基于改进YOLOv8的重参数化人体姿态估计算法OA

In order to cope with the common challenges in human pose estimation such as occlusion,complex background and body type diversity,the YOLOv8-DRS model is proposed based on the YOLOv8 model.Firstly,DBB reparameterization module is introduced into the backbone network to improve C2f,optimize parameter settings,and effectively enhance feature extraction capa-bility without increasing the number of parameters.Secondly,RFAPose is adopted in the detection head to emphasize the impor-tance of different features,so as to improve the accuracy of key point detection under occlusive and complex background conditions.Finally,SIoU loss function is introduced to improve the model convergence ability and maintain high accuracy when dealing with di-verse body shapes and poses.The experimental results on the COCO human pose estimation dataset show that the accuracy P is in-creased by 0.2%,the recall rate R is increased by 1.8%,and the AP50 is increased by 1.5%,which effectively improves the accura-cy of the model.