Research On Linux Embedded Vision System Based On ARM

Embedded vision system has important application prospects in industrial detection,medical,consumer electronics and other fields.However,for different embedded hardware and software platforms,there is no universal transplantation and application scheme.In view of the above problems,this thesis studies a Linux embedded vision system application scheme based on ARM,and applies image processing,object detection,object classification and other vision technologies in the embedded platform.The above research has a good role in promoting the sustainable development of embedded technology and computer vision technology.The main research work of this thesis is as follows:(1)A Linux development environment based on Exynos4412 processor and i Top4412 development platform is built.And the Open CV library for Ubuntu system and embedded platform is recompiled.(2)The underlying working principles of algorithms including image filtering,image edge detection,graphic geometric transformation,image feature point detection and matching,etc.are studied in detail,and the research work on the algorithm principles and implementation methods of SVM support vector machines has been completed.(3)Developed a vision system software based on the "scene-view-object element" framework and the main application plus function plug-in design method.The design and development of various image processing plug-ins such as image filtering plug-in,graphics transformation plug-in,image thresholding plug-in,object feature detection and matching plug-in,etc.have been completed.The functions of each plug-in developed are independent and do not affect each other.(4)Designed and developed a machine learning object classification plug-in,which realizes the feature extraction and classification function of screws,nuts and washers in unknown images.The plug-in design process completely follows the machine learning vision processing process,including image preprocessing,object segmentation,feature extraction and model training,etc.,providing a complete solution for subsequent research and development work related to machine learning vision.(5)The transplantation of the developed vision application software to the ARM embedded platform has been completed.The experimental test results show that the vision application software transplantation algorithm developed in the thesis runs stably on the built ARM embedded platform and realizes the expected design function.A complete set of embedded platform construction and design solutions proposed in the thesis has a good reference value for the development of embedded vision systems and image processing or machine learning functions on embedded devices.