For 3D object recognition, a discriminative point cloud descriptor is required to represent the object. The existing global descriptors encode the whole object into a vector but they are sensitive to occlusion. On the contrary, the local descriptors encode only a small neighbor of a key point and are more robust to occlusion, but many objects have the same local surface. This paper presents a novel mixture method which segments a point cloud into multiple subparts to overcome the above shortcomings. In offline training stage, we propose to build up the model library that integrates both global and local surface of partial point clouds. In online recognition stage, the scene objects are represented by its subparts, and a voting scheme is performed for the recognition of scene objects. Experimental results on public datasets show that the proposed method promotes the recognition performance significantly compared to the conventional global and local descriptors.
Video anomaly event detection is the process of finding an abnormal event deviation compared with the majority of normal or usual events. The main challenges are the high structure redundancy and the dynamic changes in the scenes that are in surveillance videos. To address these problems, we present a framework for anomaly detection and localization in videos that is based on locality-constrained affine subspace coding (LASC) and a model updating procedure. In our algorithm, LASC attempts to reconstruct the test sample by its top-k nearest subspaces, which are obtained by segmenting the normal samples space using a clustering method. A sample with a large reconstruction cost is detected as abnormal by setting a threshold. To adapt to the scene changes over time, a model updating strategy is proposed. We experiment on two public datasets: the UCSD dataset and the Avenue dataset. The results demonstrate that our method achieves competitive performance at a 700 fps on a single desktop PC.
Correlation filter-based tracking has exhibited impressive robustness and accuracy in recent years. Standard correlation filter-based trackers are restricted to translation estimation and equipped with fixed target response. These trackers produce an inferior performance when encountered with a significant scale variation or appearance change. We propose a log-polar mapping-based scale space tracker with an adaptive target response. This tracker transforms the scale variation of the target in the Cartesian space into a shift along the logarithmic axis in the log-polar space. A one-dimensional scale correlation filter is learned online to estimate the shift along the logarithmic axis. With the log-polar representation, scale estimation is achieved accurately without a multiresolution pyramid. To achieve an adaptive target response, a variance of the Gaussian function is computed from the response map and updated online with a learning rate parameter. Our log-polar mapping-based scale correlation filter and adaptive target response can be combined with any correlation filter-based trackers. In addition, the scale correlation filter can be extended to a two-dimensional correlation filter to achieve joint estimation of the scale variation and in-plane rotation. Experiments performed on an OTB50 benchmark demonstrate that our tracker achieves superior performance against state-of-the-art trackers.
This paper proposes an absolute attitude measurement approach by utilizing a monostatic wideband radar. In this approach, the three-dimensional electromagnetic-model (3-D em-model) and the parametric motion model of a target are combined to estimate absolute attitude. The 3-D em-model is established offline based on the target’s geometric structure. Scattering characteristics such as radar cross section and radar images from one-dimension to 3-D can be conveniently predicted by this model. By matching the high-resolution range profiles (HRRPs) of measurements with the HRRPs predicted by the 3-D em-model, the directions of the lines of sight relative to the target at different measuring times are first obtained. Then, based on the obtained directions and the parametric motion model of the target, the target absolute attitude at each measuring time can be acquired. Experiments using both data predicted by a high-frequency em-code and data measured in an anechoic chamber verify the validity of the proposed method.
Electromagnetic model (em-model) provides a concise and physically relevant description of target through representative scatterers. In a forward built em-model, detailed information about each scatterer’s position, scattering amplitude along with its provenance can be predicted. This makes em-model a good candidate for use in synthetic aperture radar (SAR) automatic target recognition (ATR). In this paper, we introduce scatterers’ provenance as attributed information into target recognition, and an attributed em-model based target recognition method is proposed. Firstly, according to the purpose of ATR, each scatterer in em-model is endowed with an importance factor based on its provenance. Secondly, a detection is implemented to decide whether the em-model predicted scatterer has a corresponding scatterer in measured data. If the scatterer exist in measured target, evaluate how similar the scatterer pair resembled with each other. Next, similarities of all the scatterer pairs are synthesized as a whole match score between em-model and SAR data. In the synthesis, the importance factor servers as a weighting factor that scatterer with more attention will be more discriminative for recognition. In the end, target in measured SAR data is recognized as the model type or not based on the match score. The novelty of this method comes from taking into account of the provenance information of scatterers as attributed information and endowing the scatterers with different important factors according to their importance in recognition. This makes the attributed scatterer based recognition method pertinent to the purpose of ATR. Experiments on simulated Tank SAR data that produced by a high frequency electromagnetic simulation software verified the effectiveness of this method.
