Research On Inertial-aided Image Motion Deblurring Algorithm In Personal Navigation System

In PNS,the miniaturization and intelligence of mobile devices command higher image clearness in aspect of smart photography,especially for image motion deblurring processing.In research of image deblurring algorithms,the main image processing algorithms are one-frame image processing algorithms and multi-frame image processing algorithms.Based on whether PSF is known or not,image restoration problems can be classified as blind image restoration problems and non-blind image restoration problems.In the thesis,considering capturing environmental information images in PNS dynamically,blind multi-image restoration problem is the main research problem and inertial-aided image motion deblurring algorithm is the main research algorithm.In research of inertial-aided image motion deblurring algorithms,Joshi first introduced this conception by realizing deblurring of shaking camera in 2010.Meanwhile,Kodak Research Team of China introduced an algorithm extracting head file to realize motion deblurring.In 2011,Myung Hwangbo et al.,coming from Carnegie Mellon University,realized auto-calibration and GPU implementation of motion cameras by tracking image features using gyroscopes-aided algorithm.In 2013,?indelá? first realized it in smart mobilephone of Android System using the same algorithm as Joshi.Then Daniel Cremers’ team realized it in i Phone of i OS system,and introduced VIO conception using IMU sensors-aided camera to do velocity and distance measurements.Comparing with traditional Visual Odometry,it has higher precision and better real-time requirement.In the thesis,inertial-aided image motion deblurring algorithms have been researched and analyzed based on inertial information matching multi-frame image sequence.The main work and conclusions are as follow.First,based on inertial sensors,especially MEMS gyroscopes and accelerometers of three-axis,either MTi-G-700 of Xsens Corporation or Personal Mobile Devices of Android mobilephone and pad,aided image motion deblurring,new mathematic model has been built,and the calculation of camera’s position,pose,velocity is based on the data of inertial navigation in short time,in order to revise pixel points’ offset position.Then,it can be used in motion deblurring processing,and other image processing such as denoising is done.Consequently,image motion deblurring problem has been solved,and new designed algorithm has been proved to be scientific by MATLAB simulation.Second,implementation of three operation systems,Windows,Ubuntu and Android,motion deblurring image restoration algorithm is analyzed.The algorithm is proved by MATLAB simulation of Windows System,and it is also realized to catch imagesequence and MIMU data of ROS in Ubuntu System.Besides,PTAM is also analyzed in it.Most importantly,the algorithm of Weiner Filter is realized in mobile platform,such as mobilephone and pad by using JNI and NDK to realize Java and C++ mixed compilation of Android System.In addition,it can get and save MEMS sensors’ data of motion information as well as image sequences synchronously.Third,the quality of all image processing results in three different platform are analyzed qualitatively and quantitatively.The evaluation values are basically corresponding to normal human vision.Besides,the effect of iamge quality on environmental information images feature matching problem of Personal Navigation System is discussed.Last but not least,quanterion error is analyzed by using gyroscopes’ disconcrete data.Three algorithms,single sample algorithm and uncompressed frequency-series coning algorithm of three-sampling as well as explicit frequency shaping algorithm of three-sampling,are analyzed and compared under coning environments.Additionally,theoretical equation is deduced relating to the variances of quantization errors and attitude errors.It is shown from the derived formula and results of Monte-Carlo simulation that low-pass filtering and higher attitude update frequency can decrease the influence on attitude error caused by sampling quantization error,apart from increasing resolution of gyros’ angular increments.Image quantization problem is also discussed in the thesis.