Inference on Arm CPUs
With HALCON 19.05, inference for all three deep learning technologies – image classification, object detection, and semantic segmentation – now runs out-of-the-box on Arm® processors. As this removes the need for special components like a powerful GPU or a desktop CPU, HALCON significantly broadens the range of possible deep learning applications.
Enhanced Object Detection
HALCON’s deep-learning-based object detection localizes trained object classes and identifies them with a surrounding rectangle. HALCON 19.05 now also gives users the option to have these rectangles aligned according to the orientation of the object. This results in a more precise detection, as rectangles now match the shape of the object more closely.
Improved Surface-based Matching
Edge-supported surface-based matching is now more robust against noisy point clouds: Users can control the impact of surface and edge information via multiple min-scores. Additionally, in case that no xyz-images are available or in case of noisy point clouds, a new parameter now allows switching off 3D edge alignment entirely. In case of noisy point clouds, this enables users to eliminate the influence of insufficient 3D data on matching results, while keeping the valuable 2D information for surface and 2D edge alignment.
Enhanced Shape-based Matching
With HALCON 19.05, users can now specifically define so-called “clutter” regions when using shape-based matching. These are areas within a search model that should not contain any contours. Adding such clutter information to the search model leads to more robust matching results, for example in the context of repetitive structures.
Speedups
Various operators in HALCON have been sped up. For example, depending on image type and settings, affine_trans_image is now up to 230% faster on AVX2 processors. Furthermore, polar_trans_image_ext can be executed up to 160% faster, depending on the interpolation method.