Everyone is looking for ways to reduce the time required for designing complete machine vision and deep learning applications while maintaining the highest possible performance. One of the possible solutions is choosing an environment, which does not require sophisticated programming skills but allows to quickly prototype & create machine vision applications instead.
At Adaptive Vision we solve these problems since 2007, when we started the development of our graphical programming environment – Adaptive Vision Studio. From the very beginning, our goal was to provide machine vision engineers with an environment that is intuitive, powerful and adaptable at the same time. It is reflected in its capabilities – it allows for programming image processing algorithms, communication with external devices, and designing Human Machine Interfaces without any line of C++/C#/Python code. Naturally, a user-friendly interface, HMI designer and the possibility of acquiring images from most cameras available on the market would be worthless without a reliable and comprehensive imaging library containing over 2000 functions for traditional machine vision techniques. In 2016 we decided to enrich our product with tools based on deep learning technology, which are highly optimized for industrial purposes and do not require expertise in machine learning.
New Deep Learning Tools
Recently released version 5.1 perfectly shows that the company is not slowing down. It features deep learning tools for OCR and object detection, exciting improvements in the edge-based template matching, improved 3D data previews, new HMI controls, support for Ethernet/IP, extended debugging capabilities, and many new features in the IDE. The DL_ReadCharacters tool is for highly demanding OCR applications. It achieves ~98% accuracy out of the box, without any prior training and can deal with problems like complex background, non-uniform lighting, blurred or damaged characters, also on reflective surfaces.
There is also a new tool, DL_LocateObjects, for object detection based on oriented rectangles, a great counterpart of traditional template matching technique. To further improve the Deep Learning workflow, the new Automatic Training feature allows training multiple models and perform detailed comparisons. The 5.1 Deep Learning editor also comes with the new Model History functionality, which enables fast switching between all previously trained models.
