Check out more Sensors Converge coverage. This video is also part of the TechXchange: AI on the Edge.
Synaptics' Katana-based edge artificial-intelligence (AI), low-power SoC is powerful enough to analyze a video stream with machine-learning (ML) models that can visually detect actions like falling. Ananda Roy highlights this feature in the video.
The demonstration was done using Synaptics evaluation kit (Fig. 1). The platform, based on the Katana SoC, adds audio input via two microphones and a Himax QVGA video module along with an ambient-light sensor and accelerometer. The Wi-Fi/Bluetooth module provides wireless connectivity. There's even an OLED display for immediate user feedback.
Katana is a multicore SoC with a dual DSP to handle ML models in real-time (Fig. 2). It also manages audio and video data. A Cortex-M3 core provides system-management support.
The SyNAP toolkit is a full-stack, open-source platform that supports ONNX and AI frameworks like TensorFlow and TensorFlow Lite.