Research On Human Posture Information Extraction Algorithms For Semantic Communications

Human posture information extraction plays an important role in the field of computer vision such as behavior recognition and motion capture.However,when using traditional communication methods to transmit posture information,because the traditional communication needs to transmit the original image data,it has high requirements for the bandwidth and quality of the channel,so it is easy to appear that it can not work normally in bad channel environment.Due to the mode of understanding before transmission,semantic communication can complete the task of posture information extraction only by using semantic feature information,and has strong adaptability.In this context,this paper studies the human posture information extraction algorithm for semantic communication,which has important application value.Based on Openpose network,this paper proposes a semantic communication model for human posture information extraction.The sender uses semantic encoder to extract semantic features,and the receiver uses semantic information to extract posture information in semantic decoder.In order to reduce the computational complexity of the model,the model is lightened by deeply separable convolution and residual network.The simulation results show that under AWGN channel,the proposed semantic communication model has better performance than the traditional communication scheme in the information extraction of human key points,especially when the SNR is less than 0dB,the traditional communication scheme can no longer achieve the attitude information extraction,while the semantic communication scheme can still complete the task well;At the same time,the lightweight work of the model also achieved good results.In experiments with a signal-to-noise ratio of 20dB,the lightweight model greatly reduced the overall computational complexity of the model with minimal performance loss.Based on the self-coder structure in the joint source-channel coding technology,this paper proposes a semantic communication model using reconstructed images to extract attitude information.In order to improve the adaptability of the system to the complex channel environment,an optimization scheme is proposed to improve the model structure and the joint training of the signal-to-noise ratio by using the noise-reducing selfcoder.The simulation results show that the semantic communication scheme has better attitude information extraction performance in Gaussian white noise channel and Rayleigh channel;In terms of image reconstruction performance,when the signal-to-noise ratio is 0dB,semantic communication has a significant gain on PSNR compared to traditional communication;In the ablation experiment,the optimized scheme using signal-to-noise ratio joint training showed an overall improvement in attitude information extraction performance compared to the single signal-to-noise ratio training schemes of 5dB and 20dB;Finally,the proposed semantic communication scheme also has better performance in the high-level intelligent task of using posture information for posture recognition and classification,and has great application value for the practical application of human posture information extraction.