Design of pull-up test platform based on semantic segmentation and human body posture estimation
In order to promote the intelligent development of body measurement,a pull-up test system combining semantic segmentation and human body pose estimation is designed. On the basis of fully considering the pull-up counting rule and hardware deployment,the feature extraction part Xception of the semantic segmentation model DeepLabV3+is lightweight improved by ShuffleNetV2,and the efficient channel attention(ECA)mechanism is introduced into the decoding module,which is applied to the segmentation task of the pull-up horizontal bar as a whole. The pose estimation model BlazePose is used to detect the key point information of the human body,and the discriminant algorithm is written through the transformation characteristics between its position coordinates to complete the pull-up test function. Finally,the horizontal bar target segmentation task and the human body key point detection task are successfully deployed on the edge computing platform Jetson nano,and TensorRT is used to accelerate the reasoning and achieve smooth operation of 18 frames.
万方数据
维普
