MVTec Software GmbH, a leading international provider of machine vision software, will release version 23.05 of its standard machine vision software HALCON on 23 May 2023. The new version focuses on deep learning methods. The main function is Deep Counting, which is a deep learn-based method that can count a large number of objects stably and reliably. In addition, the new HALCON version incorporates improvements to Deep learning techniques for 3D grab point detection and Deep OCR training. HALCON 23.05 is now available to further optimize the underlying deep learning networks that have been pre-trained for user-owned applications using industry-relevant imagery. This allows for a more stable recognition rate for Deep OCR applications and helps applications using 3D grab point detection to more reliably detect suitable grab surfaces. In addition, there are many other beneficial improvements, such as the fact that it is now easier to integrate external code into HALCON.

“We are seeing a significant increase in customer interest in integrating deep learning methods into our own solutions. We developed the new HALCON version with this in mind, and the result is new deep learning techniques and further development to help customers get more accurate results, “explains Jan Grtner, Product Manager at MVTec HALCON.

Deep Counting

Beginning with HALCON 23.05, customers can use the “Deep Counting” feature, which allows for fast and reliable counting and location detection of large numbers of objects. This deep learning-based technique has clear advantages over existing machine vision approaches: Feature deployment is very rapid because there are few objects to tag and train, and both steps can be done easily in HALCON. The technique provides reliable results even for objects made of highly reflective amorphous materials. You can use Deep Counting to count a large number of objects, such as glass bottles, tree trunks, and food.

Deep OCR training

Deep OCR can read text very stably, even without being affected by direction and font. The technology first detects relevant text in the image and then reads it. With HALCON 23.05, text detection can now also be fine-tuned by retraining pre-trained networks using application-specific images. The result is more stable and opens up new possibilities. For example: detecting text of any print type or character type not seen before, and improving reading ability in low contrast noisy environments.

Training of 3D grab point detection

3D grab point detection can reliably detect any surface on any object suitable for a suction grab. In HALCON 23.05, it is now possible to retrain pre-trained models with application-specific proprietary image data. This allows for more stable identification of grabable surfaces. Using MVTec deep learning tool, necessary annotation can be completed easily and efficiently.

Simple extension interface

With the HALCON extension pack, external programming languages can be integrated. The integration of external code in HALCON 23.05 is more convenient. With the new Easy Extensions Interface, users can now use functions they wrote in.NET code in HDevelop and HDevEngine in just a few steps. You can even use known data types and HALCON operators from the Halcon /.net language interface. Customer benefits: HALCON can now do more than just image processing. This increases the flexibility and application possibilities of HALCON.