Research on fast network of human pose estimation in underground coal mine based on HigherHRNet

Fast estimation of human pose in underground coal mine is an important prerequisite for intelligent safety detection of underground operation. Aiming at the problems of dusty and foggy, insufficient illumination and color blending in underground coal mine, this study conducts an in-depth study on the lightweight design and key point assignment of the HigherHRNet model and proposes a new Optimising HigherHRNet(OH-HRNet) fast network model in order to improve the accuracy of the key point assignment for human pose estimation as well as the network operation speed.The OH-HRNet model proposes a memory convolution module based on attention mechanism and a key point assignment algorithm with reinforced skeletal constraints, and the loss function of the algorithm is improved. Experiments on the coal mine underground scenario dataset and COCO public dataset show that, OH-HRNet is 1.06 times faster than LitePose in terms of GPU speed, with a 7.4% increase in mAP and a 14.0% increase in mAR, which can achieve more effective intelligent safety detection.