Feature extraction and matching are two important steps in synthetic aperture radar automatic target recognition. This paper uses the binary target region as the feature and proposes a matching scheme for the target regions using binary morphological operations. The residuals between the testing target region and its corresponding template target regions are processed by the morphological opening operation. Then, a similarity measure is defined based on the residual remains to evaluate the similarities between different targets. Afterward, a Bayesian decision fusion is employed to fuse the similarities gained by different structuring elements to further enhance the recognition performance. The nonlinearity of the opening operation as well as the Bayesian decision fusion makes the proposed method robust to the nonlinear deformations of the target region. Experimental results on the moving and stationary target acquisition and recognition dataset demonstrate the validity of the proposed method.
Various approaches have been proposed for robust visual tracking, among which compressive tracking (CT) yields promising performance. In CT, Haar-like features are efficiently extracted with a very sparse measurement matrix and modeled as an online updated naïve Bayes classifier to account for target appearance change. The naïve Bayes classifier ignores overlap between Haar-like features and assumes that Haar-like features are independently distributed, which leads to drift in complex scenario. To address this problem, we present an extended CT algorithm, which assumes that all Haar-like features are correlated with each other and have multivariate Gaussian distribution. The mean vector and covariance matrix of multivariate normal distribution are incrementally updated with constant computational complexity to adapt to target appearance change. Each frame is associated with a temporal weight to expend less modeling power on old observation. Based on temporal weight, an update scheme with changing but convergent learning rate is derived with strict mathematic proof. Compared with CT, our extended algorithm achieves a richer representation of target appearance. The incremental multivariate Gaussian distribution is integrated into the particle filter framework to achieve better tracking performance. Extensive experiments on the CVPR2013 tracking benchmark demonstrate that our proposed tracker achieves superior performance both qualitatively and quantitatively over several state-of-the-art trackers.
A three-dimensional electromagnetic model (3-D EM-model)–based scattering center matching method is developed for synthetic aperture radar automatic target recognition (ATR). 3-D EM-model provides a concise and physically relevant description of the target’s electromagnetic scattering phenomenon through its scattering centers which makes it an ideal candidate for ATR. In our method, scatters of the 3-D EM-model are projected to the two-dimensional measurement plane to predict scatters’ location and scattering intensity properties. Then the identical information is extracted for scatters in measured data. A two-stage iterative operation is applied to match the model-predicted scatters and the measured data-extracted scatters by combining spatial and attributed information. Based on the two scatter sets’ matching information, a similarity measurement between model and measured data is obtained and recognition conclusion is made. Meanwhile, the target’s configuration is reasoned with 3-D EM-model serving as a reference. In the end, data simulated by electromagnetic computation verified this method’s validity.
This paper proposes a robust method for the matching of attributed scattering centers (ASCs) with application to synthetic aperture radar automatic target recognition (ATR). For the testing image to be classified, ASCs are extracted to match with the ones predicted by templates. First, Hungarian algorithm is employed to match those two ASC sets initially. Then, a precise matching is carried out through a threshold method. Point similarity and structure similarity are calculated, which are fused to evaluate the overall similarity of the two ASC sets based on the Dempster–Shafer theory of evidence. Finally, the target type is determined by such similarities between the testing image and various types of targets. Experiments on the moving and stationary target acquisition and recognition data verify the validity of the proposed method.
A flexible new technique is proposed to calibrate the geometric model of line scan cameras. In this technique, the line scan camera is rigidly coupled to a calibrated frame camera to establish a pair of stereo cameras. The linear displacements and rotation angles between the two cameras are fixed but unknown. This technique only requires the pair of stereo cameras to observe a specially designed planar pattern shown at a few (at least two) different orientations. At each orientation, a stereo pair is obtained including a linear array image and a frame image. Radial distortion of the line scan camera is modeled. The calibration scheme includes two stages. First, point correspondences are established from the pattern geometry and the projective invariance of cross-ratio. Second, with a two-step calibration procedure, the intrinsic parameters of the line scan camera are recovered from several stereo pairs together with the rigid transform parameters between the pair of stereo cameras. Both computer simulation and real data experiments are conducted to test the precision and robustness of the calibration algorithm, and very good calibration results have been obtained. Compared with classical techniques which use three-dimensional calibration objects or controllable moving platforms, our technique is affordable and flexible in close-range photogrammetric applications.
