International audience

This article presents a 3D approach in plant growth monitoring and deals with the tracking of leaves of sunflower plants. Our aim is to compute time-series of individual leaf area, under water stress and control conditions. These data will then be used by biologists to study the drought resistance of various sunflower species. Our method to track the leaves in 3D has been evaluated on a set of 132 point clouds obtained via classical structure-from-motion techniques and multi-view stereo software. These 3D acquisitions have been performed on 12 sunflower plants (6 water-stressed, 6 well-watered) during a period of one month (11 measurement dates per sunflower plant). This method gives promising results for both conditions (water-stressed and well-watered), for different species and is able to follow the growth of the plants, as well as to detect new leaf emergence and leaf decay.

International audience

International audience

3D perception has known impressive advances in the past 3 years; it corresponds to several technological improvements, plus many new development teams providing open sources. First of all, researchers in Robotics and 3D Perception have made profit of the Kinect sensor; some works were already devoted to 3D cameras, using more expensive Time-of-Flight optical devices. Another common way to acquire dense 3D data, is by scanning the environment by a laser range finder (LRF); as for example, the Hokuyo tilting LRF integrated on the PR2 robot by Willow Garage. To build a dense geometrical model of an indoor environment, several sensors could be selected in order to acquire 3D data. This paper aims at giving some insights on this selection, presenting some pros and cons for Kinect, Hokuyo and ToF optical sensors.

Human motion is a critical aspect of interacting, even between people. It has become an interesting field to exploit in human-robot interaction. Even with today's computing power, it remains a difficult task to successfully follow the human's motion from image processing alone. New sensors were introduced, bringing depth sensing at low or no cost. Using this new technology, this paper presents a new methodology to see space with multiple depth sensors, using machine-learning technique, and features in voxel space to learn to reconstruct humans' joints in single, fused acquisitions. We back up and validate the procedure with ground truth acquired from commercial Motion Capture, and prove the approach to perform particularly well on an expansive set of motion and poses, and compare with current standard software on single depth sensors.

This paper proposes a novel approach that allows region-based active contour energy to be re-expressed combining local and global information. The basic idea of this technique consists in extracting image statistics locally from the heterogeneous region (foreground or background) and globally from the other region at each point along the curve. By exploiting benefits of both local-based and global-based statistics, this technique proves to be robust against heterogeneity and noise and shows low sensitivity to curve initialization. Experimental results for synthetic and real images reveal significant improvement compared to conventional methods.

An important challenge in service robotics is to design user interfaces and interaction capabilities that can be used intuitively to enable the potential users to benefit from the robot's functionalities and to allow effective collaborative task execution. The Interactive Behavior Operated shopping Trolley (InBOT) addresses this issue at the example of supporting complex shopping tasks in large supermarkets. The actual control of the robot is shared or traded between the robot and it's dedicated user ranging from closely coupled haptic-based interaction via observation-based interaction up to loosely coupled command-based interaction. The four modes of operation of InBOT, i.e. the manual steering, following, guiding and autonomous mode, the transitions between them and the relevant modalities of interaction are introduced. Finally the paper is completed by describing the results of user studies.

In this work, we present the implementation in a reconfigurable architecture of a dense stereo vision algorithm based on census transform. Analyzing census transform algorithm we found that size and access memory could be reduced, which consequently also reduced the latency time. Furthermore, architecture resources are optimized and efficient thanks to binary operations and integer arithmetic used by census transform directly compatible with the FPGA. Final architecture is able to construct 130 dense disparity maps per second for each corresponding pair of stereo images. A performance analysis, among other three disparity map implementations and our architecture, shows that at the end, we propose a better trade off among performance, latency, logic elements and memory size. The optimization and the resource saving rend our architecture an interesting option to solve the problem of stereo vision in real time, quite used in autonomous navigation.

This article presents a reconfigurable architecture to calculate a dense disparity map of two stereo images based on census transform. This architecture is simplified and efficient as a result of binary operations and integer arithmetic used by census transform. Our architecture was prototyped using GAUT which is a practical tool to develop high-level synthesis. We take advantage of GAUT rapid prototyping to implement different architectures and to make a general comparison among them, that lets us to optimize the architecture, and at the same time, to improve the system's performance. The optimization and the resource saving rend our architecture an interesting option to solve the problem of stereo vision in real time, quite used in autonomous navigation.

