Research On The Low Light Level Image Enhancement And Template Matching Based On FPGA

Target tracking technology has gone deep into people’s daily life, and it hasbeen widely used in agricultural production, transportation, security and other fields.However, with the expansion of domestic information demand in low lightenvironments, the low light level image, as the main tool to transmit information, hasbeen applied more widely. It is urgent to research on low light level imageprocessing techniques. Low light level image has characteristics as follows: therange of gray levels is centralized; the spatial correlation between adjacent pixels ishigh; the gradation has small changes. The information of target, background, detailsand noise in image is in the narrow concentration range of gray levels, and aftertransmission and conversion, image quality will be further reduced and the imagewill be mixed with some noise, it will cause the problem that target tracking will bedifficult to work, so it is necessary to enhance the image, then achieve the result ofmatching tracking. It can clearly be seen that research on low light level imageenhancement and matching algorithms, which is adaptively and meet real-time, hasgreat practical significance and value.Using the characteristics of low light level image, and combined withconditions for hardware implementation, this article has researched home and abroadvarious types of low light level image enhancement and matching system design schemes, and then designs a system which suitable for low light level imageprocessing. The result of this study provides strong theoretical and technical supportfor further study on low light level image processing system.Low light level image is affected by noise generally, and the gray levels areconcentrated in the low gray level side, which has made negative effect on matchingtracking, and the pretreatment of the collected image has direct influence on thematching results. The first step for image preprocessing is that selected the3×3Gaussian filter to remove noise; the second step is to enhance the image. As the lowcontrast of image, the contrast stretching enhancement methods has been selected.This article has proposed two algorithms: multi-sub histogram equalization andhistogram equalization base on GMM. By dividing the histogram, the gray levelswill be more uniform, and the texture details can be retained, than we can get thefinal enhanced image with rich information, and the target will be more obviously.The image after enhancement is conducive to related matching processing.Through analysis the MAD and NCC algorithm which theory is mature, an improvedalgorithm is proposed, in order to reduce the amount of computation, extraction lineis used on template. In this paper, System Generator is adopted to develop system,by constructing the corresponding module, algorithm is verified in simulinksimulation environment and it can automatically generate verilog hardwaredescription language. Then transplant it to the FPGA for implementation. Thisarticle selects the XC5VSX50T FPGA which in the series of Virtex-5. Result showsthat the system can meet the image enhancement and matching processing inreal-time.