NS
Nishant OS
Initializing Nishant AI
LOADING0%
All work
ResearchAssistive AI · Edge · IoT · Mar 2024 – Sep 2024

DRISHTI

A Raspberry Pi wearable that guides the visually impaired with real-time multimodal sensing.

98.42%
Detection precision
0ms
Latency
real-time
0%
Offline
no cloud
Multimodal
Sensing
vision · ultrasonic · audio

DRISHTI is a Raspberry Pi 5-based wearable that fuses YOLOv8-nano object detection, ultrasonic ranging, on-device face recognition, and image captioning into real-time navigation for visually-impaired users — reaching 98.42% detection precision at 25 ms latency, fully offline. Presented at the ICTIS 2026 Conference, Bangkok, Thailand.

The problem

Assistive navigation has to be real-time, wearable, and reliable exactly when connectivity drops — yet most solutions are bulky, cloud-dependent, or too slow to trust on the move.

The solution

A Raspberry Pi 5 platform runs YOLOv8-nano detection, ultrasonic distance sensing, KNN-based on-device face recognition with dynamic enrollment, and image captioning — fused into multimodal voice and haptic feedback for situational awareness, with zero cloud dependence.

System Architecture
Camera + UltrasonicYOLOv8-nanodetectionFace RecognitionKNN on-deviceImage CaptioningMultimodal FusionpriorityVoice + Haptic
Challenges solved
  • 1Real-time inference within a Raspberry Pi's power and compute budget.
  • 2Fusing vision, ranging, and audio into calm, non-overwhelming guidance.
  • 3On-device face recognition with dynamic enrollment and no cloud.
Highlights
  • 98.42% detection precision at 25 ms, fully offline on a Pi 5.
  • Vision + ultrasonic + face recognition + captioning, fused.
  • Published at ICTIS 2026, Bangkok, Thailand.
Technology
PythonYOLOv8-nanoOpenCVRaspberry Pi 5PyTorchKNNUltrasonic Sensors
Next project
Domain Adaptation
Transfer Learning · Computer Vision