In this work, we present a new approach for optimum design of nonlinear filters based on support vector machines. Taking advantage on the general concept of binary filters and machine learning theory, this proposed approach, is based on the concept of a new filter structured, called support vector machine filter (SVMF) and statistical data analysis. This proposed filter approach, is used as an impulsive noise image denoising. The results obtained for the application at hand show that the proposed filter outperforms a new algorithm for elimination of impulsive noise recently reported and center weighted median in the image denoising task. The proposed filter can be successfully applied for the processing of images corrupted with impulsive noise while maintaining the visual quality and a low reconstruction error.

This paper deals with the tracking of persons in a human cluttered environment. It is performed by an active perception system, consisting of a camera mounted on a pan-tilt unit and a 360° RFID detection system which are embedded on a mobile robot. Particle filters enable the fusion of heterogeneous data into the proposal distribution from which the particles are sampled. The information provided by the tracker is then used to build sensor-based dedicated control laws in order to make the robot follow the RFID tagged person. Finally, experiments on our mobile robot are presented in order to highlight the relevance and complementarity of the developed perceptual functions.

This paper presents 6-DOF monocular EKF-SLAM with undelayed initialization using linear landmarks with extensible endpoints, based on the Plu¿cker parametrization. A careful analysis of the properties of the Plu¿cker coordinates, defined in the projective space P5, permits their direct usage for undelayed initialization. Immediately after detection of a segment in the image, a Plu¿cker line is incorporated in the map. A single Gaussian pdf includes inside its 2-sigma region all possible lines given the observed segment, from arbitrarily close up to the infinity range, and in any orientation. The lines converge to stable 3D configurations as the moving camera gathers observations from new viewpoints. The line's endpoints, maintained out of the map, are constantly retro-projected from the image onto the line's local reference frame. An extending-only policy is defined to update them. We validate the method via Monte Carlo simulations and with real imagery data.

This paper presents a new variant of the classical Iterative Closest Point (ICP) algorithm. The ICP result depends heavily on the matching between 3D points of the two cloud of points to be registered : besides the 3D distance between points, many other attributes have been proposed to find good matchings, and especially, when available, color or texture attributes could be compared to confirm pairings between closest points. In our system, 3D data are acquired by a monochromatic stereovision sensor ; a simple method is proposed to associate a pseudo-color to every 3D point, only from its position in the cloud. Our ICP implementation is slower than the classic ICP algorithm but it is shown that it gives better registration, either on our stereo data or on data sets used for the benchmarking of 3D modelling algorithm.

This paper presents a Markov Chain Monte Carlo (MCMC) based particle filter to track multiple persons dedicated to video surveillance applications. This hybrid tracker, devoted to networked intelligent cameras, takes benefit from the best properties of both MCMC and joint particle filter. A saliency map-based proposal distribution is shown to limit the well-known burst in terms of particles and MCMC iterations. Qualitative and quantitative results for real-world video data are presented.

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.

Some recent stereo matching algorithms are based on graph cuts. They transform the matching problem to a minimisation of a global energy function. The minimisation can be done by finding out an optimal cut in a special graph. Different methods were proposed to construct the graph. But all of them, consider for each pixel, all possible disparities between minimum and maximum values. In this article, a new method is proposed: only some potential values in the disparity range are selected for each pixel. These values can be found using a local analysis of stereo matching. This method allows us to make wider the disparity range, and at the same time to limit the volume of the graph, and therefore to reduce the computation time.

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.

This paper describes an architecture dedicated to the real-time processing of census correlation in the context of the realization of passive stereovision sensors. Although DSP circuits have dramatically increased their performances in terms of frequency (about 600 MHz today), DSP cores (several Multipliers Accumulators) and pipelines (Super Harvard Architectures for example), FPGA circuits remain the best way to design massive parallel architectures when ultra fast algorithms computation are needed like it is the case in real time vision systems for collision avoidance.

This work describes a navigation framework for robots in semi-structured outdoor environments which enables planning of semantic tasks by chaining of elementary visual-based movement primitives. Navigation is achieved by understanding the underlying world behind the image and using these results as a guideline to control the robot. As retrieving semantic information from vision is computationally demanding, short-term tasks are planned and executed while new vision information is processed. Thanks to learning techniques, the methods are adapted to different environment conditions. Fusion and filtering techniques provide reliability and stability to the system. The procedures have been fully integrated and tested with a real robot in an experimental environment. Results are discussed.

