A Deep Learning-based Visual Privacy Measurement System For Service Robots

Privacy measurement and risk assessment is an important support for privacy protection technology.Combined with the characteristics of deep learning technology which can efficiently extract image data features,this paper takes the visual image in smart home environment as the research object and service robot as the carrier to study the scientific problem of human body privacy measurement of visual image in complex background:Aiming at the problem of visual image privacy measurement in smart home environment,a visual privacy measurement model based on set pair theory is proposed,The calculation framework of visual image privacy is built;aiming at the problem of information extraction of visual image in the model,the human body privacy measurement algorithm of deep lab V3 human body semantic segmentation based on Criss Cross fusion is proposed;aiming at the problem of low precision of semantic segmentation in the privacy measurement model,the human body semantic segmentation algorithm based on Euclidean distance measurement is proposed,which improves the accuracy of the model And robustness.In addition,the experimental platform of service robot is built,and the visual interface of human privacy measurement system is designed.The main research contents of this paper include the following aspects:(1)a human body privacy measurement model and algorithm based on set pair analysis theory is proposed.In the image privacy measurement,the size of the privacy value is affected by many factors,which makes the visual image privacy measurement model difficult to determine and inaccurate.To solve this problem in the privacy measurement model,this paper takes the information sensitivity in the image as the main factor,establishes the sensitive information connection number based on set pair analysis theory,realizes the image information privacy digitization,and establishes the human body privacy measurement model of visual image.In addition,combined with deep learning semantic segmentation to extract image information features,a human privacy measurement algorithm based on Criss Cross deep lab V3 visual image is designed.Experimental results show that the proposed model and algorithm have good robustness and feasibility.(2)A human semantic segmentation algorithm based on Euclidean distance is proposed.In the privacy measurement model of visual image,the computational accuracy of the model is affected by the effect of deep learning human semantic segmentation,which leads to the accuracy of the whole measurement process of the model.In order to improve the accuracy of the model,this paper analyzes the operation principle of inner product similarity calculation of neural network.From the point of view of low cohesion and high coupling,compared with common distance measurement methods,a human body semantic segmentation network based on Euclidean distance measurement classifier is designed,and the effectiveness and feasibility of this method are proved.Experimental results show that compared with the original model,this method has better accuracy on multiple data sets.(3)A visual image human privacy measurement system based on deep learning is designed.By integrating the theoretical results of(1)and(2)and the robot hardware platform,the visual image human privacy measurement system is designed and implemented.