Intelligent surveillance systems in human-centered environments require people behavioral monitoring. In this paper, we propose a new Bayesian framework to recognize actions on RGB-D videos by two different observations: the human pose and objects in its vicinity. We design a model for each action that integrates these observations and a probabilistic sequencing of actions performed during activities. We validate our approach on two public video datasets: CAD-120 and Watch-n-Patch. We show a performance gain of 4% in action detection on the fly on CAD-120 videos. Our approach is competitive to 2D image features and skeleton-based methods, as we present an improvement of 16% on Watch-n-Patch. Action recognition performance is clearly improved by our Bayesian and joint human-object perception.

This work presents a method that incorporates 2D and 3D cues for the estimation of head pose. We propose the use of the concept of Key-Frames (KF), a set of frames where the position and orientation of the head is automatically calculated off-line, to improve the precision of pose estimation and detection rate. Each KF consists of: 2D information, encoded by SURF descriptors; 3D information from a depth image (both acquired by an RGB-D sensor); and a generic 3D model that corresponds to the head localization and orientation in the real world. Our algorithm compares a new frame against all KFs and selects the most relevant one. The 3D transformation between both, selected KF and current frame, can be estimated using the depth image and the Iterative Closest Point algorithm in an online framework. Compared to reference approaches, our system can handle partial occlusions and extreme rotations even with noisy depth data. We evaluate the proposal using two challenging datasets: (1) an dataset acquired by us where the ground-truth information is given by a commercial Motion Capture system and (2) the public benchmark Biwi Kinect Head Pose Database.

This paper proposes a unified optimization framework to solve the time-parameterization problem of humanoid-robot paths. Even though the time-parameterization problem is well known in robotics, the application to humanoid robots has not been addressed. This is because of the complexity of the kinematical structure as well as the dynamical motion equation. The main contribution of this paper is to show that the time parameterization of a statically stablepathto be transformed into a dynamically stabletrajectorywithin the humanoid-robot capacities can be expressed as an optimization problem. Furthermore, we propose an efficient method to solve the obtained optimization problem. The proposed method has been successfully validated on the humanoid robot HRP-2 by conducting several experiments. These results have revealed the effectiveness and the robustness of the proposed method.

In this paper, we consider land vehicle positioning in an urban environment. A nonlinear and non-Gaussian model is derived as the process model whereas GPS pseudo-range and odometer measurements are used as observations. A parametric model is proposed for representing the variations of the observation noise covariance matrix. The performances of the derived Gaussian sum filter are evaluated by using experimental data.

In this paper, the imitation of human captured motions by a humanoid robot is considered. The main objective is to reproduce an imitated motion which should be as close as possible to the original human captured motion. To achieve this goal, the imitation problem is formulated as an optimization problem and the physical limits of the humanoid robot are considered as constraints. The optimization problem is then solved recursively by using an efficient dynamics algorithm, which allows the calculation of the gradient function with respect to the control parameters analytically. The simulation results using OpenHRP platform, which is a dynamical simulator for humanoid robot motions, have pointed out that the imitated motions preserve the salient characteristics of the original human captured motion. Moreover the optimization procedure converges well thanks to the analytical calculation of the gradient function.

This paper explores the possibilities of using monocular simultaneous localization and mapping (SLAM) algorithms in systems with more than one camera. The idea is to combine in a single system the advantages of both monocular vision (bearings-only, infinite range observations but no 3-D instantaneous information) and stereovision (3-D information up to a limited range). Such a system should be able to instantaneously map nearby objects while still considering the bearing information provided by the observation of remote ones. We do this by considering each camera as an independent sensor rather than the entire set as a monolithic supersensor. The visual data are treated by monocular methods and fused by the SLAM filter. Several advantages naturally arise as interesting possibilities, such as the desynchronization of the firing of the sensors, the use of several unequal cameras, self-calibration, and cooperative SLAM with several independently moving cameras. We validate the approach with two different applications: a stereovision SLAM system with automatic self-calibration of the rig's main extrinsic parameters and a cooperative SLAM system with two independent free-moving cameras in an outdoor setting.

This paper is an invitation to use mono-vision techniques on stereo-vision equipped robots. By using monocular algorithms on both cameras, the advantages of mono-vision (bearing-only, with infinity range but no 3D instant information) and stereo-vision (3D information only up to a limited range) naturally add up to provide interesting possibilities, that are here developed and demonstrated using an EKF-based monocular SLAM algorithm. Mainly we obtain: a) fast 3D mapping with long term, absolute angular references; b) great landmark updating flexibility; and c) the possibility of stereo rig extrinsic self-calibration, providing a much more robust and accurate sensor. Experimental results show the pertinence of the proposed ideas, which should be easily exportable (and we encourage to do so) to other, more performing, vision-based SLAM algorithms.

Quadratic systems are the first kind of nonlinear systems in which we are interested in order to study polynomial nonlinearities. They can be approximated by bilinear models with nilpotent structure that approximate certain nonlinearities and generate finite degree Volterra series. The hereditary identification algorithm limited until now to linear systems is extended here for identification of nonlinear systems by implementing a canonical structure to the approximant of degree two (quadratic). A NARX (Nonlinear Autoregressive eXogenous input) multidimensional expression is employed in order to perform identification by hereditary computation.

Synthesizing human motion signals is difficult because of its multi-dimensional and nonlinear nature. However, the locomotion is a synchronized motion, which means that these signals are related. Using this property, we propose a new method for modeling the human locomotion and identifying this model. The input signals of our model is the trajectory of pelvis and the outputs are corresponding motion signals of whole human body. To identify this system, we considered it as black-box, for which we propose an adapted method of identification using motion capture. We show that the identified model allows to generate various human locomotion in a fast and efficient way

Most solutions to the SLAM problem in robotics have utilised range and beating sensors as the provided perception data is easy to incorporate, allowing immediate landmark initialization. This is not the case when using bearing-only information because the distance to the perceived landmarks is not directly provided. A whole estimate of a landmark position is only possible via a set of measurements taken from different points of view. The vast majority of contributions to this problem perform a parallel task to get this estimate, and hence the landmark initialization is delayed. We give a new insight to the problem and present a method to avoid this delay by initializing the whole ray that defines the direction of the landmark. We utilize a minimal and computationally efficient form to represent this ray and a new strategy for the subsequent updates. Simulations have been carried out to validate the proposed algorithms.

This paper exhibits an algorithm based on Volterra-type processing in order to detect several independent sources on the same carrier frequency and to determine the number of them. The use of infinite impulse response (IIR) Volterra filtering to build a suitable discrimination test is dictated by the need of higher-order moments in this type of nonlinear problem, as well as the need of IIR for convergence.