Components · Software
Software turns images into decisions. A good vision application combines the right tools — locating, measuring, reading, inspecting, deep learning — with robust communication, clear operator screens and traceable data.
Phần mềm biến hình ảnh thành quyết định: định vị, đo, đọc mã, kiểm tra lỗi, học sâu — kèm giao tiếp ổn định, giao diện vận hành rõ ràng và dữ liệu truy xuất được.
How it works
Almost every application follows the same steps: acquire the image, prepare it (filters, calibration), locate the part, run the inspection, measurement or reading tools, decide pass or fail, send the result to the line and log images and data.
Rule-based tools are precise, fast and explainable — ideal for measurement, code reading and well-defined features. Deep learning learns from labeled images and excels where defects vary in shape and appearance or backgrounds are complex.
In practice, the best systems are often hybrid: rule-based tools locate and measure; deep learning classifies the difficult, natural-looking defects.
Types
Open each type for how it works, where it fits, its limitations and the specifications to check.
How it works. Tools are connected visually; parameters are set with live images. Operators and engineers can adjust recipes without programming.
Best for
Limitations
Key specs to check
How it works. Programming libraries for image acquisition, processing and communication, used inside custom software.
Best for
Limitations
Key specs to check
How it works. Pattern and shape matching, blob analysis, edge and caliper tools, geometric fitting, gauging, color tools, code reading and OCR/OCV.
Best for
Limitations
Key specs to check
How it works. Classification (what is it?), detection (where is it?), segmentation (which pixels?), anomaly detection (trained on good parts only) and deep-learning OCR.
Best for
Limitations
Key specs to check
How it works. Plane fitting, height and volume measurement, 3D matching, profile tools and 3D calibration for robots.
Best for
Limitations
Key specs to check
How it works. Custom operator screens, image archives, statistics and SPC, and communication with PLC, robots and MES or a manufacturing data platform.
Best for
Limitations
Key specs to check
Compare
| Criterion | Rule-based tools | Deep learning |
|---|---|---|
| Best for | Measurement, codes, defined features | Variable, natural defects, complex textures |
| Setup | Configure tools and thresholds | Collect, label and train images |
| Explainability | High — every rule is visible | Lower — validated by test sets |
| Precision of measurement | Sub-pixel, metrological | Not intended for precise measurement |
| Handling variation | Limited to programmed cases | Generalizes from examples |
| Hardware | CPU usually sufficient | GPU often recommended |
Specifications
| Specification | What it means | How to choose |
|---|---|---|
| Camera support | Standard camera interfaces and GenICam compliance. | Freedom to choose cameras and replace them later. |
| Tool coverage | 2D, 3D, codes, OCR, deep learning. | One platform for current and future tasks. |
| Performance | Processing time per tool and parallelism. | Must fit the cycle time with margin. |
| Recipe management | Product-specific settings. | Fast changeovers without engineering support. |
| User management | Roles, passwords, audit trail. | Protects settings; needed in regulated industries. |
| Logging | Images, results, statistics. | Traceability and continuous improvement. |
| Communication | PLC, robot, TCP/IP, file and database export. | Clean handshakes with the line and MES. |
| Maintainability | Documentation, backup, versioning. | Your team can support it after handover. |
Engineering tools
Quick estimates based on standard formulas. Always confirm with a test on real parts.
Answer four questions about your inspection to get a recommended software approach.
Selection guide
| Requirement | Recommended | Why / note |
|---|---|---|
| Precise dimensions | Rule-based measurement tools | Sub-pixel edge and fit tools. |
| Codes and printed text | Code reading and OCR/OCV tools | Deep-learning OCR for difficult fonts. |
| Scratches, stains, cracks on textured surfaces | Deep learning (detection/segmentation) | Collect representative defects. |
| Unknown or rare defects | Anomaly detection | Train mainly on good parts. |
| Presence and position | Pattern matching and blob tools | Fast and simple. |
| Mixed requirements | Hybrid pipeline | Rules locate/measure, deep learning classifies. |
Examples
Illustrative setups showing how the components are combined in practice.
Application example
Application example
Application example
FAQ
It depends on the task and defect variability. Anomaly detection can start mainly with good parts; supervised detection typically needs many labeled examples per defect class. We plan the data collection with you.
Yes, within defined limits. Recipes and user levels let operators switch products while protecting critical parameters.
Yes. Results, measurements and image references can be sent to MES or a manufacturing data platform for traceability and analysis.
New products are added as recipes. For deep learning, models may need additional images and retraining, followed by validation.
Continue exploring
Send us your parts and requirements — our engineers test real samples and recommend a setup.