One of the main needs for the humanitarian demining procedures in Colombia and in the whole world is to find complementary sensing methods for the metal detectors. This paper presents a study whose main goal is the development of a complementary sensor for a metal detector, based on the analysis of images taken by an on-board camera. The images are processed by developing texture and color analysis of different types of surfaces, composed mainly by grass or small rocks. Based on the color and texture analysis, this paper shows that it is possible to determine the presence of a non-metallic landmine within the image. This is achieved by finding the image regions whose texture features variations and color compositions are considered to be anomalous. It is thought that the texture variations over the image are an indicator that the surface has been intentionally modified. This image processing strategy is being integrated on the robot 'Ursula', designed at Universidad Javeriana in Colombia.

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 article describes visual functions dedicated to the extraction and recognition of planar quadrangles detected from a single camera. Extraction is based on a relaxation scheme with constraints between image segments, while the characterization we propose allows recognition to be achieved from different view-points and viewing conditions. We defined and evaluated several metrics on this representation space - a correlation-based one and another one based on sets of interest points.

This article describes the integration in a whole navigation system of visual functions dedicated to the extraction and recognition of visual landmarks, i.e. planar quadrangles detected from a single camera. The extraction of these landmarks is based on a relaxation scheme, used to satisfy constraints between image segments. During the exploration of an indoor environment, according to the current topology, there are several ways to take advantage of these visual landmarks. Two representations are considered, relying both on laser data and or our visual landmarks system: a GVG-based model (generalized Voronoi graph) for corridors, and a composite stochastic map for open spaces. In the particular case of corridors, the corresponding modules - laser GVG construction and visual landmarks extraction and recognition - can cooperate to complete each other, as image processing can be enhanced by some structural knowledge about the scene, whereas the GVG is annotated, even as far as its edges are concerned, by qualitative visual information.

This paper describes how to integrate tracking functions on a mobile robot. Tracking is required during autonomous robot navigation in different situations: landmark tracking to guarantee real-time robot localization, target tracking for a sensor based motion or obstacle tracking to control an obstacle avoidance procedure. These objects - landmarks, targets, and obstacles - have different characteristics: this paper shows that a single tracking method could not overcome all the tracking tasks. Several methods must be integrated on the robot. A tracker controller is in charge to activate a specific method with respect to the object type and to a transition model used to recover the tracking failures. This method is validated using four trackers, based on: template differences, set of points, lines and snakes. Experimental results provided on numerous image sequences are presented.

This paper presents new techniques for improving the structural quality of automatically acquired architectural 3D models. Common architectural features like parallelism and orthogonality of walls and edges are exploited. The location of these features is extracted from the model by using a probabilistic technique (RANSAC). The relationships among the planes and edges are inferred automatically using a knowledge-based architectural model. A numerical algorithm is used to optimise the orientations of the features. Small irregularities in the model are removed by projecting the triangulation vertices onto the features. Planes and edges in the resulting model are aligned to each other. The techniques produce models with improved appearance. We show results for synthetic and real data with consideration of noise.

Presents vision functions needed on a mobile robot to deal with landmark-based navigation in buildings. Landmarks are planar, quadrangular surfaces, which must be distinguished from the background, typically a poster on a wall or a door-plate. In a first step, these landmarks are detected and their positions with respect to a global reference frame are learned; this learning step is supervised so that only the best landmarks are memorized, with an invariant representation based on a set of interest points. Then, when the robot looks for visible landmarks, the recognition procedure takes advantage of the partial Hausdorff distance to compare the landmark model and the detected quadrangles. The paper presents the landmark detection and recognition procedures, and discusses their performances.

In this paper a complete strategy for scene modelling from sensory data acquired in a natural environment is defined. This strategy is applied to outdoor mobile robotics, from environment recognition to landmark extraction. In this work, an environment is understood as a specific kind of landscape, for instance prairie, forest, desert, etc. A landmark is defined as a conspicuous object in the environment. In the context of outdoor mobile robotics a landmark has to be useful to perform localization and navigation tasks.

