Zebra Aurora Vision™ 5.3 現已上市

Release Notes: Zebra Aurora Vision Studio™ 5.3

2023.03.22

Deep Learning

New: A completely new tool is available, DL_LocateText. It does not require any training and automatically detects text (words) in images. Use it to find the ROI (region of interest) for DL_ReadCharacters whenever text location cannot be determined with simpler means. DL_LocateText detects text at any orientation in the range of ±90 degrees, and handles font sizes between 20 and 50 pixels.
Improved: The DL_ReadCharacters tool has been re-trained on an even larger database of images, now including significant amount of currency symbols ($, €, £, ¥, ¢), special characters (+, =, % etc.) and new industrial fonts (most notably the OCR A).
Improved: The DL_ReadCharacters tool now comes with re-designed character scoring function, yielding significantly less False Negatives for confusing characters (like 0 vs O or I vs l).
Improved: The DL_ReadCharacters (Fast) tool has been further optimized, giving ~10% lower inference time.
Improved: The DL_ReadCharacters tool now comes with a new model, "Scalable", that is based on "Balanced", but supports a wide range of character height (20-100 px).
Improved: The architecture of the neural network behind the DL_DetectFeatures tool has been re-optimized. We used a set of 15 industrial and non-industrial use cases as a benchmark, and performed both manual and automatic (ProxylessNAS) architecture optimization. As a result, the tools is much faster (~20% faster on GPU, twice as fast on CPU) while accuracy is preserved or better (depending on the use case).
Improved: The neural network behind the DL_DetectFeatures tool is now pre-trained with a large database of images. This makes training much faster - it is now enough to train for 10-15 iterations to get a good model. Another benefit is that training is more stable for small datasets.

New and improved filters

New: Data Matrix code reader received a new detection algorithm that may handle codes with mixed polarity modules (some modules are darker, some are brighter than the background). It works especially well with DPM with direct illumination (direct print marking).
Improved: Our 1D code reading tool has been thoroughly revised, significantly improving both speed and accuracy. The filter's interface has been simplified.
New: There is a new code reading tool, based on a very different algorithm, that can read many types of codes at one time.
Improved: We have improved speed and accuracy of the GoldenTemplate3D tool.
Fixed: PolygonIntersectionArea was returning 0 when the same non-empty polygon was put to both inputs.
Fixed: ImageLocalMaxima.inMinValue was off 3 times for RGB images.
Improved: ResizeImage_Area has been better optimized for 1-channel images, including for NEON instruction set, and for 3-channels images, SSE/AVX only.
New: A few new auxiliary filters have been added: ReorientRectangle, AlignShapeRegion, ReadEnvironmentVariable_OrNil, AppendImageChannel, EllipseArea, RegionBoundingEllipse(_OrNil), PointsBoundingEllipse.
Improved: Gray-based Template Matching has been significantly improved (as Edge-base had been in 5.2), including:

highly reduced model size
much faster model creation
slightly faster search
more consistent object scores

Improved: "Execute" filter (for running system scripts) was reimplemented; many issues were resolved.

Additionally, when executing it, Windows' console window is no longer shown.
The Execute_WaitForExit filter has a new inEncoding input, to specify what encoding to use when reading program output. The user can also specify "RAW" encoding and use bytes from new outRawResult output directly.
The Execute_WaitForExit filter now also reads standard error stream on Windows.

IDE

New: We changed the way filters without set inputs work - instead of interrupting the program execution, we mark them as "Not Ready". If the user decides to execute a program containing "Not Ready" filters, these filters work as if they were disabled.
Improved: The auto-connect feature has been optimized, yielding much less false positive connections now.
Improved: The Filmstrip control now watches file changes and can update its content automatically.
Fixed: Status bar layout has been fixed - its space is better used, call stacks can be fully displayed.

Linux & Android

New: Deep Learning Add-on (both training and inference) is now available for Linux.
New: Aurora Vision Library and Deep Learning Add-on (inference only) can now execute on the Android operating system. Contact support for details as it is not included in the publicly available distribution package.

Hardware Support

New: Added support for Zebra Scanner SDK, making it possible to have Aurora Vision Studio receiving images from Zebra's scan engines and hand-held devices.
New: Added support for Matrox® 4Sight EV6 and EV7 industrial computers. It is now possible to control their digital I/Os, timer, generate software trigger, get values from the encoder counter and configure the timer. Important note: in the next step, support for industrial protocols (Profinet, EtherNet/IP and Modbus RTU) will be added.
New: Added support for HIKROBOT industrial computers. It is now possible to control their digital I/Os, light ports (brightness, state after triggering, trigger edge, duration time) and reboot a device.
New: Added support for Nodka industrial computers (NKDIO SDK). It is now possible to control their digital I/Os, built-in light controller, and set general parameters of the device. The implementation of all SDK functionalities is done in the Nodka_Server.exe that can be found in the Aurora Vision installation path.
Update: Updated support for Vimba SDK to version 6.0.
Update: Updated support for SmartRay SDK to version 5.5.1.41.
Update: Updated support for FlyCapture SDK to version 2.13.3.61.
Update: Updated support for Spinnaker SDK to version 2.7.0.128.
Update: Updated support for SynView SDK to version 1.03.014.

Other

New: Our products are now build with Visual Studio 2022, gaining a few percent performance gain for some tools.