Search

The cascade classifier is a usual approach in object detection based on vision, since it successively rejects negative occurrences, e.g., background images, in a cascade structure, keeping the processing time suitable for on-the-fly applications. On the other hand, similar to other classifier ensembles, cascade classifiers are likely to have high Vapnik-Chervonenkis (VC) dimension, which may lead to overfitting the training data. Therefore, this work aims at improving the generalization capacity of the cascade classifier by controlling its complexity, which depends on the model of their classifier stages, the number of stages, and the feature space dimension of each stage, which can be controlled by integrating the parameter setting of the feature extractor (in our case an image descriptor) into the maximum-margin framework of support vector machine training, as will be shown in this paper. Moreover, to set the number of cascade stages, bounds on the false positive rate (FP) and on the true positive rate (TP) of cascade classifiers are derived based on a VC-style analysis. These bounds are applied to compose an enveloping receiver operating curve (EROC), i.e., a new curve in the TP–FP space in which each point is an ordered pair of upper bound on the FP and lower bound on the TP. The optimal number of cascade stages is forecasted by comparing EROCs of cascades with different numbers of stages.

This paper presents a new algorithm for the extrinsic calibration of a perspective camera and an invisible 2D laser-rangefinder (LRF). The calibration is achieved by freely moving a checkerboard pattern in order to obtain plane poses in camera coordinates and depth readings in the LRF reference frame. The problem of estimating the rigid displacement between the two sensors is formulated as one of registering a set of planes and lines in the 3D space. It is proven for the first time that the alignment of three plane-line correspondences has at most eight solutions that can be determined by solving a standard p3p problem and a linear system of equations. This leads to a minimal closed-form solution for the extrinsic calibration that can be used as hypothesis generator in a RANSAC paradigm. Our calibration approach is validated through simulation and real experiments that show the superiority with respect to the current state-of-the-art method requiring a minimum of five input planes.

Current automatic sleep stage classification (ASSC) methods that rely on polysomnographic (PSG) signals suffer from inter-subject differences that make them unreliable in facing with new and different subjects. A novel adaptive sleep scoring method based on unsupervised domain adaptation, aiming to be robust to inter-subject variability, is proposed. We assume that the sleep quality variants follow a covariate shift model, where only the sleep features distribution change in the training and test phases. The maximum overlap discrete wavelet transform (MODWT) is applied to extract relevant features from EEG, EOG and EMG signals. A set of significant features are selected by minimum-redundancy maximum-relevance (mRMR) which is a powerful feature selection method. Finally, an instance-weighting method, namely the importance weighted kernel logistic regression (IWKLR) is applied for the purpose of obtaining adaptation in classification. The classification results using leave one out cross-validation (LOOCV), show that the proposed method performs at the state-of-the art in the field of ASSC.

This paper proposes three novel training methods, two of them based on the backpropagation approach and a third one based on information theory for multilayer perceptron (MLP) binary classifiers. Both backpropagation methods are based on the maximal-margin (MM) principle. The first one, based on the gradient descent with adaptive learning rate algorithm (GDX) and named maximum-margin GDX (MMGDX), directly increases the margin of the MLP output-layer hyperplane. The proposed method jointly optimizes both MLP layers in a single process, backpropagating the gradient of an MM-based objective function, through the output and hidden layers, in order to create a hidden-layer space that enables a higher margin for the output-layer hyperplane, avoiding the testing of many arbitrary kernels, as occurs in case of support vector machine (SVM) training. The proposed MM-based objective function aims to stretch out the margin to its limit. An objective function based on Lp-norm is also proposed in order to take into account the idea of support vectors, however, overcoming the complexity involved in solving a constrained optimization problem, usually in SVM training. In fact, all the training methods proposed in this paper have time and space complexities O(N) while usual SVM training methods have time complexity O(N (3)) and space complexity O(N (2)) , where N is the training-data-set size. The second approach, named minimization of interclass interference (MICI), has an objective function inspired on the Fisher discriminant analysis. Such algorithm aims to create an MLP hidden output where the patterns have a desirable statistical distribution. In both training methods, the maximum area under ROC curve (AUC) is applied as stop criterion. The third approach offers a robust training framework able to take the best of each proposed training method. The main idea is to compose a neural model by using neurons extracted from three other neural networks, each one previously trained by MICI, MMGDX, and Levenberg-Marquard (LM), respectively. The resulting neural network was named assembled neural network (ASNN). Benchmark data sets of real-world problems have been used in experiments that enable a comparison with other state-of-the-art classifiers. The results provide evidence of the effectiveness of our methods regarding accuracy, AUC, and balanced error rate.

