Visit My Account to manage your email alerts.
This PDF file contains the front matter associated with SPIE Proceedings Volume 9812, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.
Researches have proved that flying insects such as bees can achieve efficient and robust flight control, and biologists have explored some biomimetic principles regarding how they control flight. Based on those basic studies and principles acquired from the flying insects, this paper proposes a different solution of recovering ego-motion for low level navigation. Firstly, a new type of entropy flow is provided to calculate the motion parameters. Secondly, EKF, which has been used for navigation for some years to correct accumulated error, and estimation-Maximization, which is always used to estimate parameters, are put together to determine the ego-motion estimation of aerial vehicles. Numerical simulation on MATLAB has proved that this navigation system provides more accurate position and smaller mean absolute error than pure optical flow navigation. This paper has done pioneering work in bionic mechanism to space navigation.
In the process of steel billets recognition on the production line, the key problem is how to determine the position of the billet from complex scenes. To solve this problem, this paper presents a positioning algorithm based on the feature variance of billet character. Using the largest intra-cluster variance recursive method based on multilevel filtering, the billet characters are segmented completely from the complex scenes. There are three rows of characters on each steel billet, we are able to determine whether the connected regions, which satisfy the condition of the feature variance, are on a straight line. Then we can accurately locate the steel billet. The experimental results demonstrated that the proposed method in this paper is competitive to other methods in positioning the characters and it also reduce the running time. The algorithm can provide a better basis for the character recognition.
In order to resolve false segmentation and false tracking problems caused by the influence of complex harbor background during IR moving target detection, a harbor background suppression approach is presented. Firstly, Sky-sea line region can be obtained by Otsu segmentation, which is applied to split images obtained through wavelet transform. Secondly, harbor background suppression point in sequential images can be located by multilevel filter. Finally, harbor background suppression can be realized according to those background suppression points. The proposed approach is validated by using actual IR in complex harbor background to realize background suppression. Experiment results indicate the feasibility and effectiveness of the proposed method.
The process of Reference image preparation, is to extract region of interest area from a image,to get a simplified image which be used as template image for other algorithms.Because of the complex of scene,one usually need to excute a set of different algorithms to get a final image which only has the information of interest area. This paper presents a new variational model which used L0 norm for idelity item, to keep information of interested area better, while removing other redundant information. Experiments show that,this method can remove information of grediant in some special range,to handle a more general case,compared with the original L0 gradient method which can only remove low frequency information. Compared with the same variational model but using L1orL2 norm，the proposed method can well retain the original information.Those advantages is very important for making the process of reference image preparation faster and easier
This article seeks to discover the object categories’ semantic probabilistic model (OSPM) based on statistical test analysis method. We applied this model on road forward environment perception algorithm, including on-road object recognition and detection. First, the image was represented by a set composed of words (local feature regions). Then, found the probability distribution among image, local regions and object semantic category based on the new model. In training, the parameters of the object model are estimated. This is done by using expectation-maximization in a maximum likelihood setting. In recognition, this model is used to classify images by using a Bayesian manner. In detection, the posterios is calculated to detect the typical on-road objects. Experiments release the good performance on object recognition and detection in urban street background.
Face image symmetry is an important factor affecting the accuracy of automatic face recognition. Selecting high symmetrical face image could improve the performance of the recognition. In this paper, we proposed a novel facial symmetry evaluation scheme based on geometric features, including centroid, singular value, in-plane rotation angle of face and the structural similarity index (SSIM). First, we calculate the value of the four features according to the corresponding formula. Then, we use fuzzy logic algorithm to integrate the value of the four features into a single number which represents the facial symmetry. The proposed method is efficient and can adapt to different recognition methods. Experimental results demonstrate its effectiveness in improving the robustness of face detection and recognition.
Camshift algorithm and three frame difference algorithm are the popular target recognition and tracking methods. Camshift algorithm requires a manual initialization of the search window, which needs the subjective error and coherence, and only in the initialization calculating a color histogram, so the color probability model cannot be updated continuously. On the other hand, three frame difference method does not require manual initialization search window, it can make full use of the motion information of the target only to determine the range of motion. But it is unable to determine the contours of the object, and can not make use of the color information of the target object. Therefore, the improved Camshift algorithm is proposed to overcome the disadvantages of the original algorithm, the three frame difference operation is combined with the object's motion information and color information to identify the target object. The improved Camshift algorithm is realized and shows better performance in the recognition and tracking of the target.
