ROpenPose: A Rapider OpenPose Model for Astronaut Operation Attitude Detection

This work proposes a rapider OpenPose model (ROpenPose) to solve the posture detection problem of astronauts in a space capsule in a weightless environment. The ROpenPose model has three innovations as follows: 1) It uses MobileNets instead of VGG-19 to achieve lighter calculations while ensuring the accuracy of model recognition. 2) Three small convolution kernels replace the large convolution kernel of the original OpenPose, which significantly reduces the computational complexity of the model. 3) Through the parameter sharing of a convolution process, the original two-branch structure is changed to a single-branch structure, which obviously improves the calculation speed of the model. A residual network is proposed to suppress the hidden danger of gradient disappearance. The deployment of ROpenPose greatly improves astronauts' detection efficiency while ensuring their high detection performance, and thereby realizing the real-time monitoring of their operation attitude. Experimental results show that ROpenPose runs at speed higher than and detection performance comparable to a number of the existing state-of-the-art models.