AI Smart Cane
A low-cost offline AI-powered mobility aid for visually impaired users, built on a Raspberry Pi Zero 2W. It integrates RGB visual perception and ToF distance estimation, employs an INT8-quantized SSD MobileNet V1 model, and provides dual-modality feedback through vibrotactile and audio channels.
Timeline (last 90 days)
An arXiv paper proposes an AI smart cane that costs only $88 and runs entirely offline.
The core hardware of the system is the Raspberry Pi Zero 2W microcomputer board.
The system integrates two types of sensing: RGB visual perception and Time-of-Flight (ToF) distance estimation.
The system uses an INT8-quantized SSD MobileNet V1 object detection model.
The system adopts a multi-process architecture that isolates sensor acquisition, neural network inference, and haptic feedback into separate processes, with a fail-safe mechanism incorporated.
The smart cane converts distance information measured by ToF into distance-aware vibrotactile feedback, where the closer the obstacle, the stronger the vibration.
The system's end-to-end average latency is 330 milliseconds, and peak power consumption is only 2.8 watts.
In a preliminary usability study with 12 participants, the system received a SUS score of 78.5 and was evaluated through NASA-TLX workload assessment.
The experiments in this study focused mainly on indoor scenarios with a sample size of only 12 people, and performance in real outdoor complex environments has not yet been validated.
Fully offline operation means data never leaves the device, avoiding the privacy risks associated with cloud-based solutions.
All Facts (10)
Fully offline operation means data never leaves the device, avoiding the privacy risks associated with cloud-based solutions.
50%UnverifiedThe core hardware of the system is the Raspberry Pi Zero 2W microcomputer board.
50%UnverifiedThe system integrates two types of sensing: RGB visual perception and Time-of-Flight (ToF) distance estimation.
50%UnverifiedThe system uses an INT8-quantized SSD MobileNet V1 object detection model.
50%UnverifiedThe system adopts a multi-process architecture that isolates sensor acquisition, neural network inference, and haptic feedback into separate processes, with a fail-safe mechanism incorporated.
50%UnverifiedThe smart cane converts distance information measured by ToF into distance-aware vibrotactile feedback, where the closer the obstacle, the stronger the vibration.
50%UnverifiedThe system's end-to-end average latency is 330 milliseconds, and peak power consumption is only 2.8 watts.
50%UnverifiedIn a preliminary usability study with 12 participants, the system received a SUS score of 78.5 and was evaluated through NASA-TLX workload assessment.
50%UnverifiedThe experiments in this study focused mainly on indoor scenarios with a sample size of only 12 people, and performance in real outdoor complex environments has not yet been validated.
50%UnverifiedAn arXiv paper proposes an AI smart cane that costs only $88 and runs entirely offline.
50%