Joseph Alizon

Automatic guidance of agricultural vehicles using a laser sensor

This paper presents a method of automatic guidance of an agricultural vehicle in a structured environment (windrow harvester) or an iterative structured environment (combine harvester) using a laser sensor. The sensor parameters are estimated using a correlation based approach. A filter is incorporated so as to limit the perturbations caused by dust. The robustness of the guidance system can be increased by computing a reliability criterion from the estimate model. Vegetation volume and height are calculated and can be applied to control the vehicle velocity.

Segmentation of range images into planar regions

This paper presents a hybrid approach to the segmentation of range images into planar regions. The term hybrid refers to a combination of edge- and region-based considerations. A reliable computational procedure which takes the range image discontinuities into account is presented for computing the pixels normal. The segmentation algorithm consists of two parts. In the first one, the pixels are aggregated according to local properties derived from the input data and are represented by a region adjacency graph (RAG). At this stage, the image is still over-segmented. In the second part, the segmentation is refined thanks to the construction of an irregular pyramid. The base of the pyramid is the RAG previously extracted. The over-segmented regions are merged using a surface-based description. This algorithm has been evaluated on 80 real images acquired by two different range sensors using the methodology proposed in (Hoover et al., 1996). Experimental results are presented and compared to others obtained by four research groups.

Active and intelligent sensing of road obstacles: Application to the European Eureka-PROMETHEUS project

The authors address the problem of road obstacle detection. A sensor composed of an eyesafe laser range finder coupled with a charge coupled device (CCD) camera is proposed. This sensor is mounted in front of a vehicle. The basic idea is to first determine 2-D visual targets in intensity images of the camera. The range finder is then used not only to confirm or reject the real existence of the detected visual targets but also to acquire 3-D information of the confirmed visual targets. The central problem of this strategy is the method of detection of 2-D visual targets from intensity images of a road scene. In the method, line segments are considered as significant features. The concept of a line segment of interest and the concept of a dominant line segment are used. Two-imensional visual targets can be effectively determined with the help of the identification of the dominant line segments in an image. The range finder was used to confirm or reject a 2-D visual target.<<ETX>>

Motion analysis of an articulated locomotion model by video and telemetric data

Traditional techniques of human motion analysis use markers located on body articulations. The position of each marker is extracted from each image. Temporal and kinematic analysis is given by matching these data with a reference model of the human body. However, as human skin is not rigidly linked with the skeleton, each movement causes displacements of the markers and induces uncertainty in results. Moreover, the experiments are mostly conducted in restricted laboratory conditions. The aim of our project was to develop a new method for human motion analysis which needs non-sophisticated recording devices, avoids constraints to the subject studied, and can be used in various surroundings such as stadiums or gymnasiums. Our approach consisted of identifying and locating body parts in image, without markers, by using a multi-sensory sensor. This sensor exploits both data given by a video camera delivering intensity images, and data given by a 3D sensor delivering in-depth images. Our goal, in this design, was to show up the feasibility of our approach. In any case the hardware we used could facilitate an automated motion analysis. We used a linked segment model which referred to Winters model, and we applied our method not on a human subject but on a life size articulated locomotion model. Our approach consists of finding the posture of this articulated locomotion model in the image. By performing a telemetric image segmentation, we obtained an approximate correspondence between linked segment model position and locomotion model position. This posture was then improved by injecting segmentation results in an intensity image segmentation algorithm. Several tests were conducted with video/telemetric images taken in an outdoor surrounding with the articulated model. This real life-size model was equipped with movable joints which, in static positions, described two strides of a runner. With our fusion method, we obtained relevant limbs identification and location for most postures.

Real time road mark following

We present an original method for real time road mark following to be carried out on a motorway and based on a prediction-verification-dating principle. The prediction is made using a model including three dynamic parameters acting linearly, the road is assumed flat with a continuous lateral curve. Detection is based on the analysis of a few lines (<10) of the image. On each line, a probability function allows a prediction of the width and position of the marking strip to be made. Updating uses the least squares method enabling a temporal smoothing of the vector of the models parameters. Reliable results have been obtained on recorded motorway sequences taken at speeds reaching 130 km/h. Integration of the method was made on standard equipment without the need for a special processor.

Visual Tracking by a Multisensorial Approach

Abstract The problem of road obstacle detection and tracking is addressed. The multisensorial approach is based on the use of a mixed camera/3D-sensor. The 3D-sensor is a laser range finder equiped with two mirrors to control the laser beam direction. The multisensor is endowed with two faculties which make it a smart active sensor. They are the controlled perception and the visual servoing which allow a close collaboration between the 3D-world and the intensity world. Obstacle detection is deduced from 3D-image analysis. Tracking combines intensity image processing and visual servoing. Obstacle dynamical state is obtained by a Kalrnan filter.

Tracking systems for intelligent road vehicles

The paper describes two mono-target tracking systems based on the control of a mixed camera/3D sensor. The 3D sensor is a laser range finder. Both systems are presented in an intelligent road vehicle context, for the purpose of obstacle detection. The first is applicable to the pedestrian tracking when the vehicle does not move. The data are obtained by a 3D segmentation process. The second tracking is used for mobile vehicle tracking. In this application, the experimental vehicle is moving, and this tracking is multisensorial, which means that it makes the most of the complementarity of both 3D and intensity data. In that case, two kinds of data are used: the distance of the obstacle obtained by a visual servoing, and the obstacle location in the intensity image given by an image processing. This tracking also involves space and temporal data alignments, which are necessary steps before any data fusion. The data combination is done by an extended Kalman filter.

Building an environment map around the Prolab2 vehicle using a controllable range sensor

The Prolab2 vehicle is the demonstrator for the French ProArt project, which aims at producing a drivers assistant system to increase safety on the roads. The complete system consists of several visual sensors mounted on the car which provide images to various image processing modalities. The information provided by this processing is passed to a data fusion module which generates a complete environment map, which in turn is passed to the copilot system which interprets the perceived environment for possible danger. Alarm, advisory or assistance messages are then passed to the driver via visual and audible means. This paper deals with one of the sensors on the Prolab2 vehicle, the telemetry sensor, and the data fusion module. First the individual sub-systems are described, followed by details of their connection, and finally some experimental results are presented that show the utility of both sub-systems.

SMART SERVOING OF A CONTROLLABLE RANGE SENSOR FOR TARGET TRACKING: APPLICATION TO PEDESTRIANS

Abstract This paper presents a system for 3D target tracking. A controllable range sensor is servoid in order to focus the attention of the system on the target. The 3D-sensor is a laser telemeter based range finder of which the laser beam is deflected by two flat mirrors. The deviations of the mirrors are fully programmable and can be dynamically controlled by an application program. The data from the sensor are passed to a Kalman filter which predicts the next location of the target. The predicted location of the target allows to define an area which is scanned by the laser beam to obtain a 3D image. This system is applied to pedestrian tracking in a road environment.

Real Time 3D Location of a Car from Three Characteristic Points Observed in a Video Image Sequence

Abstract The wide range of potential applications of vision techniques for autonomous vehicles has led the research world to develop efficient algorithms and to implant them in specific architectures to satisfy real time constraints. This paper describes a direct 3D location method used with a tracking algorithm based on the smoothness of motion of all features of the model. Both were implanted in a DSP processors vision machine giving real time information about the following model such as distance, orientation or next frame location.