Several applications in robotics require complex sensor arrangements that must be carefully calibrated, both intrinsically and extrinsically, to allow information fusion and enable the system to function as a whole. These arrangements can combine different sensing modalities-such as color cameras, laser-rangefinders, and depth cameras-in an attempt to obtain richer descriptions of the environment. Finding the location of multi-modal sensors in a common world reference frame is a difficult problem that is largely unsolved whenever sensors observe distinct, disjoint parts of the scene. This article builds on recent results in object pose estimation using mirror reflections to provide an accurate and practical solution for the extrinsic calibration of mixtures of color cameras, LRFs, and depth cameras with non-overlapping field-of-view. The method is able to calibrate any possible sensor combination as far as the setup includes at least one color camera. The technique is tested in challenging situations not covered by the current state-of-the-art, proving to be practical and effective. The calibration software is made available to be freely used by the research community. [ABSTRACT FROM AUTHOR]

This article presents a new hybrid motion (HM) planner, designed to allow robust indoor navigation in constrained environments of nonholonomic differential robots, such as RobChair, the brain-actuated robotic wheelchair [1] from the Institute of Systems and Robotics, University of Coimbra, Portugal. Relying on this new planning algorithm, RobChair is now able to operate in real dynamic environments and perform challenging maneuvers in narrow spaces. The HM planner integrates deliberative and reactive modules in a three-layer structure: a fast threedimensional (3-D)-global path planner, smoothing, and a new reactive local planner designated the doubledynamic window approach (D-DWA). The 3-D-global path planner consists in the A) algorithm, which defines a path composed by a sequence of (x, y) points, with an interpolation module that has the purpose of assigning an orientation to each (x, y) point. The smoothing algorithm adjusts the (x, y) points to reduce accelerations and jerk associated with the trajectory. The D-DWA is in charge of dynamically adapting the robot motion, taking into account the robot geometry (noncircular robot) and local static/dynamic obstacles, unknown from the global planner perspective. Real-time navigation is achieved because both the smoothing and the D-DWA algorithms are iteratively executed during navigation. The use of multiresolution local grid maps also contributes to the increase of computation performance. To show the effectiveness of the proposed planning algorithm, results of real navigation experiments are reported in this article. The experiments consist in steering RobChair in a real office-like scenario by different participants, using a selfpaced P300-based brain-computer interface (BCI). [ABSTRACT FROM AUTHOR]

This paper presents, in detail, the implementation of a new control strategy, Kalman-based active observer controller (AOB), for the path following of wheeled mobile robots (WMRs) subject to nonholonomic constraints. This control strategy presents some particularities as being used in discrete mode, and being robust against uncertainties and disturbances such as the ones due to the use of the input-output feedback-linearization method in discrete mode, while it was developed to be used in continuous mode. The performance of the proposed control algorithm is verified via computer simulation, and is compared with other control strategies, such as pole placement controller (PPC) and PPC with a Kalman filter observer (CKF). [ABSTRACT FROM AUTHOR]

This paper describes a data fusion architecture for robotic assembly tasks based on human sensory-motor skills. These skills are transferred to the robot through geometric and dynamic perception signals. Artificial neural networks are used in the leaning process. The data fusion paradigm is addressed. It consists of two independent modules for optimal fusion and filtering. Kalman techniques linked to stochastic signal evolutions are used in the fusion algorithm. Compliant motion signals obtained from vision and pose sense are fused, enhancing the task performance. Simulations and peg-in-hole experiments are reported. [ABSTRACT FROM AUTHOR]

