Blind person glasses identification system based on end-to-end visual inspection

CN122018677APending Publication Date: 2026-05-12LINKER
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
LINKER
Filing Date
2025-12-29
Publication Date
2026-05-12

AI Technical Summary

Technical Problem

Existing visual inspection algorithms are computationally intensive and have high latency on mobile devices, and rely on NMS steps, making it difficult to meet real-time response requirements and limiting their effective application in mobile scenarios.

Method used

An end-to-end detection method without NMS is adopted, combined with small target perception label allocation and lightweight distillation training strategy, to design a visual glasses recognition system for the blind. The system includes visual perception, feature fusion, detection prediction and lightweight distillation modules. The model is optimized by model distillation, quantization and pruning to adapt to resource-constrained hardware, and is linked with a voice interaction module to achieve real-time feedback.

Benefits of technology

It significantly improves inference speed and response efficiency, reduces detection latency by about 40%, reduces computational load and power consumption, enhances small target detection capabilities, realizes real-time scene recognition and voice-assisted feedback, and improves environmental perception and travel safety for blind users.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SMS_1
    Figure SMS_1
  • Figure SMS_7
    Figure SMS_7
  • Figure SMS_11
    Figure SMS_11
Patent Text Reader

Abstract

The invention discloses a blind person glasses recognition system based on end-to-end visual detection, and the system comprises a visual perception module which is connected with a blind person glasses high-definition camera, receives an original image, and extracts multi-scale features; the feature fusion module integrates shallow texture and deep semantic features and supports multi-scale feature parallel processing; the detection prediction module directly predicts the category, the position and the confidence coefficient of a target through an NMS-free end-to-end detection head; and the lightweight distillation deployment module is linked with the voice module for real-time feedback through distillation, quantification and pruning lightweight adaptive hardware. According to the method, the feature extraction efficiency is improved through multi-module cooperation, NMS direct prediction is not needed, hardware is adapted in a light-weight mode, real-time voice feedback is achieved, and convenience and real-time performance of blind person environment recognition are enhanced.
Need to check novelty before this filing date? Find Prior Art