In order to enhance the speed and accuracy of ellipse detection, an ellipse detection algorithm based on edge classification is proposed. Too many edge points are removed by making edge into point in serialized form and the distance constraint between the edge points. It achieves effective classification by the criteria of the angle between the edge points. And it makes the probability of randomly selecting the edge points falling on the same ellipse greatly increased. Ellipse fitting accuracy is significantly improved by the optimization of the RED algorithm. It uses Euclidean distance to measure the distance from the edge point to the elliptical boundary. Experimental results show that: it can detect ellipse well in case of edge with interference or edges blocking each other. It has higher detecting precision and less time consuming than the RED algorithm.
Small target detection plays a crucial role in infrared warning and tracking systems. A background suppression method using morphological filter based on quantum genetic algorithm (QGMF) is presented to detect small targets in infrared image. Structure element of morphological filter is encoded and the best structure element is selected using quantum genetic algorithm. The optimized structure element is used for background suppression to detect small target. Experimental results demonstrate that QGMF has good performance in clutter suppression, and obtains higher signal-to-clutter ratio gain (SCRG) and background suppression factor (BSF) than the one using the fixed structure element with the same size.
The further research of visual processing mechanism provides a new idea for contour detection. On the primary visual cortex, the non-classical receptive field of the neurons has the orientation selectivity exerts suppression effect on the response of classical receptive field, which influences edge or line perception. Based on the suppression property of non-classical receptive field and contour completion, this paper proposed a contour detection method based on brightness and contour completion. The experiment shows that the proposed method can not only effectively eliminate clutter information, but also connect the broken contour points by taking advantage of contour completion.
With the advent of SAR imaging technology, application of SAR image navigation is rapidly developed in civil life or defense fields. As target recognition plays a big role in SAR image navigation, how to detect those geography ground objects such as river, road and airport automatically and quickly is more and more important. In the last few years, much research has focused on the detection of river, road and airport from SAR images. However, those objects are commonly detected by individual. This paper presents a method capable of detecting linear object ROI such as river, road and airport. All the linear objects ROI in the scene could be extracted quickly at the same time.
An automatic target detection method used in long term infrared (IR) image sequence from a moving platform is proposed. Firstly, based on non-linear histogram equalization, target candidates are coarse-to-fine segmented by using two self-adapt thresholds generated in the intensity space. Then the real target is captured via two different selection approaches. At the beginning of image sequence, the genuine target with litter texture is discriminated from other candidates by using contrast-based confidence measure. On the other hand, when the target becomes larger, we apply online EM method to iteratively estimate and update the distributions of target's size and position based on the prior detection results, and then recognize the genuine one which satisfies both the constraints of size and position. Experimental results demonstrate that the presented method is accurate, robust and efficient.
LiveWire interactive boundary extraction algorithm based on Haar wavelet transform and control point set direction search
Based on deep analysis of the LiveWire interactive boundary extraction algorithm, a new algorithm focusing on improving the speed of LiveWire algorithm is proposed in this paper. Firstly, the Haar wavelet transform is carried on the input image, and the boundary is extracted on the low resolution image obtained by the wavelet transform of the input image. Secondly, calculating LiveWire shortest path is based on the control point set direction search by utilizing the spatial relationship between the two control points users provide in real time. Thirdly, the search order of the adjacent points of the starting node is set in advance. An ordinary queue instead of a priority queue is taken as the storage pool of the points when optimizing their shortest path value, thus reducing the complexity of the algorithm from O[n2] to O[n]. Finally, A region iterative backward projection method based on neighborhood pixel polling has been used to convert dual-pixel boundary of the reconstructed image to single-pixel boundary after Haar wavelet inverse transform. The algorithm proposed in this paper combines the advantage of the Haar wavelet transform and the advantage of the optimal path searching method based on control point set direction search. The former has fast speed of image decomposition and reconstruction and is more consistent with the texture features of the image and the latter can reduce the time complexity of the original algorithm. So that the algorithm can improve the speed in interactive boundary extraction as well as reflect the boundary information of the image more comprehensively. All methods mentioned above have a big role in improving the execution efficiency and the robustness of the algorithm.