Cooperation between road users through V2X communication is a way to improve GNSS localization accuracy. When vehicles localization systems involve standalone GNSS receivers, the resulting accuracy can be affected by satellite-specific errors of several meters. This paper studies how road-features like lane marking detected by on-board cameras can be exploited to reduce absolute position errors of cooperative vehicles sharing information in real-time in a network. The algorithms considered in this work are based on a error bounded set membership strategy. In every vehicle, a set membership algorithm computes the absolute position and an estimation of the satellite-specific errors by using raw GNSS pseudoranges, lane boundary measurements and a 2D georeferenced road map which provides absolute geometric constraints. As lane-boundary measurements provide essentially cross-track corrections in the position estimation process, cooperation enables the vehicles to improve their own estimates thanks to the different orientation of the roads. Set-membership methods are very efficient to solve this problem since they do not involve any independence hypothesis of the errors and so, the same information can be used several times in the computation. Such class of algorithm provides a novel approach to improve position accuracy for connected vehicles guaranteeing the integrity of the computed solution which is pivoting for automated automotive systems requiring guaranteed safety-critical solutions. Results from simulations and real experiments show that sharing position corrections reduces significantly satellite-specific GNSS errors effects in both cross-track and along-track components. Moreover, it is shown that lane-boundary measurements help reducing estimation errors for all the networked vehicles even those which are not equipped with an embedded perception system. [ABSTRACT FROM AUTHOR]

The article describes a reconstruction pipeline that generates piecewise-planar models of man-made environments using two calibrated views. The 3D space is sampled by a set of virtual cut planes that intersect the baseline of the stereo rig and implicitly define possible pixel correspondences across views. The likelihood of these correspondences being true matches is measured using signal symmetry analysis [1] , which enables to obtain profile contours of the 3D scene that become lines whenever the virtual cut planes intersect planar surfaces. The detection and estimation of these lines cuts is formulated as a global optimization problem over the symmetry matching cost, and pairs of reconstructed lines are used to generate plane hypotheses that serve as input to PEARL clustering [2] . The PEARL algorithm alternates between a discrete optimization step, which merges planar surface hypotheses and discards detections with poor support, and a continuous optimization step, which refines the plane poses taking into account surface slant. The pipeline outputs an accurate semi-dense Piecewise-Planar Reconstruction of the 3D scene. In addition, the input images can be segmented into piecewise-planar regions using a standard labeling formulation for assigning pixels to plane detections. Extensive experiments with both indoor and outdoor stereo pairs show significant improvements over state-of-the-art methods with respect to accuracy and robustness. [ABSTRACT FROM AUTHOR]

Artificial perception, in the context of autonomous driving, is the process by which an intelligent system translates sensory data into an effective model of the environment surrounding a vehicle. In this paper, and considering data from a 3D-LIDAR mounted onboard an intelligent vehicle, a 3D perception system based on voxels and planes is proposed for ground modeling and obstacle detection in urban environments. The system, which incorporates time-dependent data, is composed of two main modules: (i) an effective ground surface estimation using a piecewise plane fitting algorithm and RANSAC-method, and (ii) a voxel-grid model for static and moving obstacles detection using discriminative analysis and ego-motion information. This perception system has direct application in safety systems for intelligent vehicles, particularly in collision avoidance and vulnerable road users detection, namely pedestrians and cyclists. Experiments, using point-cloud data from a Velodyne LIDAR and localization data from an Inertial Navigation System were conducted for both a quantitative and a qualitative assessment of the static/moving obstacle detection module and for the surface estimation approach. Reported results, from experiments using the KITTI database, demonstrate the applicability and efficiency of the proposed approach in urban scenarios. [ABSTRACT FROM AUTHOR]