–30% improvement in production efficiency; robots process up to 10,000 parts per hour.
• Cost reduction:~50% reduction in quality control labor costs; 20–40% savings in maintenance costs.
• Defect reduction:30% decrease in defect rates; inspection time reduced from 60 seconds to ~2.2 seconds.
• Downtime reduction:30–50% reduction in downtime.
• Safety improvements:40–60% reduction in safety incidents.

Future trends 2025

The development of machine vision continues with several key trends:

1. Generative AI and Vision Transformers (ViT)— better defect detection with less training data.
2. Edge AI— processing directly in the camera, without cloud latency.
3. Multimodal integration— combining vision, language, and audio data in a single system.
4. 3D vision and NeRF technology— photorealistic 3D reconstruction from 2D images.
5. Synthetic data and self-learning— model training without costly manual data labeling.

Machine vision technology has become indispensable in modern automation — from simple product presence verification to complex, AI-driven defect detection and 3D measurements. SICK cameras offer solutions for all these tasks, providing high accuracy, speed, and reliability in industrial environments.