Jean-Charles Noyer

A Hybrid Particle Approach for GNSS Applications With Partial GPS Outages

To provide an accurate positioning, the land vehicle navigation applications are based on global positioning system (GPS). The addition of a digital road map allows locating the vehicle continuously and helps the driver to get the best path. These systems are usually enhanced with dead reckoning sensors due to GPS outages in urban areas in particular. For instance, the odometer sensors can be used to correct the vehicle location in this case. We present here a global estimation method of solving the fusion problem of the GPS, odometer, and digital road map measurements in the presence of GPS outages. It relies on a hybrid filter that takes advantage of the combination of a Kalman filter, which computes the linear part of the state equations and a particle filter to provide an optimal resolution scheme. When GPS fails, the filter fuses all available pseudorange measures to improve the vehicle positioning. In the case of an urban transport scenario, the results show that the number of particles is significantly reduced to achieve the same performance of a single particle filter in terms of accuracy. Moreover, software solutions can be developed for real-time applications.

Feature Extraction in Scanning Laser Range Data Using Invariant Parameters: Application to Vehicle Detection

This paper presents a feature extraction method in scanning laser range data. Many authors have studied this problem by proposing solutions that rely on a modeling of the scene in Cartesian coordinates. These methods are based on the computation of the interscan distance between two consecutive measurements, which, in practice, is not very easy to estimate. Our proposed method, i.e., segmentation using invariant parameters (SIP), deals with laser measurements in natural coordinates, which avoids any preprocessing stage that could modify the measurement noise statistics. This approach is founded on the use of an invariant description of the feature and leads to the definition of a criterion of line-segment detection that only depends on the sensor intrinsic parameters.

Formulating Robust Linear Regression Estimation as a One-Class LDA Criterion: Discriminative Hat Matrix

Linear discriminant analysis, such as Fishers criterion, is a statistical learning tool traditionally devoted to separating a training dataset into two or even several classes by the way of linear decision boundaries. In this paper, we show that this tool can formalize the robust linear regression problem as a robust estimator will do. More precisely, we develop a one-class Fischers criterion in which the maximization provides both the regression parameters and the separation of the data in two classes: typical data and atypical data or outliers. This new criterion is built on the statistical properties of the subspace decomposition of the hat matrix. From this angle, we improve the discriminative properties of the hat matrix which is traditionally used as outlier diagnostic measure in linear regression. Naturally, we call this new approach discriminative hat matrix. The proposed algorithm is fully nonsupervised and needs only the initialization of one parameter. Synthetic and real datasets are used to study the performance both in terms of regression and classification of the proposed approach. We also illustrate its potential application to image recognition and fundamental matrix estimation in computer vision.

One class proximal support vector machines

Recently in Dufrenois 1, a new Fisher type contrast measure has been proposed to extract a target population in a dataset contaminated by outliers. Although mathematically sound, this work presents some further shortcomings in both the formalism and the field of use. First, we propose to re-express this problem from the formalism of proximal support vector machines as introduced in Mangasarian and Wild 2. This change is far from harmless since it introduces a suited writing for solving the problem. Another limiting factor of the method is that its performance relies on the assumption that the density between the target and outliers are different. This consideration can easily prove to be over-optimistic for real world datasets making the method unreliable, at least directly. The computation of the decision boundary is a time consuming part of the algorithm since it is based on solving a generalized eigenvalue problem (GEP). This method is therefore limited to medium sized data sets. In this paper, we propose appropriate strategies to unlock all these shortcomings and fully benefit from the interest of the approach. Firstly, we show under some conditions that generating appropriate artificial outliers allows to stay within the constraints of the method and thus enlarges the conditions of use. Secondly, we show that the GEP can be advantageously replaced by a conjugate gradient solution (CG) significantly decreasing the computational cost. Lastly, the proposed algorithm is compared with recent novelty detectors on synthetic and real datasets. HighlightsWe study the extraction of a target population from a dataset contaminated by outliers.To this end, we propose a new Fisher type contrast measure.We reconsider this problem from the formalism of proximal support vector machines.An approximation of the contrast measure is done using a conjugate gradient method.No matrix inversion is needed which lowers the computational complexity.

Object detection and tracking using the particle filtering

In this paper, we present a method for detecting and tracking rigid moving objects in a monocular image sequence. The originality of this method lies in a state modelling of this estimation problem which is solved in an unified way. This hybrid estimation problem leads to nonlinear state equations that are solved by the particle filtering. A particle filter is set for each shape model (modes). It estimates the motion and position parameters, tracks the object in the sequence and also computes at each time the probability of all modes.

Non-linear matched filtering for object detection and tracking

In this paper, we present a method for detecting and tracking rigid moving objects in a monocular image sequence. The originality of this method lies in a state modelling of this estimation problem which is solved in an unified way. This hybrid estimation problem leads to non-linear state equations that are solved by the particle method. A particle filter is set for each shape model (modes). It estimates the motion and position parameters and tracks the object in the sequence. The algorithm also computes at each time the probability of all modes. This method is then applied to synthetic and real image sequences in order to evaluate the estimation accuracies and the robustness of the tracking procedure.

Automatic feature extraction in laser rangefinder data using geometric invariance

This paper presents a feature extraction method in scanning laser rangefinder data. Whereas many popular methods use Cartesian coordinates to detect features, the proposed method use a geometric invariant in natural (polar) coordinates. It leads to a membership condition that only depends on the sensor properties (angular resolution and range measurement error).

A particle filtering approach for joint vehicular detection and tracking in lidar data

This paper presents a method for joint detection and tracking of vehicles in scanning laser range data. Many methods use a solution that processes the raw data in a detection procedure and then tracks the detected object in an association/tracking procedure. The proposed approach uses a preclustering stage (SIP) as an input of the tracking process that allows to manage the displacement of the center-of-gravity and the changes in the apparent shape from object and motion modeling. The global problem is then described using a state-space modeling which is solved by a nonlinear filtering method.

An hardware architecture for 3D object tracking and motion estimation

We present a method to track and estimate the motion of a 3D object with a monocular image sequence. The problem is based on the state equations and is solved by a sequential Monte Carlo method. The method uses a CAD model of the object whose projection can be compared directly with the pixels of the image. The advantage is to obtain a better accuracy and a direct estimation of the pose and motion in the 3D world. However, this algorithm needs a massive load in computing. For real-time use, we develop in this paper a distributed algorithm that dispatches the processing between the central processing unit (CPU) and the graphics processing unit (GPU) of a consumer-market computer. Some experimental results show that it is possible to obtain an accurate 3D tracking of the object with low computing costs.

Visual tracking using sequential importance sampling with a state partition technique

Sequential importance sampling (SIS), also known as particle filtering, has drawn increasing attention recently due to its superior performance in nonlinear and non-Gaussian dynamic problems. In the SIS framework, estimation accuracy depends strongly on the choice of proposal distribution. In this paper we propose a novel SIS algorithm called PF-SP-PEKF that is based on a state partition technique and a parallel bank of extended Kalman filters designed to improve the accuracy of the proposal distribution. Our results show that this new approach yields a significantly improved estimate of the state, enabling the new particle filter to effectively track human subjects in a video sequence where the standard condensation filter fails to maintain track lock. Moreover, because of the improved proposal distribution, the new filter can achieve a given level of performance using fewer particles than its conventional SIS counterparts.