DeepView Vision Packs provide the vision pipeline solutions for your edge computing and embedded machine learning applications. Whether targeting NXP i.MX Application Processors and i.MX RT Crossover MCUs, Au-Zone tunes these pipelines for maximum efficiency. Vision Packs provide end-to-end optimizations, including sensor support and pre-processing and full integration with Au-Zone’s DeepViewRT runtime engine. For prototyping or commercial deployment, Vision Packs run out-of-the-box with DeepView Vision Starter Kits.
MPU Vision Pack
The VisionPack for MPU allows you to directly deploy ML workloads to any desired compute core in NXP's i.MX family of Applications Processors and leverage hardware acceleration blocks to quickly implement a fully optimized vision pipeline in you project.
Based on application requirements, features can include pre-processing functions such as color space conversion, image resize, DMA-based zero copy buffers for routing image data to different processing engines, and GStreamer integration to expedite application development.
The VisionPack’s image pipeline processing is tightly coupled with Au-Zone’s DeepViewRT inference engine to minimize the latency and overhead for deep learning applications. The pipeline maintains data synchronization across frames for each processing block to ensure metadata is coherent with raw image data.
MCU Vision Pack
The VisionPack for MCU includes integrated sensor drivers and image pipeline processing optimized for NXP’s i.MX RT crossover MCU product family. The runtime software includes HDR sensor support for challenging lighting situations faced by surveillance and monitoring systems.
Default sensors include OmniVision OV10626 industrial sensor with very good low light sensitivity and 120dB HDR modes. The image pipeline supports both raw and compressed data formats for different use cases and applications. The On Semi AR0144 sensor is available for applications that require a global shutter sensor, enabling better capture of fast moving or fast changing events. Additional sensors and drivers can be requested with additional engineering cost.
The VisionPack uses hardware acceleration blocks such as the i.MX RT’s dedicated Pixel Processing Pipeline (PXP) to implement color space conversion, image resizing, and DMA-based memory copies that is critical to reduce the CPU processing load for efficient image pre-processing.