Anomaly detection (AD) becomes increasingly important in hyperspectral imagery analysis with many practical applications. Local orthogonal subspace projection (LOSP) detector is a popular anomaly detector which exploits local endmembers/eigenvectors around the pixel under test (PUT) to construct background subspace. However, this subspace only takes advantage of the spectral information, but the spatial correlat ion of the background clutter is neglected, which leads to the anomaly detection result sensitive to the accuracy of the estimated subspace. In this paper, a local three dimensional orthogonal subspace projection (3D-LOSP) algorithm is proposed. Firstly, under the jointly use of both spectral and spatial information, three directional background subspaces are created along the image height direction, the image width direction and the spectral direction, respectively. Then, the three corresponding orthogonal subspaces are calculated. After that, each vector along three direction of the local cube is projected onto the corresponding orthogonal subspace. Finally, a composite score is given through the three direction operators. In 3D-LOSP, the anomalies are redefined as the target not only spectrally different to the background, but also spatially distinct. Thanks to the addition of the spatial information, the robustness of the anomaly detection result has been improved greatly by the proposed 3D-LOSP algorithm. It is noteworthy that the proposed algorithm is an expansion of LOSP and this ideology can inspire many other spectral-based anomaly detection methods. Experiments with real hyperspectral images have proved the stability of the detection result.
KEYWORDS: 3D acquisition, Synthetic aperture radar, Scattering, 3D image processing, 3D modeling, Feature extraction, Data modeling, Data centers, Automatic target recognition, Error analysis
Additional information provided by three-dimensional (3-D) scattering centers (SCs) is useful in automatic target recognition (ATR). An approach is proposed for 3-D SC extraction from multiple-resolution synthetic aperture radar (SAR) measurements at arbitrary azimuths and elevations. This approach consists of a feature-level extraction and a signal-level optimization. In the feature-level extraction, two-dimensional (2-D) SCs are first extracted at each aspect, then 3-D SCs are coarsely generated from these 2-D SCs by a clustering method. This clustering method contains a particular distance equation and an ingenious clustering strategy which is developed based on some basic properties of scattering physics and the geometric transformation of 3-D SCs and 2-D SCs. Exploiting the sparsity of SCs in the feature domain, such a method efficiently extracts 3-D SCs. In the signal-level optimization, 3-D SC parameters are directly re-estimated using the measurement data. This improves the precision of 3-D SC parameters and provides reliable reconstructions. Finally, the experimental results of data generated by the GTD model and the high-frequency electromagnetic magnetism code exhibit the effectiveness of the proposed approach. In addition, we apply our approach to multiple-pass circle SAR. The reconstructed 3-D SCs exactly depict the shape of the target.
KEYWORDS: Target detection, Matrices, Principal component analysis, Detection and tracking algorithms, 3D acquisition, Hyperspectral imaging, Sensors, 3D modeling, 3D image processing, Hyperspectral target detection
Research on target detection in hyperspectral imagery (HSI) has drawn much attention recently in many areas. Due to the
limitation of the HSI sensor’s spatial resolution, the target of interest normally occupies only a few pixels, sometimes are
even present as subpixels. This may increase the difficulties in target detection. Moreover, in some cases, such as in the
rescue and surveillance tasks, small targets are the most significant information. Therefore, it is very difficult but
important to effectively detect the interested small target. Using a three-dimensional tensor to model an HSI data cube
can preserve as many as possible the original spatial-spectral constraint structures, which is conducive to utilize the
whole information for small target detection. This paper proposes a novel and effective algorithm for small target
detection in HSI based on three-dimensional principal component analysis (3D-PCA). According to the 3D-PCA, the
significant components usually contain most information of imagery, in contrast, the details of small targets exist in the
insignificant components. So, after 3D-PCA implemented on the HSI, the significant components which indicate the
background of HSI are removed and the insignificant components are used to detect small targets. The algorithm is
outstanding thanks to the tensor-based method which is applied to process the HSI directly, making full use of spatial
and spectral information, by employing multilinear algebra. Experiments with a real HSI show that the detection
probability of interested small targets improved greatly compared to the classical RX detector.
