Research On The Infrared Ground Moving Target Detection Technology Based On Biological Vision

Infrared imaging technology has been widely applied in the field of infrared imagingguidance, infrared reconnaissance and surveillance due to its advantage of high resolution,good elusiveness and strong anti-interference capacity. However, most of infrared imagesare affected by the disadvantage of low SCR, blurred target edge and low contrast betweenthe target and the background. Moreover, the infrared detectors are often in state ofmovement results to the dynamic background, which increases the difficulty of targetdetection. As a result, Infrared target detection is a key issue of the infrared imagingtechnology. The researchersin the field of neurophysiologyhave found the lateral inhibitionmechanism within the visual system of the Limulus and the elementary motion detectionmechanism in the visual system of the drosophilas. The lateral inhibition mechanism couldextract the edge of the target by enhancing the image contrast, while the elementary motiondetection mechanism could detect the moving target under the dynamic background anddetermine the direction of the movement.In this thesis, the mechanisms of lateral inhibitionand elementary movement detection are meticulously studied and have been applied in themoving infrared target detection under the dynamic background. The main contents andcontributions of this thesis are listed as follows.Firstly, a lateral inhibition model based on the double butterflytype nonclassicalreceptive filed is proposed to solve the problem of blured edge of the target in the infraredimages. The lateral inhibition mechanism in the limulus is analyzed firstly. The inhibitingfactor and the inhibiting threshold are the key factors in the Hartline inhibition model, andthe effects of these two factors to the response of the model are evaluated then. A doublebutterfly type nonclassical recetive field model which consists of inhibition area andfacilitation area is established to solve the problem of inhibition within the edges.Inaddition, the anisotropicinhibition coefficients are weighted by four factors: range,grayscale contrast, orientation contrast and direction constrast, which is beneficial todistinguish the target from the background. In order to verify the proposed model, theactrual infrared images are used to evaluate the performance of the model compared with other three models. The simulation results showe that the proposed model could inhibit thebackground maximally, give a good performance of target edge detection and greatenhancement of the infrared image contrast.Secondly, the elementary motion detection model proposed by Reichardt has beenanalyzed thoroughly. The mathematical model of EMD is deduced. The parameters of themodel including the space between the receptors, the delay time, velocity, timespan, spatialfrequency and temporal frequency are discussed carefully. In the simulation, the effects ofthe parameters stated above are evaluated to the output the the EMD. The analysis andsimulation lay a solid foundation for the work of modification to the EMD.Thirdly, for detecting aninfrared target in a dynamic background, a two-dimensionalmotion detector model (2DMD) is proposed based on the region growing theory. The2DMD model is sensitive to the movement, which could be used to locate the seed point byfinding the largest output of the response. The difference of the meanvalue and the maxvalue is used as the growing rule. Then the target could be detected by morphologicalreconstruction. The proposed method is verified by actrual data. The results showed that theproposed method could filter the dynamic background interference and detect the targetwith higher SCR and faster speed compared with the other three motion detection methods.Finally, According to the latestbiological neural theory of motion detection, a modifieddual-channel motion detector model (MDCMD) is proposed for adapting to the changing ofatmospheric absorption of infrared radiation.Combining the edge detection scheme with themotion detection scheme, the basic structure of the MDCMD model is designed detailedly.The input signals is preprocessed by the LMC stage which could weaken the effect of theabsolute luminance of the images. And then the preprocessed signal is splited into twochannels and then computes the output response by the weighted EMD. The mathematicalmodel is deduced and verified by the experiment. The results illuminate that the MDCMDcould detect the target effectively and determine the direction of the movement accuratelywithout being affected by the variation of the grayscale.