Infrared small target is easy submerged in the complex background, to improve the ability of detecting target, which by inhibiting the background to enhance the target signal. Focusing on the shortcomings of the isotropic background prediction method, a kind of improved anisotropic infrared background prediction method (IABP) is proposed. According to differences of local gradient character among target region, smooth background region and undulate background region, the edge stopping function of anisotropic partial differential equation is improved. Then the mean of the two least values of the edge as the prediction value of the background. Finally in order to extract the candidate target and reduce the false alarm rate of the real target, which the difference between the background image and the original image is processed. Experimental results show that: 1) improved anisotropy of background prediction for different scenes can obtain good background prediction effect; 2) improved anisotropic background predication for the signal-noise ratio (SNR) was lower than 2.3db could be loyal to the true background of the original image to the maximum extent, presenting a superior overall performance to other background prediction methods.
It’s necessary that higher education experimental teaching reforms on the basis of general education. This paper put forward the experimental teaching reform mode of optical fiber communication in the context of general education. With some reform measures such as improving the experimental content, enriching the experimental style, modifying the experimental teaching method, and adjusting the evaluation method of experimental teaching, the concept of general education is put throughout the experimental teaching of optical fiber communication. In this way, it facilitates the development of students and improvement of experimental teaching quality.
Infrared dim and small target detection plays an important role in infrared search and tracking systems. In this paper, a novel infrared dim and small target detection method based on Boolean map saliency and motion feature is proposed. Infrared targets are the most salient parts in images, with high gray level and continuous moving trajectory. Utilizing this property, we build a feature space containing gray level feature and motion feature. The gray level feature is the intensity of input images, while the motion feature is obtained by motion charge in consecutive frames. In the second step, the Boolean map saliency approach is implemented on the gray level feature and motion feature to obtain the gray saliency map and motion saliency map. In the third step, two saliency maps are combined together to get the final result. Numerical experiments have verified the effectiveness of the proposed method. The final detection result can not only get an accurate detection result, but also with fewer false alarms, which is suitable for practical use.
In the structural local sparse model, every candidate derived from the particle filter framework is divided into several overlapping image patches. However, in the tracking process, the structural characteristics of the target may change due to alterations in appearance, resulting in unstable pooled features and therefore drifting and false tracking. We propose a method to correct the changed part of the target using atoms in the patched dictionary by adding a global constraint. If the target is corrupted, this constraint term will weaken the influence of variation and strengthen the stability of the pooled features. Otherwise, the method is based on the whole target and will protect its spatial continuity. Both qualitative and quantitative evaluations on challenging benchmark image sequences demonstrate that the proposed algorithm has excellent tracking behavior, displaying robustness and stability with little drifting on a target with altering appearance and partial occlusion.
Optical systems such as telescopes are very complex, and their model usually with the uncertainty. To deal with the uncertainty of adaptive optics system and improve system robust stability, the mixed sensitivity H-infinity control has been introduced to design system controller. In order to testify the validity, wavefront aberration correction capability, as well as the robust stability, has been compared between the mixed sensitivity H-infinity controller and the classic integral controller. The computer simulation results demonstrate that the system with the mixed sensitivity H-infinity controller, while can’t guarantee a better correction performance, has greater robust stability than the one with the classic integral controller. That is to say, greater robust stability is achieved at the expense of the correction capability in the system with H-infinity controller. Moreover, the greater the uncertainty is, the more proceeds the mixed sensitivity H-infinity controller will produce. It proves the efficiency of the mixed sensitivity H-infinity controller in dealing with the uncertainty of adaptive optics system.
In the charge-coupled device (CCD)-based tracking control system of fast steering mirrors (FSMs), high control bandwidth is the most effective method to enhance closed-loop performance, which, however, usually suffers a great deal from time delay induced by a low CCD sampling rate. Moreover, mechanical resonances also limit high control bandwidth. Therefore, a tentative approach to implementing a CCD-based tracking control system for an FSM with inertial sensor-based cascade feedback is proposed, which is made up of acceleration feedback, velocity feedback, and position feedback. Accelerometers and gyroscopes are all the inertial sensors, sensing vibrations induced by platforms, in turn, which can contribute to disturbance supersession. In theory, the acceleration open-loop frequency response of the FSM includes a quadratic differential, and it is very difficult to compensate a quadratic differential with a double-integral algorithm. A lag controller is used to solve this problem and accomplish acceleration closed-loop control. The disturbance suppression of the proposed method is the product of the error attenuation of the acceleration loop, the velocity loop, and the position loop. Extensive experimental results show that the improved control mode can effectively enhance the error attenuation performance of the line of sight (LOS) for the CCD-based tracking control system.
