Research On Human Pose Estimation Methods In Natural Scenes

In recent years,with the significant progress of deep convolutional neural network in the field of computer vision,especially in the estimation of human posture through convolutional neural network,good results have been achieved,and significant progress has also been made in the field of image processing.Images captured in natural scenes may be disturbed by various noises(such as haze,raindrops,and white Gaussian noise,etc.),which seriously affect the observability of the characters in the images.In this paper,image dehazing,image denoising and human posture estimation in advanced computer vision are studied.In view of the poor accuracy of model parameter estimation and color distortion in most image de dehazing algorithms,a end-to-end dense connected dilated convolutional neural network is proposed.First of all,this paper uses multi-layer dense connection structure to increase the feature utilization of the network and avoid the gradient disappearance when the network deepens.Secondly,by using the dilation convolution with different dilation rate in the dense blocks,the network can fully aggregate the context feature information without losing the spatial resolution,and avoid the generation of mesh artifacts.Finally,in order to improve the ability of the algorithm,we divide the network into several stages,and introduce the side output module in each stage,so as to obtain more accurate feature information.The experimental results show that the proposed dehazing algorithm has a good dehazing effect on both the synthetic data set and the real data set.The recovered color is closer to the ground truth,and the quantitative evaluation indexes peak signal to noise ratio(PSNR)and structural similarity index(SSIM)are better than other comparison methods.Aiming at the problem of noise and blind denoising robustness in digital image acquisition,a residual dense connection extended convolutional neural network with high density connections is proposed.In this network,the residual-intensive expansion module is the backbone of the network.Firstly,different expansion rates are used to enhance the receptive field and improve the utilization rate of spatial context feature information.Secondly,it adopts the form of dense connection to ensure the authenticity of extracted features from the shallow layer.Finally,the idea of residual error is introduced in the dense connection structure to further improve the validity and authenticity of feature information extracted from each layer of the network.In order to further improve the performance of the network,the low-level feature information in the shallow layer is used for auxiliary training to improve the processing capacity of the network on unknown noise.The experimental results show that the proposed denoising network has better PSNR than other comparison methods in BSD68 and Set12 data sets,and achieves better results in the case of unknown noise level.Most of the current attitude estimation algorithms mainly change the network structure to some extent,ignoring the active role of attention mechanism and the influence of noise.This article tries to add attention to the Bottleneck Structures of the Resnet-50 network by adopting Simple Baseline network as the basic framework and tend to use the two attention Bottleneck structures,Namely,Simple Baseline + Squeeze-and-Excitation and Simple Baseline + Convolutional Block Attention Module,as the network infrastructure.By modifying the weight of feature map in each bottleneck structure of the network,the recognition effect of human posture estimation network is improved continuously.The experimental results show that the improved body posture estimation network is superior to other comparison methods in the recognition effect of two standard body posture estimation data sets,COCO and MPII.