HALCON Machine Vision Software

• Deep-learning architecture delivers up to 5x faster AI inference while maintaining high detection accuracy. This anchor-free approach improves bounding-box localization and performs reliably for small objects and varying object sizes.

• This feature allows users to provide sample images to automatically identify and remove unstable or misleading contours caused by reflections, shadows, or random textures. Focusing the model on robust contours makes matching faster, more stable, and more accurate.  

• HALCON 26.05 can compensate for geometric distortions when codes are printed on curved, deformed, or non-flat surfaces. This optional rectification step improves decoding reliability for cylindrical components and flexible materials.  

• Configurable operators integrate directly into deep-learning pipelines, replacing previous procedure-based methods. Developers can programmatically define transformations like geometric changes and colour variations to improve model robustness and reduce the need for large training datasets.  

• The latest preview of MVTec’s next-generation integrated development environment (IDE) is included, offering advanced tools and improved workflows for vision developers.  

• Allows models to be updated or refined with only a few images per class.
• Supports adding new classes without retraining the entire model.
• Prevents catastrophic forgetting and runs on edge devices.

• Displays detected and undetected edges in different colours.
• Shows how various model contours contribute to the total score.
• Useful for applications such as analysing overlapping or stacked objects.

• Faster and more resource-efficient text recognition.
• Maintains high accuracy while reducing hardware demands.
• Operates in real time on low-power devices.

• Adds support for MobileNetV4 architectures.
• Enables classification and object detection on limited hardware.
• All models pretrained on industrial data.

• QR code detection more robust on curved or deformed surfaces.
• Bar code reading tolerant to irregular bar widths.
• Supports PQI standards ISO/IEC 15415:2024 and ISO/IEC 29158:2025.

• Provides Software Bills of Materials (SBOMs) in SPDX JSON format.
• Enables transparent tracking of software components.
users
• Simplifies compliance verification and audits.

The new OOD detection feature helps identify unknown or unexpected inputs in deep-learning-based classification systems. When a classifier encounters objects not represented in the training data, HALCON now flags these as “Out of Distribution” and assigns an OOD score to indicate how far the object deviates from the trained classes.

• Detects unseen or abnormal objects during production
• Enables automated interventions (e.g., machine stop or alert)
• Supports dataset improvement by identifying high-value new training images

HALCON 24.11 introduces “Extended Parameter Estimation,” a patent-pending method that improves usability and execution speed for shape-based matching applications. This update makes the tool accessible to users without deep vision expertise while improving performance in production environments.

• More intuitive parameter tuning
• Faster object location
• Suitable for high-speed and precision tasks

The QR code reader in HALCON has been optimized for better performance in difficult conditions:

• Improved detection rate in cluttered or low-contrast environments
• Reduced evaluation time, even when multiple codes are present
• Enhanced robustness against image texture interference

HALCON 24.11 includes a preview of HDevelopEVO, the upcoming replacement for the classic HDevelop IDE. HDevelopEVO is built for modern development workflows, with a focus on productivity and ease of use.

• Modern, streamlined user interface
• Integrated Git version control
• Smart auto-completion and inline documentation
• “Edit & Continue” support for rapid prototyping
• Designed for faster, more efficient development

This release includes an innovative deep-learning-based 3D matching feature for bin-picking and pick-and-place applications. It offers robust object detection and orientation estimation with minimal setup effort.

• Requires no real-world data; training is done with synthetic CAD models
• Supports standard 2D cameras depending on accuracy requirements
• Fast execution with high precision
• Suitable for industrial automation and robotics tasks

With this release, HALCON's GigE Vision interface supports the RoCEv2 network protocol, which enables increased performance in image transmission.

HALCON introduces the first iteration of extended parameter estimation for shape matching. This enhancement improves accuracy and robustness when detecting objects using shape-based matching tools.

The bar code reader is improved to better handle stacked bar codes, increasing reliability in demanding or cluttered environments.

