Research On Hardware Acceleration Algorithm For Target Recognition

The infrared image target recognition technology based on embedded devices has bright application prospects in both military and industrial at present.The traditional embedded development method is mainly based on ARM,DSP or other processor platforms to implement algorithms by software.This method is subject to some objective factors such as power consumption and thermal condition,which usually leads to a problem of running slowly.In recent years,the SoC device,based on intergrating ARM and FPGA in one chip,provides a new approach,so that researchers could easily make use of programmable logic resources to provide flexible and customizable computing assistance for algorithmic programs of the processor system.The purpose of this paper is to use the SoC platform,hardware and software co-design approach to target recognition algorithms for hardware acceleration.The outline of this paper is organized as follows: 1.Owing to the highly integrated structural advantage of SoC,we have designed the overall structure of the system.The ARM works.as the core processor,while the programmable logic works as auxiliary hardware acceleration resources,and external mass storage is used as the data sharing platform.The functions of each subsystem are divided properly;2,According to the characteristics of the application background,this paper presents a specific introduction to the target recognition algorithm to be accelerated by dividing the whole algorithm into three steps: detection of region of interest,regional feature extraction and classification.The theoretical tools and implementation methods involved in each step are explained.;3,Combined with the analysis of the parallelization,data type and resource consumption of the whole algorithm's implementation,we have designed the multi-channel parallel acceleration of the two dimensional Fourier transform and multi-scale Gauss filtering.4,The experiment environment based on physical platform has been built,and the whole system are designed and implemented in this physical environment.After that we have made a brief analysis of the experimental results and improved the optimization.By comparing the effect and speed of the experiment,the function and acceleration of the whole system are verified.The hardware acceleration system based on the Zynq SoC platform can effectively speed up the target recognition algorithm running on the embedded platform,which can be used for practical engineering applications or further research on hardware acceleration.