Research on matching area selection based on multiple features fusion of full tensor gravity gradient
Submarines in the underwater sailing need a safe, reliable, high accurate, and covert well navigation system. Inertial navigation system (INS) is the core of underwater navigation. But the inertial navigation system gathers information based gyroscope, accelerometer and other sensors. In accordance to Newton's laws of mechanics, their own speed, location and other information is calculated by integral recursion. Since the recursive work of INS, positioning error gradually increases with time elapsing. Gravity and gravity gradient aided navigation as a passive autonomous navigation are more and more focused on, Selection of the gravity gradient matching area is one of the key to gravity gradient matching navigation. Earth's marine area is enormous, underwater environment is complex. Take advantage of multi-feature information fusion of gravity gradient full tensor, one hand a wider range of matching area can be got, to gain wider path planning area. on the other hand, the positioning accuracy of assisted navigation system can be inproved.
It is an important measure to observe target by using laser sensor in the field of target detection. Exact and reliable dynamic laser scatter characteristics of observing target, can not only be used for the design and development of laser sensor as well as the research of algorithm for target capture, recognition and tracking, but also can offer reference bases for the test flow programming. A set of simulation, measurement system for the dynamic laser scatter characteristics of observing target is introduced in this paper. The simulation problem of dynamic laser scatter characteristics of observing target is solved, under the circumstance of laboratory with different azimuth angle and pitch angle of observation. The dynamic laser scatter characteristics of observing target can be obtained directly by such system, the test data can be used for the verification of the analyzing model for the laser scatter characteristics of observing target, and can also offer basis for the development of laser detecting sensor.
Automatic targets recognition(ATR) of artificial objects in high resolution remote sensing images can be divided into two categories by the properties of targets. The first such building, a harbor which has fixed location and stable out looking. The other one, for example aircraft, whose location and posture is sensitive to the moment. Due to the variable sizes, colors, orientations, and complex background, aircraft detection is a difficult task in high resolution remote sensing images. In this paper, a simple and effective aircraft detection method with a single template is proposed, which exactly locates the object by outputting its geometric center, location and orientation. Compare to traditional method,this method only needs critical feature in the local areas of target and a binary template of aircraft. Compare to traditional Feature + Classifier method, it’s easy, simple and don’t need outline training,but also get high precision and low false rate in the same complicate background.
Navigation without GPS and other knowledge of environment have been studied for many decades. Advance technology have made sensors more compact and subtle that can be easily integrated into micro and hand-hold device. Recently researchers found that bee and fruit fly have an effectively and efficiently navigation mechanism through optical flow information and process only with their miniature brain. We present a navigation system inspired by the study of insects through a calibrated camera and other inertial sensors. The system utilizes SLAM theory and can be worked in many GPS denied environment. Simulation and experimental results are presented for validation and quantification.
Building losses assessment for Lushan earthquake utilization multisource remote sensing data and GIS
On 20 April 2013, a catastrophic earthquake of magnitude 7.0 struck the Lushan County, northwestern Sichuan Province, China. This earthquake named Lushan earthquake in China. The Lushan earthquake damaged many buildings. The situation of building loss is one basis for emergency relief and reconstruction. Thus, the building losses of the Lushan earthquake must be assessed. Remote sensing data and geographic information systems (GIS) can be employed to assess the building loss of the Lushan earthquake. The building losses assessment results for Lushan earthquake disaster utilization multisource remote sensing dada and GIS were reported in this paper. The assessment results indicated that 3.2% of buildings in the affected areas were complete collapsed. 12% and 12.5% of buildings were heavy damaged and slight damaged, respectively. The complete collapsed buildings, heavy damaged buildings, and slight damaged buildings mainly located at Danling County, Hongya County, Lushan County, Mingshan County, Qionglai County, Tianquan County, and Yingjing County.
Real-time detection of generic objects using objectness estimation and locally adaptive regression kernels matching
Our purpose is to develop a detection algorithm capable of searching for generic interest objects in real time without large training sets and long-time training stages. Instead of the classical sliding window object detection paradigm, we employ an objectness measure to produce a small set of candidate windows efficiently using Binarized Normed Gradients and a Laplacian of Gaussian-like filter. We then extract Locally Adaptive Regression Kernels (LARKs) as descriptors both from a model image and the candidate windows which measure the likeness of a pixel to its surroundings. Using a matrix cosine similarity measure, the algorithm yields a scalar resemblance map, indicating the likelihood of similarity between the model and the candidate windows. By employing nonparametric significance tests and non-maxima suppression, we detect the presence of objects similar to the given model. Experiments show that the proposed detection paradigm can automatically detect the presence, the number, as well as location of similar objects to the given model. The high quality and efficiency of our method make it suitable for real time multi-category object detection applications.
