Human Pose Estimation Based On Deep Convolutional Neural Network

With the development of artificial intelligence,computer vision,natural language processing,speech recognition and other emerging technologies are also developing rapidly.As a cutting-edge research direction in the field of computer vision,human pose estimation has a very wide range of applications in video surveillance,human-computer interaction,virtual reality,and video retrieval.Deep Convolutional Neural Network(DCNN)is a deep learning model that is commonly used to process and analyze visual images.In this paper,under the framework of Mask R-CNN,a feature extraction network is designed,which combines the deep level feature extraction capabilities of Dirac Nets and introduces Detail-preserving Pooling operations.Then,combining the advantages of integral pose regression to avoid quantization errors,the Ranger optimizer and Mish activation function are introduced for algorithm optimization,and a model of human pose estimation is designed.The main work completed in this article is as follows:First,a human pose estimation model based on Dirac Nets and Detail-preserving Pooling(DDHPE)is constructed.By analyzing the principle and structure of Dirac Nets,Dirac Nets with autonomous learning ability is introduced as a feature extraction network.On this basis,in view of the problem that the general pooling layer is easy to ignore detailed features,the feature extraction network can be further optimized by using the DPP operation that can amplify the spatial variation.Taking the optimized feature extraction network and feature pyramid network as the backbone network of Mask R-CNN,a human pose estimation model with deep feature extraction capability is proposed.On the CIFAR-10 and CIFAR-100 datasets,the effectiveness of the feature extraction network combining the Dirac Nets and DPP operation was verified.At the same time,the experimental results on the MSCOCO2014 and MPII datasets show that the proposed human pose estimation model improves the prediction accuracy.Then,the optimization algorithm of human body pose estimation model DDHPE is proposed.For human pose estimation tasks,integrated pose regression that can avoid quantization errors is introduced to optimize the human pose estimation model based on Dirac Nets and Detail-preserving pooling,and the DDHPE model based on integrated pose regression(IPR-DDHPE)is constructed.In addition,the IPR-DDHPE model is further optimized by using the Ranger optimizer that can optimize the gradient propagation of the network and the Mish activation function that optimizes the network structure.Experimental results on the MSCOCO2014 dataset and MPII dataset show that the optimized RM-IPR-DDHPE model improves the accuracy of human pose estimation.Finally,a prototype system for human pose estimation is designed and implemented.The human body pose estimation task is applied in video detection,and the video dataset Human3.6m is used as the processing object.The human key point detection demonstration system is designed under the Django application framework.The prototype system realizes the detection of the target person by detecting every second in the video,and on this basis,estimates the key points of the human posture.The paper has 27 drawings,9 tables and 92 references.