Research On Aerial Infrared Small Target Detection And Hardware Acceleration

The infrared small target detection under complex background is an important research content in infrared image processing field. At present, the research about the aerial infrared small target detection has made many excellent achievements. But due to the complexity of the background and the lack of shape texture features, infrared small target detection under complex background can not achieve good results. In addition, the data volume is larger and larger, the real-time processing request put forward higher requirements on the image processing system, so the acceleration implementation of processing algorithms has become a inevitable trend. This article made some improvements on the traditional infrared small target detection algorithm to improve the algorithm's adaptability under complex background, the design of hardware acceleration implementation of the enhanced algorithm also will be described. This article main research content including:(1) The Local Contrast Method(LCM) algorithm has some deficiency for infrared small target detection under complex background, on the analysis of pixel gray value characteristics in small target neighborhood, we define a parameter called the local biggest difference to improve the detection performance of the algorithm. This parameter can use more information of neighborhood pixels; the local numerical consistency as well as the numerical range dissimilarity of maximum difference and local minimum contrast has been making full use of in the process of calculation, which can achieve more effective target enhancement and background suppression. Experiments show that, the enhanced algorithm greatly improved the infrared small target detection performance in the complex background compared with the traditional algorithm.(2) Since the traditional serial processor has deficiency in power consumption and processing speed in the image processing application, we reaearch on the image algorithm hardware implementation based on FPGA. The improvement algorithm belongs to the bottom of the image processing, and it can be split into independent function modules, the simple operation structure and the high repeatability make it suitable to be parallel processed by FPGA. FPGA can map the function modules into independent circuit structure, data stream processing approach and the use of pipeline can support the algorithm operates in low clock frequency, which will reduce the power consumption of the system. Experiments show that, the FPGA can play a role of image processing accelerator compared with the traditional serial processors.(3) In order to achieve the need of high resolution target tracking performance, we designed a collaborative image processing system based on DSP+FPGA. The bottom hardware communication and hardware acceleration enables the system can achieve real-time processing of high-definition video. Experiments show that, under the collaborative work of FPGA and DSP, the system can complete the real-time detection and tracking task of small target in high resolution image.