HALCON adds support for importing 3D object models from the STEP format, expanding compatibility with common CAD systems and workflows.

The OpenVINO™ Toolkit AI² plug-in supports the latest Intel hardware and improves inference speed for deep learning applications running on Intel platforms.

HALCON introduces cloud-based licensing, allowing flexible deployment in both public and private cloud environments, no hardware dongle required.

The structured light model is enhanced to deliver faster and more precise 3D reconstructions, particularly for diffuse and non-reflective surfaces. This improves cycle times in production-grade 3D inspection tasks.

HALCON adds deep-learning-based object counting. This new tool reliably counts large numbers of visually similar objects (e.g. bottles or packages) even in overlapping or cluttered scenes.

Deep OCR and 3D Gripping Point Detection now support additional training with user-specific datasets. This enables better adaptation of pre-trained networks to application-specific requirements.

The new Easy Extensions Interface makes it simpler to integrate external code or libraries into HALCON, streamlining the customisation and extension of HALCON's native capabilities.

HALCON`s Deep OCR enables users to efficiently solve text reading applications in a multitude of use cases. In HALCON 22.05 the training functionality has been extended to offer application specific training with the user's own application dataset. This extends the ability of users to solve complex applications like reading text with low contrast (e.g., on tires).

Another advantage of HALCON's Deep OCR is with the ability to train very rarely used special characters or printing styles. Training for Deep OCR significantly improves the performance and usability and makes applications more robust.

HALCON 22.05 allows for detection of logical anomalies in images. This builds on the deep learning anomaly detection feature that HALCON software already provides. Previously, it was possible to detect local, structural anomalies. However, the new “Global Context Anomaly Detection” is a technology that can understand the logical content of the entire image. As with HALCON's existing anomaly detection, the “Global Context Anomaly Detection” tool only requires good images for training, eliminating the need of data labeling. This technology makes it possible to detect entirely new variants of anomalies like missing, deformed, or incorrectly arranged components.

Print Quality Inspection (PQI) refers to the evaluation and grading of certain aspects of printed barcodes and data codes according to international standards. HALCON supports 1D and 2D codes and in HALCON 22.05, the PQI of data codes has been further improved. It is now up to 150% faster. In addition, the module grid determination for print quality inspection of ECC 200 has been improved.

• New operator that performs adaptive histogram equalization to improve contrast locally in an image.
• HDevelop’s Matching Assistant now generates the code based on Generic Shape Matching.
• A number of new tools have been made faster
• Small improvements to HDevelop

A typical issue associated with current AI Accelerator hardware is the proprietary API's that currently have no set standard. To combat this MVTec introduced the AI² Interface. AI² is future-proof and allows the impentation of plug-ins for specific hardware. Currently TensorRT™ and OpenVINO™ tookits are available, but expect more to come.

OpenVINO™ is a toolkit for high-performance deep learning on Intel hardware. The toolkit extends workloads across Intel® hardware (including accelerators) and maximizes performance. OpenVINO™ provides plugins for Intel hardware such as:

• Intel CPUs
• Intel GPUs
• Intel VPUs (e.g., Intel® Movidius™ Neural Compute Stick) Using the OpenVINO™ toolkit plugin can shorten system runtime and lower memory consumption.

HALCON 21.11 allows codes with a larger amount of blur to be read.

HALCON 21.11 delivers a new technology called “instance segmentation”. This combines semantic segmentation and object detection. Instance segmentation allows objects to be assigned to different classes with pixel accuracy. This technology is useful in applications where objects are very close to each other, touch or overlap.

Dictionaries make it easy and convenient to manage complex data in HALCON. For example, different data types such as images, Regions Of Interest and parameter settings can be bundled in a single dictionary. Dictionaries can now be initialized with a single operator call and the syntax for adding and retrieving elements has been simplified. In addition, the auto-completion now also suggests the keys contained in the dictionary, which further speeds up and simplifies working with dictionaries.

