MODEL CALIBRATION

Real-time model based geometric reasoning for vision-guided navigation. Mach Vision Appl, 2:31{44 15. Tomasi C, Kanade T (1991) Shape and motion from image streams: a factorization method { 3. Detection and tracking of point features. Road obstacle detection and tracking by an active and intelligent sensing strategy. where he did his thesis work in the elds of computer vision and machine learning for robot navigation. In the past he has worked on image analysis for robotic applications and automated vehicle monitoring. His current major involvements are in computer vision techniques for surveillance applications. His current research interests are mostly focused on pattern recognition, machine learning, and image processing. in the eld of speech coding. He joined the Computer Vision Group at IRST in 1991, where he worked for 3 years on applications of computer vision for surveillance systems, and for 2 years on computer vision for automated vehicle monitoring. Since 1996 he has been involved in applications of image processing for surface inspection. at IRST, working for 5 years on planning and control of mobile robots. He is currently interested in probabilis-tic reasoning for vision-based monitoring and document image analysis. m frames D 9b Fig. 9. Frontal free space measurements are reported for two sequences of 300 frames (about 37 s) each. In a, the vehicle is moving on the rst lane; other vehicles overtake it and then move to the same lane in correspondence with points A, B, and C. In b, the vehicle approaches a slower car from behind and overtakes it in correspondence to point D. The upper limit of the sensor was set to 100 m through which the geometric model constrains the selection of the relevant visual features corresponding to the lane boundaries. At present, however, the parameters controlling the imaging steps (binarization) are set \globally" for the whole image. Further progress may be made by implementing a more \local" low-level processing driven by the available geometric knowledge. Study of row-by-row adaptive thresholding methods for the search of white lines is under way. Frontal free space: The principle on which the module is currently based (the detection of dark spots within a certain area of interest) shows serious shortcomings in presence of sharp shadows or very dark asphalt patches. A rst way of relieving such diiculties would be that of modelling the expected motion of the preceding \object" with respect to the vehicle. Introduction of …