Gimbals and Fast steering mirrors (FSMs) are commonly used to stabilize the line-of-sight (LOS) of the electro-optical tracking system mounted on moving platforms .The gimbal is used to restrain the vibration of low frequencies, and the FSM is used to restrain the vibration of high frequencies. The restraining performance of the Electro-Optical tracking system is equal to the multiplication of the restraining performance of the gimbal and the FSM. The vibration of high frequencies is mainly restrained by the FSM, and so the performance of the FSM is very important to the Electro-Optical tracking system. There are two ways to improve the stabilization accuracy and bandwidth of the FSM, one way is to improve the accuracy and bandwidth of inertial sensors, and the other way is to use low weight inertial sensors to reduce the load of FSM and increase the mechanical resonance frequency. And so the inertial sensors of high accuracy, high bandwidth and low weight are the key to improve the stabilization accuracy and bandwidth of the FSM.
Phase diversity (PD) can not only be used as wavefront sensor but also as image post processing technique. However, its computations have been perceived as being too burdensome and it is difficult to achieve its real time application on a PC platform. In this paper, we carried out parallel analysis on the algorithm and task assignments on the heterogeneous platform of CPU-GPU, and then implement parallel programing optimization on GPUs. The optimization strategies of the cost function on GPU are introduced. The process of OTF is improved to make the amount of calcuation reduced by 11% compared to the original method. In order to demonstrate the speedup of PD, two images, 128x128 pixels and 256x256 pixels in dimension, are tested on CPU platform and CPU/GPU heterogeneous platform respectively. The results show the time costs have the improvenments of 13x and 28x for the implementation of PD based on GPU in contrast with that based on CPU.
We propose an improved Lucas-Kanade template tracking method with drift correction, which can be applied in rigid extended object. Due to error accumulation, primary template tracking method leads to template drift and loses object gradually. In order to alleviate template drift, SIFT (Scale Invariant Feature Transform) feature is used to correct the template drift. SIFT feature is invariant to scale, rotation even affine transformation, so, according to matching SIFT key-points between frames, the affine parameters of object transformation can be computed, then the current template position can be obtained by affine parameters and primary template position. The experiment results prove that the improved template tracking method based on SIFT drift correction can more accurately track the rigid extended object and can alleviate the tracking position drifting effectively.
Image fusion method directly affects the stitching effect. If there are moving objects in the images, the same object may
add together after fusion and cause ghosting. A fusion method which can effectively eliminate the exposure difference
and ghosting is essential to image mosaic. The method of optimal seam can eliminate integrated ghosting. But the
optimal seam may include individual error points and its overall strength value is the smallest. The line is taken as the
optimal seam incorrectly. So the ghosting may still exist. In order to avoid the impact of error points an improved
optimal seam based on feature point is used in this paper. The scale invariant feature transform and the random sample
consensus are used to ensure the detected feature points are accurate. The improved optimal seam do not regard the
original optimal seam regards the point with smallest guidelines value as extended direction this paper regards the
feature point as the extended direction. The feature point's strength value should have an appropriate weight value. It
makes the optimal seam contain more feature points. In this way it can avoid the error points and dynamic elements.
Both sides of the optimal seam have exposure differences, an image fusion method should be taken to achieve smooth
and natural mosaic. Poisson fusion method can achieve the synthesis of the fragment. So Poisson fusion is used to
eliminate exposure difference in this paper.
In this paper, a novel age estimation method by using active appearance model (AAM) combining with local texture
feature is presented, which overcomes the drawbacks of the AAM. Use the multi-scale local binary patterns (MLBP) as
the local texture descriptors to get the rotation invariant texture features. Build the combined AAM model using MLBP
features. In this way, both global face features and local texture features are used. The support vector regression (SVR) is
used to estimate the facial age. The face aging data set FG-NET is used. Experimental results demonstrate the AAM
combined MLBP method performing a lower mean-absolute error (MAE) and high accuracy of estimation comparing to
other method results.
Usually, we use a single template in traditional template matching. But single template is affected easily by geometric
distortion. The distortion of some pixels would affect the whole template. For solving the problem, in this paper we
divide the original template into five small templates. The small template has fewer pixels than the original template;
geometric distortion just affects the relevant small template. We match the other templates with the image by a new
matching method based on the improved maximum pixel count (MPC) criterion which considers not only the number of
similar points but the error value. Experimental results demonstrate that the method proposed in this paper has better
accuracy, precision, robust.
The tracking method based on Co-Training framework considers the object tracking as a semi-supervised learning
problem. This paper proposes a new on-line tracking method based on Co-Training framework. The method fuses two
features to describe the object and do randomizing affine deformation with positive examples to increase the number of
positive examples. Experimental results demonstrate, the on-line tracking method based on Co-training framework can
work robustly in long-term tracking and the drift of tracking can be effectively avoided.