Pavement crack detection is affected by much interference in the realistic situation, such as the shadow, road sign, oil stain, salt and pepper noise etc. Due to these unfavorable factors, the exist crack detection methods are difficult to distinguish the crack from background correctly. How to extract crack information effectively is the key problem to the road crack detection system. To solve this problem, a novel method for pavement crack detection based on combining non-negative feature with fast LoG is proposed. The two key novelties and benefits of this new approach are that 1) using image pixel gray value compensation to acquisit uniform image, and 2) combining non-negative feature with fast LoG to extract crack information. The image preprocessing results demonstrate that the method is indeed able to homogenize the crack image with more accurately compared to existing methods. A large number of experimental results demonstrate the proposed approach can detect the crack regions more correctly compared with traditional methods.
The scene matching based navigation is an important precision navigation technology for unmanned aerial vehicles (UAV). Selection of interest area where reference image is made has an important influence on the precision of matching result besides the performance of match algorithm. In this paper, a method to select interest area based on structured edge detection is proposed. We use a data driven approach that classifies each pixel with a typical structured edge label. We propose a method that combines these labels into a feature measuring suitable to match of a region. Then a SVM classifier is trained to classify the features and get the final result of the selection of interest area. The experimental result shows that the proposed method is valid and effective.
In this paper, we proposed a novel method to extract shape feature based on dual-tree complex wavelet. First, with the two level dual-tree complex wavelet transformations, we can get two low frequency components of the first level, which are used as wavelet moment invariants formed from approximation coefficients. Then, we calculate means and variance for each of the six detailed components in the second level since it contains different directions information of the shape. Using the Principal Component Analysis (PCA), twenty features can be reduced to five maximum useful features which contribute to shape matching.
The reconstruction of the 3D scene in the monocular stereo vision needs to get the depth of the field scenic points in the picture scene. But there will inevitably be error matching in the process of image matching, especially when there are a large number of repeat texture areas in the images, there will be lots of error matches. At present, multiple baseline stereo imaging algorithm is commonly used to eliminate matching error for repeated texture areas. This algorithm can eliminate the ambiguity correspond to common repetition texture. But this algorithm has restrictions on the baseline, and has low speed. In this paper, we put forward an algorithm of calculating the depth of the matching points in the repeat texture areas based on the clustering algorithm. Firstly, we adopt Gauss Filter to preprocess the images. Secondly, we segment the repeated texture regions in the images into image blocks by using spectral clustering segmentation algorithm based on super pixel and tag the image blocks. Then, match the two images and solve the depth of the image. Finally, the depth of the image blocks takes the median in all depth values of calculating point in the bock. So the depth of repeated texture areas is got. The results of a lot of image experiments show that the effect of our algorithm for calculating the depth of repeated texture areas is very good.
Aircraft recognition is of great theoretical and practical significance in fields like remote sensing, navigation and traffic monitoring. It seems difficult to recognize aircraft in low-resolution SAR imagery because of difference between real image and simulated template induced by poor image quality and inherent simulation error. Aiming at this problem, an aircraft recognition method using peak feature matching is proposed. Firstly, the scattering centers of detected target are extracted in low-resolution SAR imagery using an adaptive threshold. Secondly, the extracted peak features are used to estimate the aircraft azimuth angle, which can be used to reduce the searching space in template database dramatically. Finally, a novel peak feature matching method using spatial distribution information of entire peak set is proposed to measure the similarity between detected target and simulated template. Experimental results demonstrate the good performance of the proposed method on a variety of low-resolution SAR imageries.
The radar targets number increases from one to more when the ballistic missile is in the process of separating the lower stage rocket or casting covers or other components. It is vital to identify the warhead target quickly among these multiple targets for radar tracking. A fast recognition method of the warhead target is proposed to solve this problem by using kinematic features, utilizing fuzzy comprehensive method and information fusion method. In order to weaken the influence of radar measurement noise, an extended Kalman filter with constant jerk model (CJEKF) is applied to obtain more accurate target’s motion information. The simulation shows the validity of the algorithm and the effects of the radar measurement precision upon the algorithm’s performance.
