改进的双流多模态信息融合坐姿识别方法

不正确的坐姿通常会导致青少年近视、脊柱侧弯和退行性疾病。研究能够快速、准确识别不规律坐姿的智能监测技术,有助于保持正确的姿势并预防健康问题。为了解决RGB图像易受光照强度以及遮挡因素的干扰并造成的识别率不高等问题,通过采用双流RGB-D图像作为双输入,利用ResNet网络中的残差结构改进EfficientNet基线网络结构,提出了一种基于改进R-EfficientNet的双流RGB-D多模态信息融合的坐姿识别方法。试验结果表明,提出的R-EfficientNet融合方法模型对8种坐姿的识别均值平均精度(mean average precision, mAP)达到了98.5%。与CNN、Vgg16、ResNet18、EfficientNet、RGB-D不同的输入方法相比,所提方法获得了最高的识别率。该方法不仅可以用于坐姿客观监测,具有医学和社会效益,此外还为人体工学研究者们提供改进办公家具的方案。

An improved dual-stream multi-modal information fusion method for sitting posture recognition

Improper sitting posture often leads to myopia, scoliosis and degeneration in adolescents. Research on intelligent monitoring technology that can quickly and accurately identify irregular sitting postures can help maintain correct posture and prevent health problems. In order to solve the problem that the RGB image is easily disturbed by the hint strength and occlusion factors and the recognition rate is not high, by using the dual-stream RGB-D image as the double input, the residual structure in the Resnet network is used to improve the Efficientnet network structure, and a dual-stream RGB-D multi-modal information fusion test based on the improved R-Efficientnet is proposed. mAP) reached 98.5%. Compared with different input methods such as CNN, Vgg16, Resnet18, Efficientnet, and RGB-D, the method in this paper has achieved a higher recognition rate. The method not only can be used for monitoring the sitting posture, but also has a medical and sociological sitting posture recognition method. In addition, it also provides better automated work solutions for ergonomic people.