A novel motion target detection method based on the Fourier descriptors of the fractal edge is presented in this paper,
The blanket covering fractal method is used to detect the edge features of targets in the high scale, and get binarization of
the features with the maximum entropy threshold segmentation, through labeling targets and removing small area noise
targets, then extract the elliptic Fourier descriptors of targets shape, cumulating the descriptors of multi-frame targets, the
frequency of Fourier descriptors can decide which one is the moving target and which is not.
We focus on the method of image registration and fusion and introduce all kinds of existing registration and fusion
algorithm based on lifting wavelet in detail in this paper. Based on the characteristics of the infrared and visible images,
this paper presents a registration approach using lifting wavelet transform to extract edge feature points, and improve the
fusion algorithm based on features of human vision system (HVS). The methods refer much other knowledge such as
lifting scheme, edge detection, affine transformation, HVS and fusion rules. A fast multi-resolution image fusion method
based on visual features for infrared and visible images is proposed in this paper. The source images are decomposed
using CDF9/7 lifting wavelet transform, respectively. Then it calculated the visual features of each sub-image and chose
the different rule of fusion based on visual features. Finally, a fused image is reconstructed by using inverse lifting
wavelet transform. Experimental results demonstrate that the proposed method has apparent advantage in information
preservation and resolution even if the source images have low signal to noise ratio (SNR), and the algorithm is more
effective in computational speed.
A new approach was proposed for detecting and segmenting the beacons from low contrast infrared image captured from
two CCD detectors fixed on two sides of deck, respectively. Adaptive thresholding is used for obtaining binary image of
the infrared beacons by rough segmentation in preprocessing step, which probably contains some non-beacons areas.
Then line encoding is conducted for binary image containing beacons and non-beacons, which is a look-up table (LUT)
with little data quantity. Refined beacons extraction and recognition based on various features are conducted within LUT
data. Finally, the geometric features (e.g. rectangularity, circularity, eccentricity) combining with the moment invariants
is used to recognize the real beacons by minimum euclidean distance and prior information of artificial beacon.
Experimental results show that the proposed method can preserve the perfect shape of beacons and effectively detect and
track it for opto-electronic guiding system.
Camera calibration is an important step for vision-based measurement applications. A well-known flexible camera calibration method is analyzed that uses the checkerboard pattern plane and in which the camera can be moved freely. When using a perspective projection camera model, characteristics of both the objective plane and the image plane are utilized and accurate results can be obtained. However, the method's results may fail when the rotation angles of the planar pattern are small, and the distortion coefficients obtained under the perspective projection model can not be used for a real-time vision application. We solve the ill-conditioned equations using the genetic algorithm, and the correct camera parameters are always obtained. We compute the distortion coefficients of the inverse projection model, which can be used for general vision applications. The influence of the corner detection precision is taken into consideration. Simulation shows that the best results may be obtained when the planar pattern is placed in a close range and its rotation angle is small. Simulations and real-world experiments illustrate that the improved calibration algorithm can always obtain robust and accurate results.
This paper presents a novel method for automatically segmenting and detecting targets in complex environment using the
improved unit linking pulse coupled neural networks (ULPCNN) combining with contour tracking. On the one hand, the
typical ULPCNN model is improved including linear modulate, linear attenuation of dynamic threshold and the
attenuation parameter matrix Δ , which is more suitable for segmenting and detecting the target under complex
environment. On the other hand, we determine the iteration times and obtain the optimal segmentation result using
contour tracking based on maximum line contour point. In order to verify the efficiency, various simulations were
conducted for different images acquired from real scenes. Experimental results show, as compared to the conventional
approaches, the proposed method can overcome the drawbacks of PCNN and obtain the good results for segmenting and
detecting targets against complex background.
In a real-time vision navigation system, an accurate and fast convergent pose estimation algorithm is required for the video guidance sensor. The orthogonal iteration (OI) algorithm is fast and globally convergent, but its results have a large translation error at a close range, and sometimes it fails to give a correct rotation matrix when the data are severely corrupted, when using the 3-D feature points. When the rotation matrix solution in the OI algorithm has been refined, an efficient pose estimation algorithm is derived. Simulation of the improved algorithm shows that the rotation matrix is always proper, which in turn improves the accuracy of the translation vector. The noise resistance and the outlier tolerance are enhanced by using the improved algorithm. The two algorithms are applied to our experimental system for an unmanned vehicle rendezvous and docking simulation separately. The comparison experiments show that the relative distance error is less than 0.28% from 1.5 to 5 m, and the rotation angle error is within ±0.7 deg in 5 m using the improved algorithm. These are better than the results using the OI algorithm.
