Lightweight Multiplayer Pose Detection Model with Improved YOLOpose
2D human posture estimation is of great significance for safety production,intelligent interaction and other research.Aiming at the current human posture estimation model with large computation and slow detection speed,this paper proposes a lightweight im-provement algorithm based on the YOLOpose model.Firstly,the GSConv convolution module,which is more delicate in operation,is introduced to replace the ordinary Conv convolution,which greatly reduces the model computation and complexity;then the UPSample module is replaced by the CARAFE module to complete the up-sampling work,and at the same time,the CBAM attention mechanism module is introduced to avoid the problem of reduced accuracy brought by the model lightweighting.The experimental results show that after the YOLOpose model is improved by the above lightweighting,the model volume is reduced to 135.6MB,which is reduced by about 15.8％,and the GFLOPS is reduced to 86.9,which is reduced by about 15.0％,with a significant reduction in the model computation volume,and then the addition of the CBAM attention mechanism has a small effect on the model accuracy,and the im-proved model can ensure the accuracy of recognition and also realize the lightweight of the detection algorithm.
NETL
NSTL
万方数据
