Research On The Application Of Gaussian Process Regression In The Prediction Based On Image And Video Data

With the continuous development of machine learning methods in recent years,the Gaussian process(GP)model has attracted more and more attention and favor of the experts and scholars.GP model can be applied to both regression and classification.In this thesis,the application of Gaussian process regression(GPR)in the prediction of image and video data will be studied.In recent years,the prediction of image and video data has become a research hotspot in the field of computer vision,new technologies emerge in an endless stream,and more and more enterprises apply these technologies to practice production.Although a lager number of academics and enterprises have conducted in-depth research on the prediction of image and video data,and achieved fruitful results,many key problems still need to be solved.In this thsis,we take the Gaussian Process Regression as the research object,and propose an improved application in the scene of image and video respectively.In video prediction: this paper takes the traffic statistics in video as the research object,and combines the dynamic texture model(DTM)and the GPR.The DTM is used to segment the motion foreground and the static background in the video sequence.By extracting the features from the foreground segmentation segments,the GPR is used to learn the mapping relationship between features and traffic flow,and finally make regression prediction for the vehicle number.From the overall perspective of vehicle flow,the method avoids missing and false detection caused by single vehicle counting,and achieves the goal of reducing errors and improving statistical accuracy.Second,In the prediction of the image: this paper takes the age estimation of face image as the research object,using active shape model to locate the 68 feature points in each image,then,combining the coordinate information of feature point and face measurement template,obtain facial geometric proportion features,then,using GPR to learn the mapping relationship between the geometric proportion features and the real age value,and make regression estimation for the age value.In addition,this thesis also focuses on the analysis of eigenvalues in the eigenvectors.In the proposed scheme,by comparing the output results with a large number of experiments,a better feature combination is selected to reduce the dimension of the eigenvector,thus reducing the computational complexity of the regression algorithm,and also guarantee the accuracy of the regression estimation.In this paper,the relevant simulation experiments are designed for the proposed scheme,and the experimental results are compared with the existing algorithms.The experimental results show that the proposed scheme has a significant improvement in accuracy,in prediction of image and video data,GPR has satisfactory performance.