Scholarlink

<p>The limited computational capacity of exoskeleton controllers and the high demands of visual detection models create challenges in real-time applications, particularly in stair ascent and descent where stringent real-time constraints are imposed. To address this issue, we first propose the Dynamic Sparse Channel Gating (DSCG) method, which realizes adaptive channel pruning according to image content. Subsequently, leveraging the YOLOv11 framework, we design the DSCG-YOLO, a lightweight visual detection model that integrates the proposed DSCG method. To support real-time detection in stair-walking scenarios, we construct the Exoskeleton Stair Walking Dataset (ESWD), with annotated individual stair steps and depth map filtering to accurately measure the distance between the exoskeleton and the steps. Experimental results show that the proposed model achieves a mAP@0.5 of 96.4% and an inference speed of 68 frames per second (FPS) on an NVIDIA GTX 1650 GPU, a 51% improvement over the baseline. As a lightweight detection model with strong real-time detection capabilities, DSCG-YOLO can be effectively applied to real-time scenarios, to enhance the walking safety of exoskeleton users.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DSCG-YOLO: a lightweight visual inspection model for exoskeleton in stair walking scenarios

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Tianhong Luo,
  • Wei Yao,
  • Shuang Liang,
  • Xiangyu Ma,
  • Yang Li,
  • Qi Tuo

摘要

The limited computational capacity of exoskeleton controllers and the high demands of visual detection models create challenges in real-time applications, particularly in stair ascent and descent where stringent real-time constraints are imposed. To address this issue, we first propose the Dynamic Sparse Channel Gating (DSCG) method, which realizes adaptive channel pruning according to image content. Subsequently, leveraging the YOLOv11 framework, we design the DSCG-YOLO, a lightweight visual detection model that integrates the proposed DSCG method. To support real-time detection in stair-walking scenarios, we construct the Exoskeleton Stair Walking Dataset (ESWD), with annotated individual stair steps and depth map filtering to accurately measure the distance between the exoskeleton and the steps. Experimental results show that the proposed model achieves a mAP@0.5 of 96.4% and an inference speed of 68 frames per second (FPS) on an NVIDIA GTX 1650 GPU, a 51% improvement over the baseline. As a lightweight detection model with strong real-time detection capabilities, DSCG-YOLO can be effectively applied to real-time scenarios, to enhance the walking safety of exoskeleton users.