Ettore Stella

Mobile robot position determination using visual landmarks

This paper is concerned with the problem of determining the position of a mobile vehicle during navigation. In order to achieve this objective a multisensor navigation system for self location of the robot has been developed. By tracking a few known landmarks with a vision module, the system is able to monitor continuously its position and to integrate these estimates with the measures provided by the vehicle odometers. This paper describes in detail the vision module used by the navigation system. >

Visual recognition of fastening bolts for railroad maintenance

This paper presents a vision-based technique to automatically detect the absence of the fastening bolts that secure the rails to the sleepers. The images are pre-processed by using several combinations of WT and PCA methods.The final detecting system has been applied on a long sequence of real images showing a high reliability and robustness.

Self-location for indoor navigation of autonomous vehicles

Accurate position estimation is a fundamental requirement for mobile robot navigation. The positioning problem consists of keeping in real-time a reliable estimate of the robot location with respect to a reference frame in the environment. A fast landmark-based position estimation method is presented in this paper. The technique combines orientation of the mobile robot from a heading sensor (a compass) with observations of landmarks from a vision sensor (a CCD camera). Knowing the position of the landmarks in a fixed coordinate system and the orientation of the optical axis of the camera its possible to recover the robot position by simple geometric considerations. The experiments made in our laboratory demonstrate the reliability of the method and suggest its applicability in the context of autonomous robot navigation.

Ball detection in static images with Support Vector Machines for classification

We present a general method for detecting balls in images at the aim of automatically detecting goals during a soccer match. The detector learns the object to detect by using a supervised learning scheme called Support Vector Machines, in which the examples are views of the object. Due to the attitude of the camera with respect to football ground, the system can be thought of as an electronic linesman which helps the referee in establishing the occurrence of a goal during a soccer match. Numerous theoretical and practical issues are addressed in the paper. The first one concerns the determination of negative examples relevant for the problem at hand and the training of a reference classifier in the case of an unbalanced number of positive and negative examples. The second one focuses on the reduction of the computational complexity of the reference classifier during the test phase, without increasing its generalization error. The third issue regards the problem of parameter selection, which is equivalent, in our context, to the problem of selecting, among the classifiers the machine implements, the one having performances similar to the reference classifier. Experimental results on real images show the performances of the proposed detection scheme.

Mobile robot navigation by multi-sensory integration

A strategy and a control architecture to allow a mobile robot to navigate in an indoor environment on a planned path is described. The robots control system includes several processes which run in parallel by using specialized hardware. The navigation subsystem of the mobile robot integrates the position estimation obtained by a vision system with the position estimated by odometry using a Kalman filter framework. Obstacle detection is performed by means of a set of ultrasonic sensors.<<ETX>>

Data representations and generalization error in kernel based learning machines

This paper focuses on the problem of how data representation influences the generalization error of kernel based learning machines like support vector machines (SVM) for classification. Frame theory provides a well founded mathematical framework for representing data in many different ways. We analyze the effects of sparse and dense data representations on the generalization error of such learning machines measured by using leave-one-out error given a finite amount of training data. We show that, in the case of sparse data representations, the generalization error of an SVM trained by using polynomial or Gaussian kernel functions is equal to the one of a linear SVM. This is equivalent to saying that the capacity of separating points of functions belonging to hypothesis spaces induced by polynomial or Gaussian kernel functions reduces to the capacity of a separating hyperplane in the input space. Moreover, we show that, in general, sparse data representations increase or leave unchanged the generalization error of kernel based methods. Dense data representations, on the contrary, reduce the generalization error in the case of very large frames. We use two different schemes for representing data in overcomplete systems of Haar and Gabor functions, and measure SVM generalization error on benchmarked data sets.

Rail corrugation detection by Gabor filtering

Inspection of the rail state in order to detect defects is one of the basic tasks in railway maintenance. Rail defects exhibit different properties and are divided in various categories relating to the type and position of flaws on the rail. We propose a technique, based on texture analysis of the rail surface, to detect and classify a particular class of defects: corrugation.

Position estimation for a mobile robot using data fusion

This paper describes a position estimation technique based on the fusion of data obtained by two independent subsystems in a mobile robot navigation context. The first subsystem is a self-location one composed of an onboard camera, an onboard image processing unit and artificial landmarks; the second one is a dead-reckoning subsystem based on odometry. The robot navigation system integrates the position estimation obtained by the vision subsystem with the position estimated by odometry using a Kalman filter framework.

A Neural Network for Egomotion Estimation from Optical Flow.

In this work we consider the problem to determine qualitative information about the motion of a viewer moving in a stationary environment. First the optical flow (OF) is computed using a token based approach estimating the 2D velocity vectors only for some interesting points. Then our method estimates the motion of the viewer using only the available sparse OF. A neural network extracts information about stable points useful for the computation of vehicles heading and Time-to-Collision (TTC). A number of experiments showing the efficacy and robustness of the method have been performed both on synthetic image sequences and on real images acquired by a CCD camera mounted on a mobile platform.

Self-location of a mobile robot by estimation of camera parameters

Geometric approaches to camera localization are suitable for mobile robot navigation. Our work describes a straightforward technique to determine the location of the Center-of-Projection (CP) of the camera and orientation of optical axis using 3 landmarks and capitalizing on the excellent angular resolution of CCD cameras. Applications to mobile robot navigation are analyzed.