Three-dimensional (3D) reconstruction is one of the most attractive research topics in photogrammetry and computer vision. Nowadays 3D reconstruction with simple and consumable equipment plays an important role. In this paper, a 3D reconstruction desktop system is built based on binocular stereo vision using a laser scanner. The hardware requirements are a simple commercial hand-held laser line projector and two common webcams for image acquisition. Generally, 3D reconstruction based on passive triangulation methods requires point correspondences among various viewpoints. The development of matching algorithms remains a challenging task in computer vision. In our proposal, with the help of a laser line projector, stereo correspondences are established robustly from epipolar geometry and the laser shadow on the scanned object. To establish correspondences more conveniently, epipolar rectification is employed using Bouguet’s method after stereo calibration with a printed chessboard. 3D coordinates of the observed points are worked out with rayray triangulation and reconstruction outliers are removed with the planarity constraint of the laser plane. Dense 3D point clouds are derived from multiple scans under different orientations. Each point cloud is derived by sweeping the laser plane across the object requiring 3D reconstruction. The Iterative Closest Point algorithm is employed to register the derived point clouds. Rigid body transformation between neighboring scans is obtained to get the complete 3D point cloud. Finally polygon meshes are reconstructed from the derived point cloud and color images are used in texture mapping to get a lifelike 3D model. Experiments show that our reconstruction method is simple and efficient.
A new partially occluded target location method based on straight line is proposed. It is divided into four steps: firstly, we label the straight lines of concerned target in the history image artificially and store the line points together with the grads orientation. The labeled lines, the length of which is restricted, should distribute symmetrically. Then, transform the stored lines using the transformation model whose parameters are derived from geometry calibration result of the real-time image. Afterwards, construct pyramid structure of real-time image and search the optimal match position. The geometry coherence rule is used to gain holistic optimal match result. Lastly, compare the matching measure with the threshold to decide whether need to perform the same match process using the higher solution image, and output the match result. The experiment results, tested by real-time remote sensing images especially when part of them are occluded, are shown that the proposed algorithm for target location is accurate and effective.
KEYWORDS: Line scan cameras, Cameras, Calibration, 3D modeling, Motion models, Line scan image sensors, 3D image processing, Optical engineering, Data modeling, Image processing
A direct linear transformation (DLT) model is derived to describe the scan imagery of a line scan camera undergoing a uniform rectilinear motion. When more than five points on scan image and their corresponding three-dimensional space points are substituted in the DLT model, 11 coefficients are settled directly and linearly without any approximations. After that, the 11 physically meaningful line scan camera parameters are worked out from the 11 DLT coefficients through a group of analytical operations. The performance is tested and verified by both simulated experiment and demonstration of a real line scan camera.
A novel line-scan camera calibration method in close-range photogrammetry is proposed. Since the line-scan camera is only sensing in one dimension, it's hard to recognize the space points from the linear data captured in static state. To address this problem, the camera is fixed to a programmable linear stage. With the help of the linear stage, a scan image of the pattern is grabbed by the line-scan camera in uniform rectilinear motion state. Therefore, the image points are definitely matched with the space points on the pattern. A pair of projective equations is established to describe this dynamic imaging model, which is determined by six extrinsic camera parameters, five intrinsic camera parameters and three other motion parameters. All the fourteen parameters are estimated approximately by using the direct linear transformation of a reasonably simplified camera model firstly, and then the results are further refined by non-linear least square mean (LSM). Both computer simulated data and real data are used to test our calibration method. The robustness and accuracy are verified by lots of simulated experiments, and for the real data, the root mean square error of re-projected points is less than 0.3 pixels.
The process of relative radiometric calibration (RRC) is an important step for detecting change and monitoring environment through analyzing multi-temporal satellite images. Two key issues that focus on the RRC are how to extract invariant targets that have little or no variation in their reflectance between two images and how to acquire a linear function that expresses the relationship between the digital counts (DNs) of the two images. In this paper, an automatic method for selecting invariant targets which cover the range of bright, midrange, and dark data values is developed, and a robust estimator, which can be effective in tolerating with up to 50 percent outlier contamination, for calculating the gain and the offset of the linear function is described. The proposed methods are automatic and robust. We have applied the proposed method experimentally to synthesized images and real SPOT images, and experiment results have shown the feasibility of our algorithms.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.