Research On Dim And Small Targets Detection And Tracking Algorithms In Sequence Image

In order to improve the ability to detect and track dim small targets under low signal-to-noise ratio(SNR)and strong noise interference,this paper first utilized preprocessing to suppress the background and enhance the target signal based on the motion correlation of target in sequential image.Then,two different detection algorithms are proposed for different dynamic scenarios with different SNR.Finally,on the basis of the previous detection algorithm,in order to correlate the multi-frame target trajectories,an improved dim small target tracking algorithm is proposed to solve the problem that the target is disturbed by strong noise in dynamic scenes.The main research works are as follows:(1)In terms of background modeling and energy enhancement.With regard to the poor modeling effect in the face of non-stationary edge contour region of traditional background modeling,this paper first proposed an improved the anisotropic difference filtering algorithm,which effectively highlights the difference between the target and the edge contour region and well preserves the target signal in the differential image mainly by comparing the gradient difference between the target and the background in eight directions in the neighborhood and selecting the mean in the three directions with minimum diffusion function value to filter the image.Then,on the basis of obtaining the difference image,an energy enhancement algorithm is proposed,which integrates the target's motion characteristics of the temporal-spatial.The algorithm makes full use of the motion information of the target in the temporal-spatial,and effectively enhances the target signal.Experiments show that the mean square error of improved anisotropic background modeling in different SNR images of two scenes are less than 10.2,the structural similarity of two scenes are greater than 0.951,and the local SNR gain of two scenes are greater than 10.6.The average gray level of the target and the local signal-to-noise ratio of the image obtained by the energy enhancement algorithm combining the spatial and temporal motion characteristics are 238 and 13.65 dB,which further improves the target signal.(2)In terms of the dim small targets detection algorithm,the dim small targets detection algorithms in dynamic changes and low SNR(SNR < 3dB)scenarios are mainly studied.This paper proposed the dim small targets detection based on self-adaptive caliber temporal-spatial filtering and high order cumulant detection algorithm for dim small target based on motion direction estimation for different scenarios.On the basis of image pre-processing and the misdetection and omission due to a fixed pipe diameter in traditional detection algorithm and in combination with the advantages of scale space theory of obtaining the spatial position and scale of the target,a dim small target detection based on self-adaptive caliber temporal-spatial filtering is proposed,which adopts the scale space theory to modify the size of the search diameter in a self-adaptive and effectively eliminates noisy points and accurately detects the targets.Since the imaging system is susceptible to the variation of external illumination,the image response acquired by the system is non-uniform and the image contains more irregular shading.In addition,the imaging system is disturbed by strong noise.Such factors cause the local SNR of the target to be extremely low.In order to effectively eliminate the non-uniformity of images,this paper proposed a correction algorithm for circular perturbation of sequential image to obtain the image after the sequential correction and additionally proposed a high order cumulant detection algorithm based on motion direction estimation.This algorithm first adopts a multi-frame accumulation technique to enhance the target signal,and extracts a candidate target image in combination with the Poisson distribution theory.Meanwhile,it takes a high order cumulant detection algorithm to further extract the candidate target image.Then it compares the two candidate target images to obtain a common candidate target point.Then motion direction is estimated for the candidate target point to make accumulation along the estimated direction in order to enhance the signal of the dim small target.This algorithm has showed better effects than other algorithms in the detection of sequential image in dim small targets.Experiments show that the dim small targets detection based on self-adaptive caliber temporal-spatial filtering can effectively detect dim small targets with local SNR less than 3dB.At the same time,the local signal-to-noise ratio(SNR)of the target obtained by the the correction method of circular perturbation proposed in this paper is improved more than 2dB compared with the traditional correction algorithm,while the high-order cumulant detection algorithm based on direction of motion estimation achieves better detection than other algorithms.As a result,it can effectively detect dim small targets whose local SNR is less than 1 dB.(3)In terms of the dim small target tracking algorithm,on the basis of the aforementioned detection algorithms,this paper proposed an improved immune genetic particle filter method for tracking dim small targets to mainly solve particle degradation and tracking efficiency problems in the process of dim small targets tracking under strong noise interference.Under the immune genetic particle filter tracking algorithm,the gray feature was first obtained by constructing the spatial position weighting method.Next the motion feature that fully combines the target motion information with the gray probability was extracted,and then integrate the two features into the joint observation model to calculate the particle weight.And modify the reference templates in a self-adaptive manner during the tracking to adapt it to the dynamically changing scenarios.At the same time,in order to improve the tracking efficiency,the KLD algorithm adjusts the number of particles in a self-adaptive manner,can effectively reduce the computational complexity of the algorithm.Experiments show that the proposed tracking method can deal with strong noise,occlusion and cross-motion better than the traditional algorithm in dim small targets tracking results of sequential images in different scenarios.The root mean square error obtained by the algorithm is less than 1.5.At the same time,due to the KLD algorithm,the number of frames processed per second is higher than that of the traditional particle filter by three times,which effectively improves the tracking efficiency.