In this paper, a new target reconstruction method considering the atmospheric refraction is presented to improve 3D reconstruction accuracy in long rang surveillance system. The basic idea of the method is that the atmosphere between the camera and the target is partitioned into several thin layers radially in which the density is regarded as uniform; Then the reverse tracking of the light propagation path from sensor to target was carried by applying Snell’s law at the interface between layers; and finally the average of the tracked target’s positions from different cameras is regarded as the reconstructed position. The reconstruction experiments were carried, and the experiment results showed that the new method have much better reconstruction accuracy than the traditional stereoscopic reconstruction method.
A real-time target detection algorithm based on combination of intensity and edge for infrared search system
With regard to aircraft target detection in complex clouds background of infrared search system, this paper proposes a new target detection algorithm based on combination of intensity and edge of the target. Firstly, the algorithm segments the image by iterative OTSU segmentation method, at the same time, it detects the edge by morphological processing. Then, by the fusion decision of the combination of segmentation and edge result, it detects the real aircraft targets and eliminates the clouds false alarm. The algorithm overcomes the too much clouds false alarm problems of the traditional target detection method. The test data detection shows, the algorithm enables effective detection of aircraft target in complex clouds background with low-rate false warning. The algorithm has realized real-time processing and has been effectively applied to the development of the engineering sample of the Wide Field of View Infrared Search System.
In this paper, a line detection algorithm, which is based on direction filter and regional growth, is proposed for line detection problem in a specific angle. Our algorithm is better than the state-of-the-art algorithm in the detection of the specific direction of the line, which have a higher detection rate, less error and missing detection, lower computational complexity, higher efficiency of the algorithm and better anti-interference ability.
In this paper, a fast correlation matching method in frequency domain was presented to extract the vehicle speed from a single QuickBird (QB) satellite image. Suppose the vehicles had been extracted from 0.6m resolution panchromatic image, we transformed the panchromatic and multispectral images into frequency domain and found the maximum correlation position to determine the exact coordinates of vehicles. Then the speed of moving vehicles were calculated using the shift between the two coordinates and the time gap between the two images. The novelty of this work is that the time consumption can be significantly reduced compared with the conventional area correlation matching method. This method makes it feasible to use satellite images for traffic statistics in large scale of road network.
In this paper, we present a new direction of arrival (DOA) estimation algorithm for coherent wideband signals. This algorithm is based on the test of orthogonality of projected subspaces (TOPS) method which will fail to work in real environments where signals are highly correlated or coherent due to multipath propagation. In order to overcome the disadvantage, we combine spatial smoothing techniques with TOPS method so that the rank of covariance matrix is equal to the number of signal sources even signals received are coherent. Unlike other coherent wideband methods, such as the coherent signal subspace method (CSSM) and WAVES, the new method does not require any initial DOA estimation, thus avoiding errors brought by incorrect initial values. Simulations on computer and experiments in the anechoic chamber based on an 8-elements digital array radar test-bed operating at L & S band are carried out. Simulation and experimental results validate the effectiveness of proposed algorithm.
A new target extraction algorithm based Nonsubsampled Contourlet Transform(NSCT) is proposed according to the difficulty of weak target extraction. Paper detailed analyses and summarizes the data feature of image in NSCT domain, proposes a weak target extraction algorithm using high-frequency coefficients and mathematical morphology. The high-frequency coefficients were calculated by NSCT at first. Then the high-frequency coefficients were processed for noise cancellation by adaptive thresholds in corresponding sub-band. After these operations, the mathematical morphology method was adopted to remedy the defects of image contour. Finally, the object image is obtained by inverse NSCT. The simulation results show that this method can detect the target information fast and accurately, can meet the practical requirement.
Radar Micro-Doppler signatures are of great potential for target detection, classification and recognition. In the mid-course phase, warheads flying outside the atmosphere are usually accompanied by precession. Precession may induce additional frequency modulations on the returned radar signal, which can be regarded as a unique signature and provide additional information that is complementary to existing target recognition methods. The main purpose of this paper is to establish a more actual precession model of conical ballistic missile warhead and extract the precession parameters by utilizing Viterbi & Kalman algorithm, which improving the precession frequency estimation accuracy evidently , especially in low SNR.