HALCON 21.11 provides Generic Shape Matching that offers user-friendly access to MVTec’s industry-proven shape matching technologies. In this version users can now implement their solution much faster as the number of required operators is significantly reduced. Further, existing functionalities are enhanced to increase usability. e.g. the clutter feature has been integrated, handle inspection has been optimised, and additional parameters have been integrated and included in the automatic parameter estimation.

• HDevelop now offers a dark mode.
• Canvas now offers a zooming tool as well as the ability to fit all graphic windows into view.
• The bar code reader now allows users to set a maximum bar code width.

Deep OCR is a holistic deep-learning-based approach for OCR. Compared to existing algorithms, Deep OCR can localize characters much more robustly, regardless of their orientation, font type and polarity. The performance and usability of Deep OCR have been substantially improved in version 21.05. The handling of big images has been upgraded and the subsequent results now contain a list of character candidates with corresponding confidence values. This can be used to improve the recognition results.

HALCON 21.05 introduces Generic Shape Matching which makes MVTec's shape matching technologies more user-friendly and future-proof. In this version users can now implement their solution much faster as the number of required operators is significantly reduced.

HDevelop's new window docking has been improved in HALCON 21.05. Users are now enabled with more options to control the position when floating windows are opened. Previously the top corner of the main screen has been used as the origin. Now it’s also possible to select the upper left corner of the screen where HDevelop is located, or the upper left corner of HDevelop itself. Additionally a new feature called "Auto-hide" has been introduced, this feature allows users to quickly shrink widgets into the side bar when they don't need them and bring them back when necessary.

HALCON 21.05 includes a first version of HALCON Deep Learning framework. This allows experienced users to create their own models within HALCON.

In HALCON 21.05 the subpixel barcode reader has been improved for low-resolved codes. The decoding rate for those can now increase up to 50%.

In this version the 3D point cloud sampling now supports a new mode called "furthest point" which typically results in a more uniform sampling of a 3D object. The 3D point cloud smoothing has been extended by a new mode that uses information from the XYZ-mappings. 3D point cloud smoothing can be used as a preprocessing step to smooth point clouds and remove noise.

HALCON 20.11 offers an edge-supported surface-based 3D-matching which is significantly faster for 3D scenes containing multiple objects and edges. Additionally, the usability has been improved by removing the need to set a viewpoint.

The data code reader in HALCON 20.11 has been extended by the new code type, DotCode. This type of 2D code is based on a matrix of dots. It can be printed very quickly and is especially suitable for high speed manufacturing lines. Furthermore, the ECC 200 code reader now supports the Data Matrix Rectangular Extension (DMRE).

Deep OCR is a holistic deep-learning-based approach for OCR. This new technology brings machine vision one step closer to human reading. Compared to existing algorithms, Deep OCR can localize characters much more robustly, regardless of their orientation, font type and polarity.

In HALCON 20.11, the shape-based matching tool has been improved. This increases usability as well as the matching rate. It also increases robustness in low contrast and high noise situations.

HALCON 20.11 has implemented more viewing options for individual viewing configurations. The changes feature a dark mode with white text on a black background. A new modern window docking concept allows windows to be repositioned. Moreover themes are now improved to improve visual ergonomics and to suit individual preferences.

Deep Learning edge extraction is a new and unique method to robustly extract edges. There are two major use cases for this new tool. The first case is where images have a wide variety of edge types visible. The tool can be trained with only a few images to reliably extract all desired edges. Hence the programming effort to extract specific types of edges is highly reduced. The second major use case, is where edges are low contrast and in high noise situations. The tool is innately able to robustly detect these edges. It makes it possible to extract edges that traditional edge detection filters cannot detect.

HALCON 20.11 introduces a new HALCON/Python interface. This enables developers who work with Python to easily access HALCON's powerful operator set.

Providing increased accuracy and an extended choice of compatible platforms

HALCON 19.11 offers a new anomaly detection system which facilitates the automated surface inspection for, e.g. detection and segmentation of defects.

