Human Key-points Location Algorithm Based On Deep Learning

The human key-points location algorithm is a very important branch in the field of computer vision,and has a wide range of application scenarios in human-computer interaction,audio-visual entertainment,and beauty cameras.The human key-points studied in this paper include facial key-points and human joint points.In order to make the key-points accurately and quickly locate on devices with limited computing resources,two human key-points localization algorithms based on deep learning are proposed.Firstly,in the facial alignment algorithm,we propose a convolutional neural network based on Inception structure instead of multi-cascade neural network structure,which effectively reduces the computational cost.In order to improve the accuracy of the network,we use the multi-branch structure to increase the network width.At the same time,we introduce the residual structure to realize the fast and accurate positioning of the facial key-points.Secondly,in the human pose estimation algorithm,we combine the depthwise separable convolution with the hourglass network structure.This combination method can greatly reduce the parameter amount of the network.In order to reduce the loss of spatial information,we use the heat map as the output of the neural network.This structure can effectively improve the accuracy of positioning.In order to improve the positioning speed of facial alignment,we proposes a key-points location method based on Inception structure.This method cancels the traditional multi-cascade structure,and use a wider and deeper neural network to achieve the location of facial key-points,it can greatly improve the operation speed of the neural network.In order to compensate for the error caused by a single neural network,we design a module based on Inception structure.Firstly,the module uses multiple convolution branches to convolve the feature maps,then superimposes the feature maps of multiple branches,and finally introduces the residual structure to reduce the network gradient dispersion.The improved Inception module has better recognition of image translation,scaling,deformation,etc.,and increases the adaptability of the network.The experimental results show that the proposed algorithm can improve the positioning speed while ensuring the accuracy of key-points positioning.The algorithm achieves stable positioning in the face of different poses and different environments,and has high robustness.In order to improve the speed of human pose estimation,we propose a key-points location method based on depthwise separable convolution.The method combines the depthwise separable convolution with the hourglass network structure to improve the positioning speed of the network.The operation of depthwise separable convolution improves the speed of positioning by reducing the parameters of the network.The hourglass network structure effectively reduces the computational complexity of the network by compressing the resolution of the feature map.This algorithm uses the heat map as the output of the network which can save more spatial information and improve the generalization ability of the network.The experimental results show that the human pose estimation algorithm studied in this paper has achieved satisfactory results in both accuracy and operation speed,and has high robustness in different poses and environments.