Design And Implementation Of Embedded Image Processing Platform

Since the 1950 s,with the computer science development rapidly.The application of digital image processing is also realized quickly on the computer.At present,with the continuous development of the semiconductor industry,the processing performance of computers is also improving.During this period,a large number of image processing algorithms based on computer platform appeared.The DSP digital signal processor produced by TI is a type of processor specially designed for digital signal.For example,TI DaVinci series multi-core multimedia processor TMS320DM6467 T,the chip adopts the architecture of ARM core and DSP core integrated on one chip,a large number of matrix operations in image processing are accelerated on DSP multipliers and other hardware resources,this greatly improves the processing speed of embedded platform in specific scenarios.However,with the rapid development of semiconductor technology and embedded platform,the processing performance based on DSP platform can no longer meet the requirements of high resolution and high color space for real-time image processing.With the increasing demand of embedded image processing performance,the professional graphics and image processor based on multi-core has been widely used which named Graphics Processing Unit(GPU).The embedded image processing platform used in this thesis is the GPU-based image processing platform launched by NVIDIA Jetson Nano.Based on this platform,this thesis mainly completes the following and part of the content of the work:(1)The basic development environment based on Jetson Nano platform was built,and the development environment of software such as Opencv,Pycuda and Jupyter was installed.Appropriate power supply,camera CCD module and peripheral modules such as device network card were selected.The Jetson Nano software image was made,with an Intranet mapping function installed for remote operation of the device,etc.(2)Some basic image algorithm logic was implemented on PC computer,the correctness of the algorithm was verified and the algorithm was basically optimized.The embedded platform transplantation was carried out for the image algorithm after verification simulation,and the algorithm was run on the ARM core of Jetson Nano.(3)The related algorithm is transplanted into the Ubuntu Linux system,the algorithm for the part of the GPU acceleration operation can be implemented on GPU device of transplantation,maximize the real-time performance of the algorithm,and compared with ARM core to the operation efficiency,but also to a wide variety of programming language(python,c + +)a comparison of the test different programming languages in the differences of time efficiency,etc.(4)After the GPU-accelerated operation of the basic algorithm is completed,the basic algorithm modules involved in the target tracking algorithm are packaged and coordinated to realize the logical function of the target tracking algorithm,and the design part of the relevant algorithm is further optimized at the bottom of the algorithm to improve the real-time performance of the algorithm.