This article describes the integration in a complete navigation system of an environment modeling method based on a Generalized Voronoi Graph (GVG), relying on laser data, on the one hand, and of a localization method based on monocular vision landmark learning and recognition framework, on the other hand. Such a system is intended to work in structured environments. It is shown that the two corresponding modules - laser GVG construction and visual landmarks learning and recognition - can cooperate to complete each other, as image processing can be enhanced by some structural knowledge about the scene, whereas the GVG is annotated, even as far as its edges are concerned, by qualitative visual information.

The paper presents a video surveillance system which is able to measure the load displacements in an aircraft cargo area as well as the appearance of fires (hot surfaces, flame and smoke). To benefit from both measurements and cost savings, the system is based on an uncooled CCD camera operating in a near infrared (NIR) spectral band. In this article, we focus on load displacement measurements. The cargo compartment is not pressurized and the temperature is not controlled, thus it may vary from -55/spl deg/C to 85/spl deg/C. The influence of spectral band and temperature variations on a spatial resolution camera is studied using the physical models of a CCD camera. The classical camera geometric model is modified to take into account these variations. After the load detection, the geometric model is used to localize the load in the cargo area. Finally, we present the image processing software which computes the 3D load displacement.

This paper presents a new hybrid range image segmentation approach. Two separate techniques are applied consecutively. First, an edge based segmentation technique extracts the edge points-creases and jumps-contained in the given range image. Then, by using only the edge point position information, the boundaries are computed. Secondly, the points clustered into each region are approximated by single surfaces through a genetic algorithm (GA). The GA takes advantage of previous edge representation finding the surface parameters that best fit each region. It works in a local way, according to the boundary information, reducing considerably the required CPU time. Experimental results with different range images are presented; moreover a comparison using either the edge detection stage or not is given.

We present a new approach to sensor-based navigation in cluttered environments. In our system, tasks are specified in terms of visual goals, and obstacles are detected by a laser range finder. To effect task performance, we introduce a new gain scheduling visual servo controller. Our approach uses a diagonal gain matrix whose entries are adjusted during execution according to one of several proposed gain schedules. Obstacle avoidance is achieved by allowing the detected obstacles to generate artificial repulsive potential fields, which alter the motion of the mobile robot base. Since this motion affects the vision-based control, it is compensated by corresponding camera motions. Finally, we combine the obstacle avoiding and visual servo components of the system so that visual servo tasks can be performed as obstacles are avoided. We illustrate our approach with both simulations and real experiments using our experimental platform H/sup il/are2Bis.

We present an object tracking system based on an edge model for the target characterization. The target position is estimated by looking for the model in the current image using a Hausdorff partial distance. Target is searched only in a sub-window of current edge image. Its boundaries are determined by Kalman filter estimation that uses target dynamics to predict the current position. We use a spiral searching strategy to find the actual position. The target model is updated in each iteration by using unidirectional partial distance from the image to the model. This model is refined by an enclosure operator in order to perform the target/background discrimination. The parameters of our system can be modified in an active way along the tracking task. The system is shown to be robust to illumination changes and pose variations. The system has been also embedded in a mobile robot for personal robotics applications and integrated in a real-time OS (3 Hz).

A machine vision system based on a CCD camera combined with a light beams matrix is developed to output the 3D surface shape of vehicle cockpit occupancy. This paper focuses on the problem of image spots labeling. An initial set of possible spot/beam matching is deduced from epipolar constraints provided by the prior calibration of the relative camera/projector position. Then, using the topological constraints in the 2D mesh of illuminated dots, a constraint propagation process, based on relaxation or maximal cliques search, eliminates most of the possible combinations from the initial set of matching. Results of the labeling process, obtained by the two methods, are then discussed.

Describes the localization function integrated in a landmark-based navigation system. It relies on the use of planar landmarks (typically, posters) to localize the robot. It is based on two periodic processes running at different frequencies. One of them performs the poster tracking (based on the partial Hausdorff distance) and the active control of the camera. The other one runs on a lower frequency and localizes the robot thanks to the tracked landmarks, the positions of which have been learnt during an offline exploration step. The system has been embedded on our indoor Hilare mobile robot and works in real time. Experiments, illustrated in the paper, demonstrate the validity of the approach.