Symmetry axis extraction is an important part of the image feature detection. So far, various classical symmetry axes extraction algorithms have been proposed, such as the minimum-inertia-axis-based method, the SIFT-based method. If the input image is blurry, or it’s difficult to extract feature points or corner points from input images, however, the above algorithms are difficult to obtain satisfied results. This paper presents a gradient-based method that can robustly extract symmetry axis from visual pattern. The key points of our methods are gradient calculation, symmetric weight calculation, and Hough Transform. Our method was evaluated on several datasets, including both blurred and smooth-edged cases. Experimental results demonstrated that our method achieves a more robust performance than previous methods.
The image classification is an important means of image segmentation and data mining, how to achieve rapid automated image classification has been the focus of research. In this paper, based on the super pixel density of cluster centers algorithm for automatic image classification and identify outlier. The use of the image pixel location coordinates and gray value computing density and distance, to achieve automatic image classification and outlier extraction. Due to the increased pixel dramatically increase the computational complexity, consider the method of ultra-pixel image preprocessing, divided into a small number of super-pixel sub-blocks after the density and distance calculations, while the design of a normalized density and distance discrimination law, to achieve automatic classification and clustering center selection, whereby the image automatically classify and identify outlier. After a lot of experiments, our method does not require human intervention, can automatically categorize images computing speed than the density clustering algorithm, the image can be effectively automated classification and outlier extraction.
Touching Mycobacterium tuberculosis objects in the Ziehl-Neelsen stained sputum smear images present different shapes and invisible boundaries in the adhesion areas, which increases the difficulty in objects recognition and counting. In this paper, we present a segmentation method of combining the hierarchy tree analysis with gradient vector flow snake to address this problem. The skeletons of the objects are used for structure analysis based on the hierarchy tree. The gradient vector flow snake is used to estimate the object edge. Experimental results show that the single objects composing the touching objects are successfully segmented by the proposed method. This work will improve the accuracy and practicability of the computer-aided diagnosis of tuberculosis.
Framland parcels extraction from high-resolution remote sensing images based on the two-stage image classification
It is difficult and boring for people to artificially extract farmland parcels from high resolution remote sensing images. Therefore, automatic methods are in the urgent need to release image interpreters from such a work as well as achieve accurate results. In the past years, although many researchers have made attempts to solve this problem by using different techniques and also produced some good results, they still cannot meet the demand of practical applications. In this paper, a farmland extraction method is proposed based on a new technique of two-stage image classification. The first stage aims at producing a map of farmland area by using the supervised iterative conditional mode (ICM), where a novel mixture posterior is proposed based on the tree-structured interpretation of certain complex landscapes, e.g., farmland and building area, and the Markov random field model (MRF) is also used to make use of spatial information between neighboring pixels. The second stage extracts the farmland parcels by using the Meanshift algorithm (MS) based on the hybrid of the original image and the texture image produced by the local binary pattern (LBP) method. We applied our method to a piece of aerial image in the urban area of Taizhou, China. The results show that the proposed method has an ability to produce more accurate results than the MS method.
The 3D target tracking based on the infrared image is a challenge problem in computer vision. With the development and progress of the target tracking in recent years, key point-based target tracking is widely applied in many fields, such as augmented reality, object retrieval, human computer interaction and medical imaging. Generally, the aim of target tracking is to locate the targets in the sub-sequent frames. Because the tracking is a fundamental research topic, it should be self-adaptive under some special situations,, such as illumination variation, occlusion, background clutters and so on. Thereby it is hard to fit all scenarios by single tracking approach. And in a particle filter framework, computational cost grows linearly with the number of particles sampled. Hence, it is important to decrease computational cost but also get good performance. In the process of the tracking 3D target, traditional tracking algorithms, such as template matching using fast normalized cross correlation or HOG (Histogram of Oriented Gradients), perform not so well. The main reasons are that they are not scale and rotation invariant matching. Mean shift tracking algorithm takes advantage of a probability density to search for the position of the target, which fails in abrupt changes of luminance. Furthermore, the feature selection is important in obtaining high accuracy. So the computational cost may become a bottleneck in tracking problem. This paper proposes an effective tracking framework based on FAST BRIEF and Particle Filter (FBPF). Our approach can be divided into four steps as follows:(1) An advanced fast algorithm which includes decision trees is adopted to extract key points. (2) The BRIEF descriptor is applied to speed up the feature matching.(3)Calculate the affine transformation parameters. (4)Particle filter is adopted to predict the best position of target. The key point of traditional FAST algorithm is defined as point which has different gray value with enough surrounding constant pixels. Applied in the gray images, it means there are enough and constant pixels with more or less gray value than itself. The circle area is usually selected when taking any pixel in the image and its round area. In order to decrease the cost of calculation, we employ an advanced FAST based on decision tree, which learns the key points with the maximum entropy. Based on the binary descriptor BRIEF, the original transformation matrix can be calculated efficiently. Then, in order to estimate the best position of the target, particle filter is applied. We take the parameters as particles. It provides a convenient framework for estimating and propagating the posterior probability density function of state variables regardless of the underlying distribution through a sequence of prediction and update steps. Experimental results demonstrate that FBPF achieves state-of-the-art performance on the challenge sequences. In this paper, the main contribution is the proposal of a new tracking framework which based on FAST BRIEF and Particle Filter (FBPF). First, we employ an advanced FAST using the decision tree and BRIEF is used to establish feature for decreasing time consumption. Second, we calculate the original affine matrix base on the feature matching. Finally, the particle filter is utilized to predict the best position.
