Research On Human Posture Estimation Method And Application Based On Cloud-Edge Collaboration

Human pose estimation refers to locate all keypoints of the human body and build human body representation(e.g.,skeleton model)from input data such as images and videos,which is the basic key technology of motion analysis,human-computer interaction and many other applications.Although convolutional neural network-based methods have shown good performance in human pose estimation tasks,there are still some challenges when these methods are applied in practice.First,how to ensure the performance of the method in the face of complex environments in reality(such as crowding,shading changes,occlusion,etc.).Second,most of the current methods tend to improve network complexity to ensure the improvement of method performance,while complex network models are difficult to be directly applied to scenarios such as limited computing resources and time-delay sensitive.Third,in the cloud-based environment,the training of complex network models requires more computing and broadband resources,resulting in higher latency and reduced data security.In response to the above issues,this article has conducted the following research work:(1)A human body pose estimation algorithm based on MLP-Mixer was proposed.Aiming at the problem that human posture is difficult to detect in complex environments,a new model,MLPPose,was designed in this paper.By combining the MLP-Mixer layer with the convolutional token embedding,the dependencies between different key points were obtained,and the global dependencies between key points and scenes were effectively captured,thus improving the detection ability of the model for complex junction nodes.Experiments show that compared with methods such as convolutional neural network and transformer,this algorithm has better performance and less computation and parameters.(2)A new lightweight model,Lite Net,is proposed.Aiming at the problem that the model has high complexity and is difficult to directly apply to resource-constrained terminals,a lightweight module named Long-Short Range Convolution(LSRC),which can extract global dependencies of images,was designed on the basis of MLP-Mixer.Convolution Fusion Deconv is designed to enhance the capability of the model.Experiments show that compared with self-attention mechanism,Lite Net can achieve better performance with lower model complexity.(3)A cloud-edge collaborative human posture estimation method is designed.Aiming at the problems of high resource consumption and high detection delay of human posture model in cloud computing mode.In this paper,edge computing,cloud computing and lightweight human posture estimation model are integrated,and a human posture detection scheme based on cloud-edge collaboration is designed.The scheme combines the advantages of cloud center and edge nodes and lightweight models.It can effectively reduce the consumption of computing resources,improve the detection speed,and ensure data security while meeting the requirements of resource-limited terminals and detection speed.In summary,this paper aims to study the human pose estimation method and application of cloud-edge collaboration,and the proposed method has a progressive relationship: firstly,the MLP-Mixer method can improve the detection performance of the model against crowded,shading,occlusion and other complex environments.Secondly,the model based on long distance convolution is lightweight based on the previous method,so that it can be better used in resource-limited devices.Finally,the detection scheme based on cloud-edge collaboration is applied to the specific deployment of the former lightweight algorithm,reducing resource consumption and detection delay.