A novel method for image segmentation using double-level parallelized firing pulse coupled neural networks
(DLPFPCNN) is presented in this paper. The first level (or auxiliary level) is used to enhance image by improved and
simplified PCNN model combining with boundary enhancement, which can give the better results for the second level
(or primary level) PCNN. The primary level uses a parallelized firing PCNN (PFPCNN) model to segment the enhanced
images so that can improve the adaptability to the complex environment. Parallelized firing neuron model can overcome
the drawbacks for sequential pulse burst, which is unfair for those pixels at low grayscale value areas. Finally, the
optimal segmentation results are determined by maximum Shannon entropy of image. Experimental results show, as
compared to the conventional PCNN model with single level and sequential pulse burst, the proposed method can
improve the performance of image segmentation and obtain the good results, especially suiting for those images with low
contrast, low signal-to-noise ratio (SNR) and continuously spatial-varying background.
A new detection algorithm of dim moving targets in the IR image sequences is presented. The images consist mainly of
sensor noise, drifting background clutter and low contrast targets. So it is difficult to provide reliable detection in just a
single frame. This algorithm adopts multi-scale gradient to suppress the clutter according to its spatial distribution
feature. Then the recursive maximum similarity (RMS) filter is used to accumulate targets energy in temporal domain.
The advantage of the algorithm is that it realizes the enhancement of the target gradient feature and clutter suppressing in
the same time. The results show that the algorithm can effectively detect the real small dim targets even if there is strong
A novel algorithm for detecting and tracking extended target using the projection curves analysis and correlation tracking based on the maximum matching pel count (MPC) criterion is presented. First, the projection curves of the difference image of two consecutive frames are analyzed to find the approximate areas of moving target on the entire scenes. Then correlation tracking based on the improved MPC criterion is used for target tracking against the cluttered background. Experimental results show, as compared to the conventional approaches, the proposed algorithm is more robust, has higher precision, and has simplified computational complexity for tracking an extended target against a cluttered background.
A novel algorithm for detecting and tracking extended target against cluttered background using the improved MPC
distance and searching strategy with hierarchy model is presented in this paper. Comparing with the conventional
methods, the proposed method modified the correlation distance for MPC criterion for classifying target and background
as well, and designed an image shrinking methods controlled by the ratio of median filter and mean filter, which can
improve the computation efficiency. Moreover, we interpolate the image within a small window to obtain high accuracy
with the half pixels. Experimental results show, as compared to the conventional approaches, the proposed algorithm is
more robust, higher precision and has simplified computational complexity for tracking extended target against cluttered
Real-time target detection against strong (bright) background under daytime is a challenging and leading edge subject, and also is a key technique for imaging tracking system. Strong background makes CCD image sensor work in critical saturation state, and imaging target contrast is very low. It's very difficult to accurately and stably track due to the complex characteristics of imaging target, such as strong clutter background, low contrast, and low signal to noise radio (SNR). So the key techniques for detecting and tracking target are eliminating the disturbance of diffuse reflection and beacon, synchronous detection, improving the performance of real-time image processing with high frame rate and high sampling rate.
A robust strategy for detecting and tracking day-time target was proposed in this paper. A series of efficient approaches ware presented to improve performance of detection and tracking in precision and stability, including strong background and noise suppression, image enhancement, adaptive thresholding, region merging based on morphology, recognition and tracking algorithm and so on.
In the end, we summarized and built the effictive flow for detecting and tracking target against strong background under daytime. The results of combining computer simulation with practical detection experiments show that the above-mentioned approaches are feasible and significant for real-time tracking system.
Dim target detection and tracking is a key technique for imaging tracking systems and is a challenging and leading-edge subject. A new strategy for dim target detection is presented. Various nonlinear features that are representative of dim targets are extracted from image sequences. Then, an information fusion algorithm by Karhunen-Loeve (K-L) transform is used to obtain an integrated feature. As a correlation degree, using an integrated feature calculates a target's position coordinates. Experimental results demonstrate the proposed method is feasible and effective.
Searching for the faint target is thought as a very difficult subject in imaging track system. The target shows the complex characteristics, for example low S/N ratio, low contrast, small size, blurry edge, even flicker, discontinuous image, changeful brightness during the target’s moving. It’s difficult to track the target with high-precision and wellstability. In this paper, a new algorithm searching for the target adaptively was proposed. This method fully uses the correlation of target’s multi-characteristics fusion as criterion and fitness function during searching. Then calculate its
position coordinate using global and adaptive genetic algorithm by searching for it. Experimental results demonstrate that this method improve not only detection precision, but also stability and intelligence strategy in tracking.