Presents the motion control for our mobile robot Hilare2Bis using both visual servoing and potential fields in a, cluttered indoor environment. This function is required in order to execute navigation tasks in a robust way. The camera is mounted on the robot close to the end effector of an arm with 6 DOF, used as a virtual pan and tilt platform: only two DOFs are controlled (/spl theta//sub pan/, /spl theta//sub tilt/) by the visual servoing process. The mobile platform is controlled according to the position of some features in an image, and to the distance to the obstacles detected by a 2D laser range finder.

Presents a controller for driving a mobile robot towards a target in a cluttered environment. The proposed method combines visual servoing techniques allowing the target to be tracked, with an obstacle avoidance strategy based on the information provided by a 2D laser range sensor. Simulation results are presented at the end of the paper.

Describes a two-level representation of indoor environments as an intermediate goal towards the topological description of the navigation area in which a mobile robot performs its tasks. Geometric features detected by the exteroceptive sensors of the vehicle are grouped into landmarks, which are characterized by a local reference frame. Each of these landmarks is expressed with respect to a global reference frame. The proposed approach provides a common framework for both the absolute location of the vehicle with respect to the global reference frame, and for its relative location with respect to the landmark frames. A complete representation of the environment is incrementally constructed while the mobile robot is relocalized along its trajectory. A probabilistic representation of uncertain geometric information is used (the SP-model). Experimental results obtained with the mobile robot HILARE-2bis and its 2D laser range finder are presented to validate the approach.

Deals with perception functions required in an autonomous robot which must move in unknown natural environments. From a sequence of range images acquired during the motion, the robot must incrementally build a model and correct its situation estimate. This approach is suitable for environments in which (1) the terrain is mostly flat and (2) obstacles can be distinguished from the ground. The model gives ground and obstacle representations. In previous works, some obstacles emerging from the ground were selected as landmarks; in this paper, we present a new method: the boundary lines between the ground and the obstacles, can be approximated by B-spline curves so that stable curvature information can be extracted. The robot localization is performed from matchings between these curves perceived from different positions and from the numerical fusion between their models. Several experiments on actual images acquired in a lunar-like environment, are presented and analyzed.

An important task in most 3D vision systems is camera calibration. Many camera models, numerical methods and experimental set-ups have been proposed in the literature to solve the calibration problem. We have analysed and tried many methods, and we conclude that the main problems lie in the choice of the numerical methods and on the calibration object. We propose in this paper a method which is based on a camera model that incorporates lens distortion, and involves a nonlinear minimization technique which can be performed using multiple views of a single 2D object and subpixel feature extraction. We present an application for which only a 2D calibration object can be used.

Building on previous work on incremental natural scene modelling for mobile robot navigation, we focus in this paper on the problem of representing and managing uncertainties. The environment is composed of ground regions and objects. Objects (e.g., rocks) are represented by an uncertain state vector (location) and a variance-covariance matrix. Their shapes are approximated by ellipsoids. Landmarks are defined as objects with specific properties (discrimination, accuracy) that permit to use them for robot localization and for anchoring the environment model. Model updating is based on an extended Kalman filter. Experimental results are given that show the construction of a consistent model over tens of meters.

This paper deals with the perception subsystem of a mobile robot which must navigate in a structured environment. We will consider the exploration task; the robot must build successive snapshot models from sensory data acquired from a laser range finder (LRF), and fuse them in a global model so that it can localize itself with respect to a pertinent reference frame. Structuration rules are required to limit the complexity of the modeling process. First of all, we only extract from the sensory data, useful landmarks which correspond to characteristic local feature groupings, like wall corners, doors, corridor crossings, ...; each landmark has its own frame and its own geometrical model. Then, correlated landmarks are kept together at an area level; finally, the global model is built from the relationships between area frames, providing the topological description of the world. For each level (landmark, area, environment), the model is represented by a random vector and a covariance matrix, updated through the use of an extended Kalman filter (EKF).

Mobile robot navigation in natural environments requires incremental scene modelling and robot self-localization. These two capacities are discussed in this paper. A scene model is built with heterogeneous representations to match the diversity of the environment. We focus on object-based representations, using coarse models (ellipsoids) or volumic primitives (superquadrics), and their topological relations. The method for extracting the models from a range image to provide geometrical and topological representations, and the matching of successive views during the robot's movement are presented. For robot localization, two procedures are proposed based on the coarse or fine models. Experimental results are given.

