Research On Target Detection And Fusion Display System Based On FPGA+ARM

Target detection and scene fusion have been studied and applied widely in intelligent transportation,security,medical,consumer electronics and other fields,and these studies are mainly implemented in PC or in a single-architecture system platform,resulting in the restriction of performance and power consumption.Therefore,this paper constructs an image processing platform combined with ARM and FPGA,and designs related drivers and communication mode between platforms,which ensures the reliability and stability of high-speed communication between platforms.The implementation of the image processing algorithm is mainly implemented in the following two aspects:(1)Aiming at the specific problem of vehicle target detection,based on the deformable part model(DPM)algorithm and the hardware characteristics of the image processing system,a real-time target detection algorithm is implemented on the image processing system.Considering the information redundancy in the computation of image Pyramid,a fast way to build pyramid is put forward based on the derivation relationship between adjacent layers in pyramid.According to the hardware advantages of FPGA logic resource richness and parallel architecture,we design multi convolution kernel parallel computing,high-speed data stream ring buffer and efficient data reuse module,which significantly reduces the running time of the program.A lot of experiments have been done for the unmanned aerial vehicle scene and vehicle scene.The experiments prove that the real-time and accuracy of the system designed in this paper is good.(2)In the study of multi-source scene fusion,a weighted response pyramid fusion model is built on the basis of the image pyramid algorithm and HOG features.On the basis of the original pyramid fusion algorithm,a response pyramid's construction process is proposed.The feature map of the response area is constructed by the way of up sampling,then the multi-source image pyramid is fused,and finally reconstructed to the output image.It is proved that the fusion effect of the multi-source image is improved obviously compared with the original pyramid fusion algorithm.The embedded system proposed in this paper has the advantages of low overall power consumption,strong processing performance,small volume and easy deployment.In most realistic scenarios,it can give full play to the effect of the algorithm,and has improved significantly compared with the previous system.