AI Inferencing and Multi-Camera Vision with Qualcomm SOMs
Interested in measuring depth without stereo hardware? Or real-time object detection on mobile and edge devices? Critical Link has developed a series of demos running open source AI models on the MitySOM-QC6490. The SOM is built around the Qualcomm QCS6490, targeting the 6th Gen Qualcomm AI Engine and Hexagon DSP for acceleration. Pipeline includes camera capture, model inference, and overlay rendering at the edge, giving you a concrete baseline for throughput, latency, and resource utilization on this class of SoC. From an engineering perspective, it’s a working reference for image capture, pre-processing, model execution, and secure storage in a single embedded platform. It’s a practical starting point if you want to swap in your own models, adjust input resolution, or integrate application logic for use cases like quality inspection, security/access control, smart kiosks, retail analytics & personalization, smart agriculture, or on-device analytics in medical and smart city devices.
Our AI-driven single-camera depth perception demo is shown below. This demo uses Qualcomm’s MiDaS V2 implementation on the QCS6490, generating real-time relative depth maps and visualizing distance as a heat map. It’s designed for robotics, surveillance, and safety/security use cases where size, cost, or mechanical constraints limit sensor options.
See other demos and explore integration options: www.criticallink.com/demos
Check out the MitySOM-QC6490 System on Module specifications or browse available Development Kits
See more MitySOM Demos: www.criticallink.com/demos
Use our SOM Selector