← All Projects
Accepted · EIFCOM 2026
Edge AI Assistive Tech Embedded Systems

EchoVision

Hybrid NPU-CPU Assistive Navigation

An auditory navigation aid for the visually impaired that runs entirely on a Raspberry Pi 5 with a Hailo-8 NPU — no cloud, no tether. Transformer attention won't compile to INT8, so we split the model at the CNN–attention boundary.

EchoVision: assistive navigation system detecting a nearby person and alerting the user via audio
19.7 FPS
Peak Detection Throughput · Hailo-8 NPU
30×
Latency Reduction vs CPU-Only Baseline
>0.97
Cosine Similarity · NPU–CPU Feature Boundary
The Problem

2.5 FPS is not real time

Assistive navigation has to classify obstacles in milliseconds. On the Raspberry Pi 5 CPU, YOLOv8n manages 2.5 FPS at 404ms.

  • Offloading to an NPU is the obvious fix — but transformer attention is incompatible with INT8 fixed-point arithmetic, and existing toolchains fail to compile it.
  • EchoVision's contribution is a systematic way around that incompatibility without sacrificing segmentation quality.

System Architecture

Two compiled models share one accelerator: detection stays on the NPU, attention falls back to the CPU.

ComputeRaspberry Pi 5 · Cortex-A76
AcceleratorHailo-8 · 26 TOPS · PCIe 3
InputIMX219 camera
OutputProximity-coded audio
EchoVision system architecture: hybrid NPU-CPU pipeline
System architecture — hybrid NPU-CPU partitioning of YOLOv8n and EfficientViT-SAM with the auditory feedback path
  • YOLOv8n runs entirely on the NPU — scanning at 16.5–19.7 FPS, triggering segmentation only on high-confidence detections.
  • EfficientViT-SAM is split — CNN backbone (Stages 0–3) on the NPU; Stage 4 attention, the feature pyramid neck, and the mask decoder on the CPU in FP32.
  • Both models load simultaneously via multi-context allocation.

Key Contributions

Contribution 01

Hybrid CNN-Transformer Partitioning

First systematic method for running transformer vision models on INT8 NPUs, split at the CNN–attention boundary.

Contribution 02

Quantization Failure Taxonomy

Five documented quantization failure modes on the Hailo-8 — a reference for anyone hitting NPU compiler limits.

Contribution 03

Calibration-Based Dequantization

Linear regression from NPU uint8 to CPU float32, holding >0.97 cosine similarity across the boundary.

Hardware Platform

The Hailo-8 mounts over M.2 HAT+; the speaker runs from a custom transistor driver.

Why the speaker needs its own driver circuit

Standard GPIO pins are limited to ~16mA — insufficient for the 8Ω speaker. A 1kΩ base resistor controls the transistor gate, sourcing speaker power directly from the 5V rail, with a flyback diode for inductive spike protection.

EchoVision hardware prototype and BJT driver circuit
Hardware prototype (left) and BJT transistor driver circuit schematic for safe high-current speaker actuation (right)

Results

  • 30× latency reduction — detection moved from 2.5 FPS on the CPU baseline to 19.7 FPS at 13.4ms.
  • Bimodal operation — continuous scanning at 16.5 FPS, dropping to ~1.7 FPS while SAM segmentation runs.
  • 100% proximity-classification accuracy at 2m and beyond.

Publication

Su Ho Lim, Artemis Shaw, Huy Dinh, Nam Bui

VenueEIFCOM 2026 · ACM MobiSys
Where & WhenCambridge, UK · April 2026
AffiliationsCU Denver · CU Boulder