Use Case "Edge AI"
Led by Fraunhofer ITWM, the Edge AI use case focuses on running a deep-learning video analytics pipeline directly on CAPE edge servers. By processing data close to the cameras, the system eliminates cloud round-trip latency for real-time anomaly detection. It also preserves privacy by transmitting only compact, privacy-preserving information rather than raw video.
The pipeline addresses smart-city scenarios, detecting anomalies such as unusual human behavior, safety incidents, and irregular traffic, using advanced vision–language models (e.g. CLIP).
A key technical focus is integrating RISC-V and CXL technology into the CAPE project platform. Fraunhofer drives the design and integration of an FPGA-based RISC-V deep-learning accelerator into the Edge High Performance Server (EHPS), utilizing CXL for efficient host–accelerator synchronization.
The use case is evaluated by benchmarking representative kernels on CAPE FPGAs against commercial off-the-shelf solutions. This assessment focuses on offload overheads, performance, and energy efficiency to demonstrate how open RISC-V acceleration combined with CXL offers a practical path for low-latency edge AI.