In this paper, we propose a new pulse edge detection method based on short time Fourier transform (STFT) and difference of boxes (DOB) filter. Firstly, detect the coarse starting and ending positions in frequency domain after STFT. Then obtain the precise pulse edge through DOB filter. It achieves a better performance than the classical energy detection (ED) method, especially when signal to noise ratio (SNR) is low. Simulation results and real data application validate the effectiveness of the proposed method.
Unsupervised multi-class co-segmentation via joint object detection and segmentation with energy minimization
Multi-class co-segmentation is a challenging task because of the variety and complexity of the objects and images. To get more accurate object proposals is the key step for the existing co-segmentation methods to obtain better performance. In this paper, we propose a novel method to co-segment multiple regions from a group of images in an unsupervised way. The key idea is to discover unknown object proposals for each image via joint object detection and object-level segmentation. First, object proposals of each image are generated by object-like windows (or boxes) and object-level segmentation using graph cuts, and two Gaussian mixture models (GMMs) are employed to characterize the object proposals for all images and single image, respectively. Then, a weighted graph for each image is constructed on super-pixel level, and multi-label graph cuts with global and local energy is employed to obtain the final co-segmentation results. In contrast to previous methods, our method could obtain the object proposals with high objectness by object-level segmentation. Experimental results demonstrate the good performance of the proposed method on the multi-class co-segmentation.
Target cluster brings about a light-spot which consists of several neighborhood pixels in image, therefore it is difficult to distinguish between the targets or locate them with sub-pixel accuracy. In this paper, a pseudo oversampling-based C3PC (Covariance Constrained Constructive Particle Clustering) method is proposed to solve the closely space objects problem. As a classical detection and location method, C3PC algorithm, presents a particle clustering decomposition technique. However, the particle distribution according to the pixel gray value yields pixel level accuracy, which will lead to location error. Thus, by using a particle distribution at sub-pixel level, substantially better position accuracy can be obtained. According the characteristic of oversampling, an improved interpolation algorithm which simulating the oversampling techniques of sensor is brought forward. Simulation experiment results show that the positioning accuracy of CSOs in our algorithm is higher than that of C3PC algorithm.
Aiming at the graph optimization-based monocular SLAM, a novel design for the front-end in single camera SLAM is proposed, based on the recursive SOM. Pixel intensities are directly used to achieve image registration and motion estimation, which can save time compared with the current appearance-based frameworks, usually including feature extraction and matching. Once a key-frame is identified, a recursive SOM is used to actualize loop-closure detecting, resulting a more precise location. The experiment on a public dataset validates our method on a computer with a quicker and effective result.
In this paper, a method of spindle extraction of target in inverse synthetic aperture radar (ISAR) image is proposed which depends on Radon Transform. Firstly, utilizing Radon Transform to detect all straight lines which are collinear with these line segments in image. Then, using Sobel operator to detect image contour. Finally, finding all intersections of each straight line and image contour, the two intersections which have maximum distance between them is the two ends of this line segment and the longest line segment of all line segments is spindle of target. According to the proposed spindle extraction method, one hundred simulated ISAR images which are respectively rotated 0 degrees, 10 degrees, 20 degrees, 30 degrees and 40 degrees in counterclockwise are used to do experiment and the proposed method and the detection results are more close to the real spindle of target than the method based on Hough Transform .
