OpsisVision: A Multimodal AI Assistance System for the Blind and Visually Impaired with Hybrid Edge-Cloud Architecture

Assistive technologies for the blind and visually impaired have reached enormous potential through advances in artificial intelligence (AI) and wearable computing systems. However, many existing solutions remain limited by high costs, proprietary hardware, or dependence on pure cloud services. This paper presents OpsisVision, a multimodal AI assistance system based on a low-cost, 3D-printed glasses frame with standard hardware. The system combines local real-time obstacle detection using YOLOv11 with precise depth measurement through stereoscopic vision (Stereo Vision). For deeper semantic understanding of the environment, a hybrid architecture is used that combines local object detection with visual analysis by a Large Language Model (GPT-4o) in the cloud. Interaction is entirely via voice commands in Greek, enabled by local wake-word detection (Vosk) and cloud-based speech recognition (Whisper). We describe the system's architecture, its implementation on an NVIDIA Jetson Orin, and the results of a successful test run. A detailed cost analysis shows that OpsisVision, with estimated hardware costs of approximately 470 euros (8GB cloud version) to 2,200 euros (64GB offline version), is significantly cheaper than commercial alternatives such as Envision Glasses ($3,500) or OrCam MyEye Pro ($6,000). Furthermore, we discuss the ethical implications, particularly with regard to the EU General Data Protection Regulation (GDPR) and the European Accessibility Act (EAA) 2025. OpsisVision demonstrates a scalable and cost-effective approach that can be evolved from a pure cloud prototype to a fully offline-capable system, thereby significantly increasing the accessibility and independence of the blind and visually impaired.