Human Pose Recognition Based On Blurred Images

Currently,the aging of the population is a focus topic all over the world.China's aging population has the characteristics of large base and fast growth.As the family structure shrinks rapidly,adult children usually do not live with the elderly,resulting in an increase in the number of empty nesters.In order to ensure the safety of living alone,the importance of intelligent monitoring is becoming more and more prominent.By acquiring images and understanding the video content,the intelligent monitoring technology grasps the life dynamics of the elderly at home in real time,timely discovers the abnormal situation of the elderly and feeds back to the children and service organizations to ensure the personal safety of the elderly living alone.The existing intelligent monitoring for the family uses a clear camera,and children or related old-age care institutions can find abnormal conditions in time by acquiring video images.However,the monitoring of clear video often makes the monitored person unacceptable.In addition,the image may be intercepted or cracked by illegal elements resulting in privacy leakage.This paper proposes a human pose recognition method based on fuzzy images,which minimizes the violation of the privacy of the elderly under the premise of protecting the safety of the elderly.Ordinary blur maps would hide the details of life,in the case of video images being illegally stolen by others,technology can restore part of the image relying on image restoration.This paper adopts the physical blur method,which can not restore the clear image through calculation.Its irreversible characteristics shield the security risks caused by video leakage from the root.While protecting the privacy of the elderly,the different human poses in the blurred image are identified to form a complete set of human pose recognition solutions based on blurred images.The main research work of this paper includes the following three aspects:(1)Study existing image blurring and restoration algorithms.In this paper,the image is blurred by adding the physical method of non-smooth lens to the ordinary lens.Compared with the blurred image,the design of fuzzy lens has the characteristics of randomness and information reduction.Therefore,the generated blurred image cannot be restored according to algorithms.It protects the privacy of the elderly,and blocking the security breaches caused by video content leakage;(2)The key point information of different parts of the human body in the video is extracted by the single person pose estimation model as the input of dynamic bone modeling.Based on the single-person pose estimation of the convolutional pose machine,the stacked smallsized convolution kernel replaces the large convolution kernel,deepens the network structure and reduces the parameter size,and obtains image features with better expressive power.It has a good positioning effect on the key points of different parts of the human body under the blurred image,creating a good precondition for the attitude recognition model;(3)The human body pose recognition method uses the dynamic skeleton modeling of the space-time graph convolution network,and constructs the spatial-temporal map by using the results of single-person pose estimation.Traditional bone modeling methods require manual division and definition of traversal rules,while spatial-time map convolution relies on the natural connection of human joints,with stronger expression and generalization ability.Through the previous physical fuzzy method,the behaviors based on the blurred image are collected to achieve the purpose of different human pose recognition.According to the work of this paper,fuzzy images can be applied to intelligent monitoring,identify the sudden behaviors that occur in the elderly living alone,such as falling,waving hand for help and so on.It would be feedback to children timely or elderly care institutions.At the same time,the use of irreversible physical blurring to protect the privacy of the elderly,to avoid the security risks caused by malicious recovery of video leaks.