This paper concerns computer vision dedicated to autonomous manipulation tasks. A 6 d.o.f. manipulator is provided with a wrist mounted camera; the grasping task on a priori known objects must provide the best accuracy for the object position in the gripper. A "Look and Move" strategy is proposed to move the gripper towards the optimal grasping position. We try to evaluate the accuracy for the global operation; the different sources of error (camera calibration, hand-eye calibration and object localization) are identified and measured by specific experiments, in order to minimize the final error in the gripper position with respect to the object.

In this paper, we focus on some perceptual functions required by a generic task "GOTO" in natural environments; in previous works, only geometrical modeling has been used to deal with two fundamental tasks: landmark extraction and recognition for sensor-based motion control or robot localization, and terrain modeling for motion planning. Geometrical representations alone lead to a bulky model, and, after some iterations, to a combinatorial explosion. We present here, higher level representations; from a range image, given some assumptions on the perceived scene (even ground with few objects), we propose a segmentation algorithm to extract simple semantical representations for the ground and the objects; then, we can analyze the relative positions of the objects to build a topological scene description. Both models constitute the scene model, needed for further incremental environment modeling.>

This paper deals with the perception subsystem of an autonomous mobile robot which must navigate in structured or semistructured environments. For intervention in high risk zones, mismatches between an a priori known model of the environment and the perceived scenes must be considered, as well as the possibility that the ground is not necessarily horizontal. The incremental building of a model used for localization tasks only is addressed. It contains the representations of particular 3-D features extracted from depth images acquired by a laser range finder. Perceiving the same features again allows reestimation of the positions of the robot and the localization features in the environment. The problem of using simultaneously both previous kind of localization models is particularly considered.

As an autonomous robot navigates in an unknown environment, the perception subsystem must be able to perform the incremental modeling of this environment as well as robot self-location. In this paper, the authors have chosen to combine a polyhedral representation of the world with its digital elevation model, in order to deal with autonomous navigation in semi-structured environments. The authors describe algorithms needed to build snapshot models from noisy and sparse range data, to perform 3D data fusion between these snapshot models, in order to incrementally build a reliable 3D model from which a path planner could generate safe trajectories. The authors present a fully implemented modeling strategy based on range data processing, and describe the different modeling services required in order to perform the indoor scene navigation of the robot HILARE-2.

The authors consider the perception subsystem of an autonomous mobile robot which must be able to navigate in 3D terrain. They describe their approach to building a rough geometric model of a 3D terrain accounting for the locomotion capabilities of the vehicle, using a laser range finder. This model may be used as direct input for the robot's path planner. The terrain model relies on two grid-based representations: the local elevation map and the local navigation map. Both are incrementally built at arbitrary resolution using new interpolation and localization methods and other 3D vision techniques. The authors validate the proposed approach by presenting some comprehensive results using real range images of an indoor structured environment.>

The authors show how color and texture image segmentation, automatic image classification, camera calibration, and 3D scene representation can cooperate to solve a complex problem such as selective cutting. Attention is focused on the field of ore recognition, where significant improvement has been obtained by texture information. The research project considered involves the automation of the cutting operation of a roadheader for selective cutting in an underground potash mine near Barcelona in Spain. The system described is based on the use of computer vision to discriminate the different ore types found in the face (sylvinite, carnalite, and salt). Using the information about the ore distribution, paths are then planned for the computer-controlled cutting boom. It was shown that color information was important for minerals identification but texture information was absolutely necessary to get a good identification in all cases. The two kinds of information cooperate in an automatic image classification algorithm, which has been validated on many images of the mine face.>

A model-based method for identifying convex or nonconvex polyhedral objects from a single image is described. An object is known by a region-edge-vertex model and an aspect graph. A few hypotheses are searched in a topological compatibility graph in which each node corresponds to a matching between chains of segments extracted from the image and model faces. The construction of this graph depends only on qualitative or topological criteria (number of edges for a face, syntactic description, perceptual organization, visibility, etc.). The search for hypotheses in the graph relies on the maximal cliques technique modified to deal with the global topological compatibility.>