一种基于多信息图像的足部压力检测算法

By using a foot pressure detection algorithm based on multi-information images, and leveraging cameras and multimodal models for non-contact foot pressure prediction, this method solves the problems of high cost and complex operation of traditional equipment, achieving high-precision pixel-level pressure distribution detection, and is applicable to fields such as medical and sports science.

CN117017268BActive Publication Date: 2026-07-17FUDAN UNIVERSITY

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
FUDAN UNIVERSITY
Filing Date
2023-08-14
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Traditional foot pressure detection devices are expensive, complex to operate, and difficult to install. They are also difficult to expand into various scenarios, have poor comfort, and cannot provide pixel-level pressure distribution information.

Method used

A foot pressure detection algorithm based on multi-information images is adopted. Multi-angle RGB images are acquired by a camera, and combined with a segmentation network and a multimodal pre-trained model. Through cross-attention mechanism and feature pyramid decoder, non-contact foot pressure prediction is achieved.

Benefits of technology

It achieves low-cost, easy-to-use, and high-precision foot pressure distribution detection, providing pixel-level pressure information, avoiding the discomfort and interference of traditional devices, and is suitable for medical, sports science, and sports fields.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117017268B_ABST
    Figure CN117017268B_ABST
Patent Text Reader

Abstract

本发明公开了一种基于多信息图像的足部压力检测算法,设计的足部压力预测网络模型采用了一个多输入、多分支的架构,包括三个输入:原始RGB图像、脚的分割图和脚的动作文本提示,输入的图片使用图卷积进行特征提取。模型编码器为ResNet50,解码器为FPN,模型中还引入了CLIP模型用于建立足部动作文本与图像之间的关系,再引入交叉注意力机制用于不同输入信息的交互,得到特征送入解码器;在训练过程中,使用交叉熵损失函数,经过推理,得到图像上每个像素的压力估计,最终得到整个图像的压力预测图,足部压力预测模型可以应用于医疗、运动科学、体育等领域,具有广泛的应用前景。
Need to check novelty before this filing date? Find Prior Art