Computing occlusion-free viewpoints

Abstract

This paprr presents inethoi-ls to compuie the locils of nll viewpoints froin which feoturcrs on known polyhedral nhjecls con be viewed in their entirety without being nccluAcd by ailything in ihe eizvironrni?nt. Convcx and coiicavc polyhedra with or withorit holes mid the viewing inndel of perspective projection are eniploycd in this work. h s r d on properties that are dcterrnincd (,f ihe occlusionfree nnd occItiAed loci of viewpoints, two tnethod,r to construct tlie!sc loci together with iheir cninplmity nnalysi%s nre preseriicd. I n the first inerhod, a hounilory rqwesentntion of thc occlii.rioizfree locus i.s ohtnincd. In the second ineihod, thr locus of occluded viewpoints is expressed in terms of a construciive solid geornc.try (CSG) representntion t h n~ cnnsisis of a union ?f coniponent so1id.s. Iinplementation results are shown. I n trodrictioii Wlicti attempting to view a feature of an object from a particular viewpoint, it may happen that some or all points of tlie feature may be occluded by either some part of the object on which it lies, or by other objects in (he environment. 111 this paper, we shall determine the viewpoints liom whicli object features of interest are visible in their entirety in a knowrr polyhedral world. 'I hat is, the features are not occluded by anything in the environment. O w approach has coinmon poiilly with research in the at ea of aqpect graphs [S, 9, lo]. Aspect graphs provide a parlition of' tlircc-dimerisional space iiito sets of viewpoints yielding topologically equivalent views aticl in this sense drfitte a twoader problem. J'his paper cniphasizes that tlie visibility pioldem should be viewed in its own riglit rathrr than as a spccial case of (tie more peticral aspect graph problem. Among the difkrences hetueen the work reportrd liet e and that of aspect graph reseat cli iricludc hie followitig: The pnrccllatiori of viewp4iit <pace in the aspect graph research is done using thc elitire platte deliiietl by the edge and vertvx pair that defines the visual event. Ilowever, as is pointed out i n this paper, it is only tlie locally separating fnrrs (i.e. triangles or trape7oid~) that are nceded. This observation

A General Method for Sensor Planning in Multi-Sensor Systems: Extension to Random Occlusion

“…Several researchers (Cowan and Kovesi 1988;Stamos and Allen 1998;Reed and Allen 2000;Tarabanis et al 1996;Maver and Bajcsy 1993;Yi et al 1995;Spletzer and Taylor 2001;Wixson 1994;Abrams et al 1999) have studied and incorporated more complex constraints based on several factors not limited to (1) resolution, (2) focus, (3) field of view, (4) visibility, (5) view angle, and (6) prohibited regions. The set of possible sensor configurations satisfying all such constraints for all the features in the scene is then determined.…”

“…Several systems take a generate-and-test approach (Sakane et al 1987(Sakane et al , 1992Yi et al 1995), in which sensor configurations are generated and then evaluated with respect to the task constraints. order Another set of methods take a synthesis approach (Anderson 1982;Tarabanis et al 1991Tarabanis et al , 1995bTarabanis et al , 1996Cowan 1988;Cowan and Bergman 1989;Cowan and Kovesi 1988;Stamos and Allen 1998), in which the task constraints are analytically analyzed, and the sensor parameter values that satisfy such analytical relationships are then determined. Sensor simulation is another approach utilized by some systems (Ikeuchi and Robert 1991;Raczkowsky and Mittenbuehler 1989).…”

Shape recovery and viewpoint planning for reverse engineering

“…Also, estimates of the viewpoint location may or may not be available. In case of intensity cameras used in inspection tasks, there is a nearly perfect estimate of the geometry and viewpoint [11][12][13]. In assembly tasks [14] and model-based recognition [15][16][17], the geometry is also known.…”

Visibility-Based Planning of Sensor Control Strategies

“…Tarabanis and Tsai [32] examine the question of complete visibility for general polyhedral environments in three dimensions. For a feature polygon of size m and a polyhedral environment of size n, they present an O(m 3 n 3 ) algorithm for computing the locus of all viewpoints from which the fixed feature polygon can be entirely seen.…”