Drishti – Vision for the Blind: An AI-Driven Assistive Navigation System

Drishti – Vision for the Blind presents an AI-driven assistive navigation system designed to improve mobility and independence for visually impaired individuals. By combining real-time object detection (YOLOv8), voice command recognition (Whisper), scene understanding, text reading (OCR), and contextual reasoning via a Hybrid Brain powered by Gemini 2.5 Flash-Lite, Moondream2, and Florence-2, the system delivers rich spoken environmental awareness through a TTS Output pipeline. The Vision Consumer backend layer ensures smooth coordination between video frame ingestion, audio command processing, and the AI Engine, making the overall system responsive and scalable. The results demonstrate that the proposed system can effectively detect obstacles and provide timely audio guidance. With further improvements, Drishti has the potential to become a widely adopted assistive technology for visually impaired users.

[1] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016, pp. 779–788. [2] G. Jocher et al., “Ultralytics YOLOv8,” GitHub Repository, 2023. [Online]. Available: https://github.com/ultralytics/ultralytics [3] A. Radford et al., “Robust Speech Recognition via Large-Scale Weak Supervision,” in Proceedings of the 40th International Conference on Machine Learning (ICML), Honolulu, HI, USA, 2023, pp. 28492–28518. [4] Google DeepMind, “Gemini 2.5 Flash-Lite: Efficient Multimodal Reasoning,” Google AI, 2025. [Online]. Available: https://deepmind.google [5] X. Xiao et al., “Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks,” Microsoft Research, 2024. [Online]. Available: https://www.microsoft.com/en-us/research/ [6] V. Kumar, “Moondream2: Lightweight Visual Question Answering Model,” 2024. [Online]. Available: https://github.com/vikhyat/moondream2