A method of automatic threshold selection for image segmentation is presented. An optimal threshold is selected in order to preserve edge of image perfectly in image segmentation. The shortcoming of Otsu's method based on gray-level histograms is analyzed. The edge energy function of bivariate continuous function is expressed as the line integral while the edge energy function of image is simulated by discretizing the integral. An optimal threshold method by maximizing the edge energy function is given. Several experimental results are also presented to compare with the Otsu's method.
The micro-motion of ballistic missile targets induces micro-Doppler modulation on the radar return signal, which is a unique feature for the warhead discrimination during flight. In order to extract the micro-Doppler feature of ballistic missile targets, time-frequency analysis is employed to process the micro-Doppler modulated time-varying radar signal. The images of time-frequency distribution (TFD) reveal the micro-Doppler modulation characteristic very well. However, there are many existing time-frequency analysis methods to generate the time-frequency distribution images, including the short-time Fourier transform (STFT), Wigner distribution (WD) and Cohen class distribution, etc. Under the background of ballistic missile defence, the paper aims at working out an effective time-frequency analysis method for ballistic missile warhead discrimination from the decoys.
Infrared small target detection is difficult due to several aspects, including the low signal-to-clutter ratio of the infrared image, and the small size, lack of shape and texture information of the target. a novel method, which is based on spatial-temporal association, is presented for infrared target detection. The algorithm consists of the three steps: Firstly, 2-dimensional histogram of entropy flow field is computed to estimate the background motion. Secondly, the difference image through background motion compensation is obtained. Finally, the targets are detected by spatial-temporal filter. The experiment results demonstrate that the proposed algorithm is robust to noise, and also fit to detect small targets under moving backgrounds in infrared image sequences.
A novel monocular visual navigation method for cotton-picking robot based on horizontal spline segmentation
Visual navigation is a fundamental technique of intelligent cotton-picking robot. There are many components and cover in the cotton field, which make difficulties of furrow recognition and trajectory extraction. In this paper, a new field navigation path extraction method is presented. Firstly, the color image in RGB color space is pre-processed by the OTSU threshold algorithm and noise filtering. Secondly, the binary image is divided into numerous horizontally spline areas. In each area connected regions of neighboring images’ vertical center line are calculated by the Two-Pass algorithm. The center points of the connected regions are candidate points for navigation path. Thirdly, a series of navigation points are determined iteratively on the principle of the nearest distance between two candidate points in neighboring splines. Finally, the navigation path equation is fitted by the navigation points using the least squares method. Experiments prove that this method is accurate and effective. It is suitable for visual navigation in the complex environment of cotton field in different phases.
Visible image, compared with SAR image and infrared image, has the advantage of high resolution, clear details, etc. So it can be selected for object extraction. Water objects play an important role in locating bridges, dams and other typical buildings. This paper presents a segmentation method for visible image based on gradient of the original image, and combined with the features of the water targets. According to the feature of water targets, gray uniform, smaller entropy, and smaller local variance, water objects can be extracted automatically and effectively by using clustering method from image segmentation result.
The traditional Canny operator cannot get the optimal threshold in different scene, on this foundation, an improved Canny edge detection algorithm based on adaptive threshold is proposed. The result of the experiment pictures indicate that the improved algorithm can get responsible threshold, and has the better accuracy and precision in the edge detection.
Infrared image generation technology is being widely used in infrared imaging system performance evaluation, battlefield environment simulation and military personnel training, which require a more physically accurate and efficient method for infrared scene simulation. A parallel multiple path tracing method based on OptiX was proposed to solve the problem, which can not only increase computational efficiency compared to serial ray tracing using CPU, but also produce relatively accurate results. First, the flaws of current ray tracing methods in infrared simulation were analyzed and thus a multiple path tracing method based on OptiX was developed. Furthermore, the Monte Carlo integration was employed to solve the radiation transfer equation, in which the importance sampling method was applied to accelerate the integral convergent rate. After that, the framework of the simulation platform and its sensor effects simulation diagram were given. Finally, the results showed that the method could generate relatively accurate radiation images if a precise importance sampling method was available.
In this paper, a fast object tracking algorithm based on template matching and region information fusion extraction is proposed. In the prediction framework, the data connection task is achieved by object template and object information extraction. And then the object is tracked accurately by using the object motion information. We handle the tracking shift by using the confidence estimation strategy. The experiments show that the proposed algorithm has robust performance.