The market for AI in image processing will continue to grow. This should not surprise anyone – image processing with its complex image data lends itself particularly well to the use of AI-based algorithms. This is supported by the available and easy-to-use Software Development Kits (SDKs) such as TensorFlow and Caffe. AI models specially trained for use cases that perform tasks such as bulk material monitoring or object position recognition in image processing applications also support this trend. However, classical main processors (CPU, Central Processing Unit) are not very suitable for working with neural networks. Therefore, high-performance graphics processors (GPU, Graphics Processing Unit) or specially optimized, application-specific integrated circuits (ASIC, Application-Specific Integrated Circuit) are often used, which are better suited for the necessary network calculations due to their highly parallel architecture. The AX smart cameras integrate these two aspects directly and are equipped with the market-leading platform NVIDIA Jetson, which in addition to a high-performance GPU also provides ASICs specially designed for AI in the form of DLA cores (Deep Learning Accelerator cores). For example, they achieve up to 300 fps for AI-supported object classification. Among other things, such high image and classification rates help to make the collaboration of fast-operating robots with humans even safer. In addition to AI tasks, smart cameras can also be used for classic image processing. For example, error images can be compressed in JPEG format and sent directly to the cloud to further train and improve the existing neural network. Another advantage: Thanks to a Linux-based approach, the programming language can be freely selected according to the application, image processing libraries or APIs (Application Programming Interfaces) from third-party suppliers can be used flexibly, and custom algorithms can also be used in a protected manner.