HALCON's newly implemented Grad CAM supports you in analysing which parts of an image have a strong influence for the inference into a certain class. Furthermore, the Grad-CAM heatmap is very fast compared to the previously offered method.

Within the 19.11 edition CNNs can be exported into the ONNX (Open Neural Network Exchange) format. From 19.11 onwards HALCON is able to read data on ONNX format allowing to use previously created 3rd party networks within HALCON.

HALCON 19.11 is significantly accelerated for multi-core systems.

A new feature in HALCON 19.11 is the generic box finder for pick and place applications. This feature allows users to find boxes of different sizes based on 3D space eliminating the need to train a model for each required box size.

Two editions available- HALCON Steady (as successor of HALCON 13) offered as a perpetual license and HALCON Progress - available as a 12 month subscription with a fast six-month release cycle.

Users are able, seamlessly, to train their own classifier using pretrained CNNs included in HALCON which have been highly optimised for industrial applications.

Improved recognition rate can be increased by 5% (data based on MVTec's internal ECC 200 benchmark consisting of more than 3,700 images from various applications). The ECC 200 reader is now able to read codes with disturbed quiet zone and codes against complex backgrounds can be found and read faster and more robustly.

HALCON features an improved version of the automatic text reader, which now detects and separates touching characters more robustly.

HALCON now features optimized edge detection, which improves the ability to read bar codes reliably with very small line widths as well as those with strongly blurred codes. the quality of the bar codes is also verified in accordance with the most recent version of the ISO/IEC 15416 standard.

HALCON 18.11 introduces a new data structure "dictionary", an associative array that opens up new ways of working with complex data. For example, this allows bundling various complex data types (eg., an image, corresponding ROIs and parameters) into a single dictionary, making it easier to structure programs

Double-clicking a handle variable now returns all parameters associated with the handle and their current settings.

HDevelop provides a new library export that makes the use of HALCON procedures from C++ and .NET as easy and intuitive as calling any other C++/.NET function. This new library export also generates CMake projects which can easily be configured to output project files for many popular IDEs

HALCON now makes it much more comfortable to work with handles by clearing these automatically once they are no longer required. This significantly reduces the risk of creating memory leaks and makes writing "safe code" much simpler.

HALCON now offers a method that fuses multiple 3D point clouds into one watertight surface. This new method combines data from various 3D sensors, even from different types like a stereo camera, a time of flight camera, and fringe projection. This technology is especially useful for reverse engineering.

HALCON now offers optimized functions for surface-based 3D matching. These can be used to determine the position of objects in 3D space more reliably, making development of 3D applications easier.

A new camera model within HALCON now allows the corrections of distortions in images that were recorded with hypercentric (also known as pericentric) camera lenses.

HALCON 18.11 includes new operators, which enable the user to inspect specular and partially specular surfaces to detect defects by applying the principle of deflectometry.

This powerful machine vision software suits all types of applications. HALCON 13 not only masters complex 3D machine vision tasks but also offers supplier's experience, a comprehensive set of technologies and the flexibility of your choice of hardware.

HALCON views information simultaneously from multiple cameras leading to a more robust reconstruction than was possible using previous stereo methods.

HALCON's HDevEngine applications can now be de-bugged directly, making error tracking, and even remote error-tracking, much easier.

New deep-learning technology achieves higher reading rates than with previous classification methods. HALCON 13's automatic text reader is faster and supports reading of dot print characters and barcodes even if large parts of the code are defective or not visible. Dealing with blur and distortion is easier.

Traditionally one the the more challenging machine vision tasks, dealing with an objects different characteristics has proved problematic. HALCON 13 offers an easy-to-use texture inspection, that automatically identifies defects by comparison with an image of flawless materials

HALCON 13 also offers significant speedups for all related technologies, i.e., shape-based 3D matching, local and perspective deformable matching, and component-based matching.